Publish gemma-4-26B-A4B-it distribution-fidelity artifact (part 3)

Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/inspect.json

@@ -1,6 +1,6 @@
 {
   "model": "/media/fmodels2/Neural-ICE/Gemma-4-26B-A4B-it-NVFP4",
-  "inspect_version": 4,
+  "inspect_version": 5,
   "weights_sha256": "e833541d24828cb56aa5190656ced93552a81d5e50f463b9e4a957d22ca15f32",
   "weights_bytes": 16423205676,
   "quant_method": "modelopt",
@@ -8,6 +8,8 @@
   "detected_scheme": "nvfp4",
   "detected_block": 16,
   "quant_algorithm": "modelopt",
+  "scheme_mix": null,
+  "kv_cache_scheme": null,
   "coverage": {
     "router": {
       "weights": 30,

Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/manifest.json

@@ -775,6 +775,7 @@
     "lm_head.safetensors": "3a92448a235b5cd0f5c187ff34835d0d23a0b0e35b59b60f723d602c5a3c79f4"
   },
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "lm_head": {
     "runtime": {
       "state": "unquantized",
@@ -875,7 +876,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },

Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.md

@@ -1,12 +1,12 @@
 # gemma-4-26B-A4B-it / Neural-ICE-Gemma-4-26B-A4B-it-NVFP4: distribution fidelity
 
-**Mean KLD(reference || candidate) = 1.81880328** over 1572096 scored positions.
+**Mean KLD(reference || candidate) = 1.82281421** over 1572096 scored positions.
 
 | Mean | Median | p90 | p99 | Max | Top-1 agreement |
 |---|---|---|---|---|---|
-| 1.81880328 | 0.67534816 | 5.17001915 | 13.29482074 | 46.64394379 | 57.7526% |
+| 1.82281421 | 0.67840236 | 5.18726063 | 13.24978852 | 49.14295197 | 57.7098% |
 
-Reverse direction, KLD(candidate || reference): 2.10499252.
+Reverse direction, KLD(candidate || reference): 2.11411994.
 
 ## Identity
 
@@ -18,7 +18,7 @@ Reverse direction, KLD(candidate || reference): 2.10499252.
 | Candidate weights SHA-256 | e833541d24828cb5 |
 | Suite | gemma4-31b-it-fidelity-1024x2048-v1 |
 | Suite token SHA-256 | 81d66409b44937fa |
-| Capture manifest SHA-256 | d1f03dea74e9a9b9 |
+| Capture manifest SHA-256 | b5ab5ede01f46551 |
 | Tokenizer | None |
 | Scored vocabulary | 262144 |
 | Declared vocabulary | 262144 |
@@ -27,31 +27,32 @@ Reverse direction, KLD(candidate || reference): 2.10499252.
 | Model runner | V1 |
 | Tensor parallel | 1 |
 | Eager enforced | True |
+| KV cache | bfloat16 |
 | Prefix caching | False |
 | max_num_seqs | 1 |
 | vLLM | 0.1.dev20446+gb2bc9171d |
-| vLLM commit | c798225a7f19 |
+| vLLM commit | a2ae11e29e34 |
 | vLLM dirty digest | e3b0c44298fc1c14 |
 | Compiled extensions | f2fbc7537b0f01f6 |
 | FlashInfer | 0.6.17 |
 | torch | 2.13.0+cu132 |
 | Driver | 580.173.02 |
 | GPUs | NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, NVIDIA RTX PRO 6000 Blackwell Workstation Edition |
-| Laws version | 12 |
+| Laws version | 14 |
 | Partition | analysis |
 
 ## Routed-model intervention: QxQ and BxQ
 
 **The first axis is routing source; the second Q is the same unchanged quantized candidate in both runs.** QxQ uses the student's natural expert IDs. BxQ forces the BF16 teacher's expert IDs while the student computes its own gating weights for those experts.
 
-| Run | Routing | Mean KLD | Support |
-|---|---|---|---|
-| QxQ | student natural IDs; student gating weights | 1.81880328 | [report](report.json) |
-| BxQ | BF16 teacher IDs; student gating weights | 1.58869884 | [report](report.json) |
-| QxQ - BxQ | paired routing-intervention delta; not additive attribution | +0.23010445 |  |
-| Exact repeat | two QxQ, two forced-natural, two BxQ samples | certified |  |
+| Run | Routing | Mean KLD | Measurability | Support |
+|---|---|---|---|---|
+| QxQ | student natural IDs; student gating weights | 1.82281421 | measured | [report](report.json) |
+| BxQ | BF16 teacher IDs; student gating weights | 1.58512109 | measured | [report](report.json) |
+| QxQ - BxQ | paired routing-intervention delta; not additive attribution | +0.23769312 |  |  |
+| Exact repeat | two QxQ, two forced-natural, two BxQ samples | certified |  |  |
 
-Natural QxQ selections changed from the teacher at **74.4082%** of (token, layer) choices. BxQ protocol 4; routing trace `65a995be0ed63037`.
+Natural QxQ selections changed from the teacher at **74.4778%** of (token, layer) choices. BxQ protocol 5; routing trace `0fcde10d9c35222c`.
 
 Natural control used exact-repeat certification: max position delta 0.000e+00 / 0.000e+00; absolute mean delta 0.000e+00 / 0.000e+00; natural route-repeat flips 0.000%.
 
@@ -71,6 +72,7 @@ No cell here holds expert selection fixed. Rounding any weight changes the resid
 | Declared method | modelopt |
 | Algorithm | modelopt |
 | Activation scheme | n/a |
+| KV cache | unquantized (none declared; scoring pins unquantized regardless) |
 
 | Component | Quantized weights | Coverage |
 |---|---|---|
@@ -91,14 +93,14 @@ Two terms sit outside every cell above, because no cell can hold them. They are
 
 | Term | Value | Share | What it is | Derived in |
 |---|---|---|---|---|
-| Beyond weight rounding | +0.38238287 | 21% of the deployed mean | activation quantization and kernel arithmetic | [Beyond weight rounding](#beyond-weight-rounding) |
-| Routing, measured | saturated | n/a | expert selection changed at 74.4082% of choices and every scored position rerouted, so no held-routing population remains to measure the excess against | [Routing divergence](#routing-divergence) |
+| Beyond weight rounding | +0.38639380 | 21% of the deployed mean | activation quantization and kernel arithmetic | [Beyond weight rounding](#beyond-weight-rounding) |
+| Routing, measured | saturated | n/a | expert selection changed at 74.4778% of choices and every scored position rerouted, so no held-routing population remains to measure the excess against | [Routing divergence](#routing-divergence) |
 
 The selections column is why none of these cells is routing-free: each one reroutes some tokens purely as a consequence of rounding weights upstream of a router it never touched.
 
 ### Beyond weight rounding
 
-**+0.38238287** (21% of the deployed mean). Every cell above rounds weights and runs on BF16 kernels. The deployed checkpoint differs from the composite cell by this much. The checkpoint quantizes weights only, so this term is kernel arithmetic. A weight-only analysis, which is what any QDQ cell is, cannot see it.
+**+0.38639380** (21% of the deployed mean). Every cell above rounds weights and runs on BF16 kernels. The deployed checkpoint differs from the composite cell by this much. The checkpoint quantizes weights only, so this term is kernel arithmetic. A weight-only analysis, which is what any QDQ cell is, cannot see it.
 
 The deployed checkpoint leaves the router in BF16, so router weight precision costs exactly nothing here. That is not the same as routing costing nothing: an identical router fed perturbed activations selects different experts, which is measured under [Routing divergence](#routing-divergence).
 
@@ -124,15 +126,15 @@ This is the observed selection divergence in the normal QxQ run, not the BxQ int
 
 | Measure | Rate | Of what |
 |---|---|---|
-| Selections changed | 74.4082% | of 47162880 (token, layer) choices |
-| Expert slots changed | 41.0994% | of individual expert slots |
-| Highest-weighted expert changed | 26.7558% | includes reordering within an unchanged selection |
+| Selections changed | 74.4778% | of 47162880 (token, layer) choices |
+| Expert slots changed | 41.1550% | of individual expert slots |
+| Highest-weighted expert changed | 26.7992% | includes reordering within an unchanged selection |
 | Positions with any flip | 100.0000% | 1572096 of 1572096 scored positions |
 
 | Positions where | Mean KLD | Count |
 |---|---|---|
 | Routing held | n/a | 0 |
-| Routing flipped | 1.81880328 | 1572096 |
+| Routing flipped | 1.82281421 | 1572096 |
 
 **Ranking floor undefined.** Every one of the 1572096 scored positions rerouted in at least one layer, so there is no held-routing population to compare the flipped ones against. The floor is not small here, it is unmeasurable, and no rescore changes that: the perturbation is large enough that routing divergence is the whole picture rather than a correction to it. Rank against this candidate on the experts-only cell, and treat the deployed mean as unranked.
 
@@ -140,10 +142,10 @@ Divergence rises with how many layers rerouted, which is what distinguishes a si
 
 | Layers rerouted | Positions | Mean KLD |
 |---|---|---|
-| 4-7 | 19 | 0.81431452 |
-| 8+ | 1572077 | 1.81881542 |
+| 4-7 | 18 | 0.74741639 |
+| 8+ | 1572078 | 1.82282652 |
 
-Per-layer selection change runs from 28.33% to 90.55% (worst at layer 29), averaging 44.54% across the first quarter of routed layers and 88.37% across the last. Rerouting that grows with depth is accumulated perturbation, not router precision.
+Per-layer selection change runs from 28.33% to 90.64% (worst at layer 29), averaging 44.67% across the first quarter of routed layers and 88.42% across the last. Rerouting that grows with depth is accumulated perturbation, not router precision.
 
 ## Fidelity by domain
 
@@ -151,31 +153,31 @@ The suite is stratified, so the mean above is an average over kinds of text that
 
 | Domain | Contexts | Reference top-1 | Mean KLD | x run | Worst context |
 |---|---|---|---|---|---|
-| Other multilingual content | 36 | 70.7% | 2.36378738 | 1.30 | 4.47943237 |
-| Natural dialogue, instruction following, and assistance | 96 | 77.4% | 2.07672787 | 1.14 | 4.15724368 |
-| Scientific and technical exposition | 96 | 69.9% | 2.07479035 | 1.14 | 3.38248339 |
-| Encyclopedic and factual reference | 96 | 68.7% | 2.02309414 | 1.11 | 3.37861824 |
-| Chinese across several content types | 72 | 63.0% | 1.84919395 | 1.02 | 2.94232623 |
-| News, history, economics, legal analysis, and essays | 72 | 68.9% | 1.82956264 | 1.01 | 3.43971685 |
-| Literary, narrative, and creative writing | 72 | 63.0% | 1.65811583 | 0.91 | 3.08165872 |
-| Worked mathematics, science, and formal reasoning | 96 | 72.3% | 1.56078164 | 0.86 | 3.25403348 |
-| Structured data, tool calls, APIs, JSON, and tables | 36 | 83.3% | 1.42163395 | 0.78 | 3.19270213 |
-| Source code, tests, technical documentation, and issue discussions | 96 | 84.2% | 1.39284493 | 0.77 | 2.87263039 |
+| Other multilingual content | 36 | 70.7% | 2.36413931 | 1.30 | 4.52259375 |
+| Natural dialogue, instruction following, and assistance | 96 | 77.4% | 2.07293250 | 1.14 | 3.94908391 |
+| Encyclopedic and factual reference | 96 | 68.7% | 2.02868660 | 1.11 | 3.50657211 |
+| Scientific and technical exposition | 96 | 69.9% | 2.02530800 | 1.11 | 3.51423229 |
+| Chinese across several content types | 72 | 63.0% | 1.90327632 | 1.04 | 3.17382072 |
+| News, history, economics, legal analysis, and essays | 72 | 68.9% | 1.80855383 | 0.99 | 4.06137836 |
+| Literary, narrative, and creative writing | 72 | 63.0% | 1.70555695 | 0.94 | 3.10018999 |
+| Worked mathematics, science, and formal reasoning | 96 | 72.3% | 1.54692551 | 0.85 | 3.28942782 |
+| Structured data, tool calls, APIs, JSON, and tables | 36 | 83.3% | 1.48350580 | 0.81 | 3.17805380 |
+| Source code, tests, technical documentation, and issue discussions | 96 | 84.2% | 1.40275384 | 0.77 | 2.82508607 |
 
 The same domains measured on the other attribution cells, which shows whether a domain's weakness is a property of the model or of the format:
 
 | Domain | deployed | bxq | mxfp8 | nvfp4 | all_quantized |
 |---|---|---|---|---|---|
-| Other multilingual content | 2.36378738 | 2.01060924 | 0.68445047 | 0.89229785 | 1.79519304 |
-| Natural dialogue, instruction following, and assistance | 2.07672787 | 1.79807095 | 0.59552252 | 0.82821448 | 1.66159067 |
-| Scientific and technical exposition | 2.07479035 | 1.82912511 | 0.50141256 | 0.70339667 | 1.68508002 |
-| Encyclopedic and factual reference | 2.02309414 | 1.73666327 | 0.49743044 | 0.71954225 | 1.56808539 |
-| Chinese across several content types | 1.84919395 | 1.67080916 | 0.44706156 | 0.68167667 | 1.40434265 |
-| News, history, economics, legal analysis, and essays | 1.82956264 | 1.57566891 | 0.44719073 | 0.64140586 | 1.41631587 |
-| Literary, narrative, and creative writing | 1.65811583 | 1.49461078 | 0.39173107 | 0.62338708 | 1.35504621 |
-| Worked mathematics, science, and formal reasoning | 1.56078164 | 1.36868653 | 0.37927146 | 0.52795210 | 1.23575505 |
-| Structured data, tool calls, APIs, JSON, and tables | 1.42163395 | 1.27786368 | 0.47775227 | 0.61546437 | 1.16853193 |
-| Source code, tests, technical documentation, and issue discussions | 1.39284493 | 1.18805084 | 0.36354693 | 0.52781126 | 1.09767677 |
+| Other multilingual content | 2.36413931 | 2.12630776 | 0.68445047 | 0.89229785 | 1.79519304 |
+| Natural dialogue, instruction following, and assistance | 2.07293250 | 1.83255161 | 0.59552252 | 0.82821448 | 1.66159067 |
+| Encyclopedic and factual reference | 2.02868660 | 1.72222707 | 0.49743044 | 0.71954225 | 1.56808539 |
+| Scientific and technical exposition | 2.02530800 | 1.78720326 | 0.50141256 | 0.70339667 | 1.68508002 |
+| Chinese across several content types | 1.90327632 | 1.66541288 | 0.44706156 | 0.68167667 | 1.40434265 |
+| News, history, economics, legal analysis, and essays | 1.80855383 | 1.54854078 | 0.44719073 | 0.64140586 | 1.41631587 |
+| Literary, narrative, and creative writing | 1.70555695 | 1.50247649 | 0.39173107 | 0.62338708 | 1.35504621 |
+| Worked mathematics, science, and formal reasoning | 1.54692551 | 1.33582942 | 0.37927146 | 0.52795210 | 1.23575505 |
+| Structured data, tool calls, APIs, JSON, and tables | 1.48350580 | 1.26618476 | 0.47775227 | 0.61546437 | 1.16853193 |
+| Source code, tests, technical documentation, and issue discussions | 1.40275384 | 1.19365006 | 0.36354693 | 0.52781126 | 1.09767677 |
 
 **Other multilingual content** is this candidate's weakest domain and **Source code, tests, technical documentation, and issue discussions** its strongest, a spread of 1.7x. A deployment weighted toward the weakest domain sees more divergence than the headline mean implies; the per-source breakdown and the reading are in [strata.md](strata.md) and [strata.json](strata.json).
 
@@ -183,8 +185,8 @@ The same domains measured on the other attribution cells, which shows whether a
 
 | Component | Value |
 |---|---|
-| Trunk (candidate hidden states, reference head) | 1.81880333 |
-| Deployed (candidate's own head) | 1.81880328 |
+| Trunk (candidate hidden states, reference head) | 1.82281426 |
+| Deployed (candidate's own head) | 1.82281421 |
 | Head-associated delta (not additive) | -0.00000005 |
 | Reference head | {'state': 'unquantized', 'workers': [{'heads': [{'name': 'language_model.lm_head', 'org_vocab_size': 262144, 'quant_method': 'UnquantizedEmbeddingMethod', 'state': 'unquantized', 'weight_dtype': 'torch.bfloat16'}], 'logits_processor': {'head_dtype': 'torch.bfloat16', 'org_vocab_size': 262144, 'scale': 1.0, 'soft_cap': 30.0, 'type': 'LogitsProcessor', 'vocab_size': 262144}, 'primary': {'org_vocab_size': 262144, 'quant_method': 'UnquantizedEmbeddingMethod', 'state': 'unquantized', 'weight_dtype': 'torch.bfloat16'}, 'state': 'unquantized'}]} |
 | Candidate head | {'runtime': {'state': 'quantized', 'workers': [{'heads': [{'name': 'language_model.lm_head', 'org_vocab_size': 262144, 'quant_method': 'UnquantizedLinearMethod', 'state': 'quantized', 'weight_dtype': 'torch.bfloat16'}], 'logits_processor': {'head_dtype': 'torch.bfloat16', 'org_vocab_size': 262144, 'scale': 1.0, 'soft_cap': 30.0, 'type': 'LogitsProcessor', 'vocab_size': 262144}, 'primary': {'org_vocab_size': 262144, 'quant_method': 'UnquantizedLinearMethod', 'state': 'quantized', 'weight_dtype': 'torch.bfloat16'}, 'state': 'quantized'}]}, 'state': 'quantized', 'static': {'ignored': True, 'lm_head_dtypes': {'model.language_model.embed_tokens.weight': 'BF16'}, 'lm_head_keys': [], 'output_weight_keys': ['model.language_model.embed_tokens.weight'], 'packed_keys': [], 'quant_method': 'modelopt', 'state': 'unquantized', 'tie_word_embeddings': True}} |
@@ -193,26 +195,26 @@ The same domains measured on the other attribution cells, which shows whether a
 
 | Position range | Positions | Mean KLD |
 |---|---|---|
-| 0–511 | 393216 | 1.48458588 |
-| 512–1023 | 393216 | 1.59614646 |
-| 1024–1535 | 393216 | 1.90740159 |
-| 1536–2046 | 392448 | 2.28799558 |
+| 0–511 | 393216 | 1.49461607 |
+| 512–1023 | 393216 | 1.61429620 |
+| 1024–1535 | 393216 | 1.91318905 |
+| 1536–2046 | 392448 | 2.27002897 |
 
 ## Error by reference confidence
 
 | Reference top-1 probability | Positions | Share | Mean KLD |
 |---|---|---|---|
-| [0.00, 0.25) | 113465 | 7.2% | 3.55419825 |
-| [0.25, 0.50) | 291225 | 18.5% | 2.68340379 |
-| [0.50, 0.75) | 308315 | 19.6% | 2.15399698 |
-| [0.75, 0.95) | 317610 | 20.2% | 1.71816747 |
-| [0.95, 1.00) | 541481 | 34.4% | 0.85832234 |
+| [0.00, 0.25) | 113465 | 7.2% | 3.55395796 |
+| [0.25, 0.50) | 291225 | 18.5% | 2.68651026 |
+| [0.50, 0.75) | 308315 | 19.6% | 2.16130676 |
+| [0.75, 0.95) | 317610 | 20.2% | 1.71971882 |
+| [0.95, 1.00) | 541481 | 34.4% | 0.86327489 |
 
 ## Top-K set agreement
 
 | K=1 | K=2 | K=3 | K=4 | K=5 |
 |---|---|---|---|---|
-| 57.7255% | 22.2352% | 6.7988% | 1.7936% | 0.4376% |
+| 57.6859% | 22.0960% | 6.7311% | 1.7631% | 0.4385% |
 
 ## Law compliance
 
@@ -224,17 +226,18 @@ Zero baseline: reference against itself scored **0.00000000** over 2047 position
 | 2 | Determinism | PASS | eager enforced; prefix caching off, max_num_seqs=1 |
 | 3 | Frozen input | PASS | token hash matches suite gemma4-31b-it-fidelity-1024x2048-v1 [analysis]: 81d66409b44937fa |
 | 4 | Real vocabulary | PASS | scored 262144 real tokens of 262144 declared (0 padding rows) |
-| 5 | Manifest binding | PASS | all bound fields present; manifest d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a |
-| 6 | Provenance | PASS | torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ c798225a7f19, driver 580.173.02, 4 GPU(s) |
+| 5 | Manifest binding | PASS | all bound fields present; manifest b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef |
+| 6 | Provenance | PASS | torch 2.13.0+cu132, vLLM 0.1.dev20446+gb2bc9171d @ a2ae11e29e34, driver 580.173.02, 4 GPU(s) |
 | 7 | Storage integrity | PASS | hidden storage with bitwise-exact replay |
-| 8 | Head transparency | PASS | trunk 1.81880333, deployed 1.81880328, delta -5.164658944956102e-08 |
-| 9 | Tail and depth disclosure | PASS | mean 1.81880328, median 0.67534816, max 46.64394379, 4 depth buckets |
+| 8 | Head transparency | PASS | trunk 1.82281426, deployed 1.82281421, delta -5.102717559601899e-08 |
+| 9 | Tail and depth disclosure | PASS | mean 1.82281421, median 0.67840236, max 49.14295197, 4 depth buckets |
 | 10 | Comparability | PASS | comparability key fully resolved |
 | 11 | Freeze before qualification | NOT_APPLICABLE | partition is 'analysis' |
 | 12 | Reusable reference | PASS | suite, reference, head, and checksums present; published reference matches the scored capture on every bound field |
-| 14 | Routed-model intervention | PASS | QxQ 1.81880328, BxQ 1.58869884, QxQ - BxQ +0.23010445; natural selections changed at 74.408% of (token, layer) choices; trace 65a995be0ed63037 |
-| 15 | Domain disclosure | PASS | 10 domains disclosed; weakest other_multilingual at 2.36378738, strongest code_docs_issues at 1.39284493, spread 1.7x |
+| 14 | Routed-model intervention | PASS | QxQ 1.82281421, BxQ 1.58512109, QxQ - BxQ +0.23769312; natural selections changed at 74.478% of (token, layer) choices; trace 0fcde10d9c35222c |
+| 15 | Domain disclosure | PASS | 10 domains disclosed; weakest other_multilingual at 2.36413931, strongest code_docs_issues at 1.40275384, spread 1.7x |
 | 16 | Candidate weight binding | PASS | scored weights e833541d24828cb5 as inspected |
+| 17 | Substitution disclosure | PASS | w2_input_scale uncalibrated_experts_filled_from_layer_max across 15/30 layers, worst spread 230.6x, 26 unusable slot(s); w13_input_scale uncalibrated_experts_filled_from_layer_max across 15/30 layers, worst spread 54.17x, 52 unusable slot(s) |
 | 13 | Recorded deviation | PASS | no overrides claimed |
 
 ## Environment
@@ -245,7 +248,7 @@ Zero baseline: reference against itself scored **0.00000000** over 2047 position
 | Platform | Linux-6.8.0-137-generic-x86_64-with-glibc2.39 |
 | Python | 3.12.3 |
 | vLLM | 0.1.dev20446+gb2bc9171d |
-| vLLM commit | c798225a7f197994ed502068fc6ac1ba8a9706f3 |
+| vLLM commit | a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61 |
 | torch | 2.13.0+cu132 |
 | torch CUDA runtime | 13.2 |
 | cuDNN | 9.20.0 (92000) |
@@ -280,11 +283,6 @@ Credential values are never published. Set but redacted: `HF_TOKEN`.
 | `CUDA_HOME` | `/usr/local/cuda-13.0` |
 | `HF_TOKEN` | `<redacted>` |
 | `NCCL_DETERMINISTIC` | `1` |
-| `PYTORCH_NVML_BASED_CUDA_CHECK` | `1` |
-| `TORCHINDUCTOR_CACHE_DIR` | `/tmp/torchinductor_phaedawg` |
-| `TORCHINDUCTOR_COMPILE_THREADS` | `1` |
-| `TRITON_CACHE_AUTOTUNING` | `1` |
-| `TRITON_PTXAS_BLACKWELL_PATH` | `/usr/local/cuda-13.0/bin/ptxas` |
 | `VLLM_BATCH_INVARIANT` | `1` |
 | `VLLM_MARLIN_USE_ATOMIC_ADD` | `0` |
 | `VLLM_MOE_USE_DEEP_GEMM` | `0` |
@@ -295,26 +293,26 @@ Paths are relative to the artifact root, the same paths `checksums.txt` uses. Ve
 
 | Path | Size | What it is |
 |---|---|---|
-| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.md` | 21.67 KiB | This document. |
-| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.json` | 564.40 KiB | Every statistic behind it, machine-readable: per-bucket means, percentiles, agreement rates, and the phase timings. |
-| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/manifest.json` | 87.32 KiB | The capture manifest this result is bound to (Law 5). Scoring refuses to run if the live configuration differs from it. |
-| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/compliance.json` | 370.41 KiB | The law-by-law receipt, including the comparability key. |
-| `baselines/self-kld.json` | 6.72 KiB | The zero-baseline proof required by Law 1: the reference scored against a capture of itself. |
+| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.md` | 21.85 KiB | This document. |
+| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.json` | 596.81 KiB | Every statistic behind it, machine-readable: per-bucket means, percentiles, agreement rates, and the phase timings. |
+| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/manifest.json` | 87.35 KiB | The capture manifest this result is bound to (Law 5). Scoring refuses to run if the live configuration differs from it. |
+| `Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/compliance.json` | 404.60 KiB | The law-by-law receipt, including the comparability key. |
+| `baselines/self-kld.json` | 6.89 KiB | The zero-baseline proof required by Law 1: the reference scored against a capture of itself. |
 | `suite/suite-manifest.json` | 382.61 KiB | The frozen evaluation input's identity: token hashes per context and per partition, sources, strata, and the analysis/qualification split. |
 | `suite/tokens` | 17.32 MiB in 1024 files | The token IDs themselves. These are the evaluation input, not a description of it; retokenizing source text does not reproduce them. |
 | `suite/sources.json` | 737.48 KiB | Per-context provenance: dataset, revision, licence, source unit, and the deterministic token offset chosen within the document. |
 | `suite/validation/capability-overlap.json` | 1.48 KiB | The benchmark-contamination scan and every document it blocked. |
-| `reference/manifest.json` | 87.32 KiB | The reference capture's own manifest: geometry, vocabulary, storage mode, and a hash for every tensor file. |
+| `reference/manifest.json` | 87.35 KiB | The reference capture's own manifest: geometry, vocabulary, storage mode, and a hash for every tensor file. |
 | `reference` | 11.62 GiB in 771 files | The reusable reference distributions. Pass this directory as `--reference-logits` to score a new candidate against the same reference without loading the reference checkpoint. |
 | `reference/lm_head.safetensors` | 1.38 GiB | The reference language-model head, which turns the stored hidden states back into reference logits. |
-| `environment/runtime.json` | 5.33 KiB | Machine-readable provenance: torch, CUDA, cuDNN, NCCL, driver, devices, and the captured environment variables. |
+| `environment/runtime.json` | 5.08 KiB | Machine-readable provenance: torch, CUDA, cuDNN, NCCL, driver, devices, and the captured environment variables. |
 | `environment/summary.md` | 1.10 KiB | The same provenance as prose, plus an index of every captured file. |
 | `environment/toolchain-nvcc.txt` | 243 B | `nvcc --version` verbatim; `toolchain-gcc.txt` and `toolchain-ldd.txt` sit beside it. |
 | `environment/gpu-smi-query.txt` | 54.73 KiB | `nvidia-smi -q` verbatim: ECC state, persistence mode, clocks, and throttle reasons, any of which can move a bitwise result. |
 | `environment/pip-freeze.txt` | 4.47 KiB | Every installed package version in the scoring environment. |
-| `environment/models` | 25.42 KiB in 7 files | Checkpoint fingerprints: file listing, sizes, config and tokenizer hashes, and `config.json` verbatim for each model scored. |
+| `environment/models` | 28.70 KiB in 7 files | Checkpoint fingerprints: file listing, sizes, config and tokenizer hashes, and `config.json` verbatim for each model scored. |
 | `checksums.txt` | 1.13 MiB | `sha256sum --check` compatible over every other file here. This is authoritative for integrity (Law 12). |
-| `LAWS.md` | 31.84 KiB | The laws this artifact was produced under, including the override procedure. |
+| `LAWS.md` | 37.98 KiB | The laws this artifact was produced under, including the override procedure. |
 
 ## Scope
 

Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.json

@@ -8,78 +8,6 @@
   ],
   "groups": {
     "source_key": [
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 6,
-            "max_kld": 28.93311882019043,
-            "mean_kld": 2.5519506836519916,
-            "mean_ref_top1_prob": 0.6737162987948037,
-            "median_context_kld": 2.7713133945380797,
-            "median_context_p99": 15.75142765045166,
-            "p90_context_kld": 2.969656211924431,
-            "positions": 12282,
-            "top1_agreement": 0.48078488845464906,
-            "worst_context_id": 969,
-            "worst_context_kld": 2.969656211924431
-          },
-          "bxq": {
-            "contexts": 6,
-            "max_kld": 32.43082809448242,
-            "mean_kld": 3.099819791888436,
-            "mean_ref_top1_prob": 0.6737162987948037,
-            "median_context_kld": 3.3174707252638598,
-            "median_context_p99": 16.03008460998535,
-            "p90_context_kld": 3.804737185271593,
-            "positions": 12282,
-            "top1_agreement": 0.42794333170493404,
-            "worst_context_id": 969,
-            "worst_context_kld": 3.804737185271593
-          },
-          "deployed": {
-            "contexts": 6,
-            "max_kld": 32.0179443359375,
-            "mean_kld": 3.4288325302364404,
-            "mean_ref_top1_prob": 0.6737162987948037,
-            "median_context_kld": 3.5853966743722316,
-            "median_context_p99": 19.162126541137695,
-            "p90_context_kld": 4.246862471248152,
-            "positions": 12282,
-            "top1_agreement": 0.41393909786679695,
-            "worst_context_id": 975,
-            "worst_context_kld": 4.246862471248152
-          },
-          "mxfp8": {
-            "contexts": 6,
-            "max_kld": 33.06193542480469,
-            "mean_kld": 1.2310746371314905,
-            "mean_ref_top1_prob": 0.6737162987948037,
-            "median_context_kld": 1.1703093400484361,
-            "median_context_p99": 14.220373153686523,
-            "p90_context_kld": 2.2051642107922818,
-            "positions": 12282,
-            "top1_agreement": 0.6832763393584107,
-            "worst_context_id": 951,
-            "worst_context_kld": 2.2051642107922818
-          },
-          "nvfp4": {
-            "contexts": 6,
-            "max_kld": 33.5982780456543,
-            "mean_kld": 1.5728345065674871,
-            "mean_ref_top1_prob": 0.6737162987948037,
-            "median_context_kld": 1.7573445549604514,
-            "median_context_p99": 15.11145305633545,
-            "p90_context_kld": 2.1238786811814974,
-            "positions": 12282,
-            "top1_agreement": 0.6252239049014818,
-            "worst_context_id": 969,
-            "worst_context_kld": 2.1238786811814974
-          }
-        },
-        "key": "wikipedia_ru",
-        "label": "wikipedia_ru",
-        "relative_to_run": 1.885213517784071
-      },
       {
         "cells": {
           "all_quantized": {
@@ -97,29 +25,29 @@
           },
           "bxq": {
             "contexts": 6,
-            "max_kld": 32.78110122680664,
-            "mean_kld": 2.4811192159132895,
+            "max_kld": 31.15774917602539,
+            "mean_kld": 2.8207919863737145,
             "mean_ref_top1_prob": 0.7085505361620842,
-            "median_context_kld": 2.7373627584033318,
-            "median_context_p99": 16.09213638305664,
-            "p90_context_kld": 3.7471527233231927,
+            "median_context_kld": 3.039518434706234,
+            "median_context_p99": 16.104936599731445,
+            "p90_context_kld": 4.170437929941192,
             "positions": 12282,
-            "top1_agreement": 0.5079791564891711,
+            "top1_agreement": 0.47760950985181566,
             "worst_context_id": 946,
-            "worst_context_kld": 3.7471527233231927
+            "worst_context_kld": 4.170437929941192
           },
           "deployed": {
             "contexts": 6,
-            "max_kld": 41.130943298339844,
-            "mean_kld": 2.9473331214106677,
+            "max_kld": 49.14295196533203,
+            "mean_kld": 3.204664049557937,
             "mean_ref_top1_prob": 0.7085505361620842,
-            "median_context_kld": 2.916557520377998,
-            "median_context_p99": 17.520902633666992,
-            "p90_context_kld": 4.479432370259336,
+            "median_context_kld": 3.4375257608068144,
+            "median_context_p99": 18.00170135498047,
+            "p90_context_kld": 4.5225937464682,
             "positions": 12282,
-            "top1_agreement": 0.47736524995929003,
+            "top1_agreement": 0.45236932095749877,
             "worst_context_id": 946,
-            "worst_context_kld": 4.479432370259336
+            "worst_context_kld": 4.5225937464682
           },
           "mxfp8": {
             "contexts": 6,
@@ -150,7 +78,79 @@
         },
         "key": "wikipedia_cs",
         "label": "wikipedia_cs",
-        "relative_to_run": 1.620479330179758
+        "relative_to_run": 1.7580859495349368
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 6,
+            "max_kld": 28.93311882019043,
+            "mean_kld": 2.5519506836519916,
+            "mean_ref_top1_prob": 0.6737162987948037,
+            "median_context_kld": 2.7713133945380797,
+            "median_context_p99": 15.75142765045166,
+            "p90_context_kld": 2.969656211924431,
+            "positions": 12282,
+            "top1_agreement": 0.48078488845464906,
+            "worst_context_id": 969,
+            "worst_context_kld": 2.969656211924431
+          },
+          "bxq": {
+            "contexts": 6,
+            "max_kld": 32.99455261230469,
+            "mean_kld": 3.345258112791561,
+            "mean_ref_top1_prob": 0.6737162987948037,
+            "median_context_kld": 3.341322453586905,
+            "median_context_p99": 17.53594970703125,
+            "p90_context_kld": 4.436682889352712,
+            "positions": 12282,
+            "top1_agreement": 0.40514574173587364,
+            "worst_context_id": 975,
+            "worst_context_kld": 4.436682889352712
+          },
+          "deployed": {
+            "contexts": 6,
+            "max_kld": 34.00013732910156,
+            "mean_kld": 3.013584198435825,
+            "mean_ref_top1_prob": 0.6737162987948037,
+            "median_context_kld": 3.344363800410978,
+            "median_context_p99": 17.513042449951172,
+            "p90_context_kld": 4.171558654484496,
+            "positions": 12282,
+            "top1_agreement": 0.4503338218531184,
+            "worst_context_id": 969,
+            "worst_context_kld": 4.171558654484496
+          },
+          "mxfp8": {
+            "contexts": 6,
+            "max_kld": 33.06193542480469,
+            "mean_kld": 1.2310746371314905,
+            "mean_ref_top1_prob": 0.6737162987948037,
+            "median_context_kld": 1.1703093400484361,
+            "median_context_p99": 14.220373153686523,
+            "p90_context_kld": 2.2051642107922818,
+            "positions": 12282,
+            "top1_agreement": 0.6832763393584107,
+            "worst_context_id": 951,
+            "worst_context_kld": 2.2051642107922818
+          },
+          "nvfp4": {
+            "contexts": 6,
+            "max_kld": 33.5982780456543,
+            "mean_kld": 1.5728345065674871,
+            "mean_ref_top1_prob": 0.6737162987948037,
+            "median_context_kld": 1.7573445549604514,
+            "median_context_p99": 15.11145305633545,
+            "p90_context_kld": 2.1238786811814974,
+            "positions": 12282,
+            "top1_agreement": 0.6252239049014818,
+            "worst_context_id": 969,
+            "worst_context_kld": 2.1238786811814974
+          }
+        },
+        "key": "wikipedia_ru",
+        "label": "wikipedia_ru",
+        "relative_to_run": 1.653259110808003
       },
       {
         "cells": {
@@ -169,29 +169,29 @@
           },
           "bxq": {
             "contexts": 3,
-            "max_kld": 22.545244216918945,
-            "mean_kld": 2.0154727196300577,
+            "max_kld": 30.3443603515625,
+            "mean_kld": 1.8255684168147637,
             "mean_ref_top1_prob": 0.698265131049136,
-            "median_context_kld": 1.8069082424712604,
-            "median_context_p99": 11.676387786865234,
-            "p90_context_kld": 2.591646807757937,
+            "median_context_kld": 1.5544368652654912,
+            "median_context_p99": 10.238090515136719,
+            "p90_context_kld": 2.401359476169069,
             "positions": 6141,
-            "top1_agreement": 0.5142484937306627,
+            "top1_agreement": 0.5427454811919883,
             "worst_context_id": 967,
-            "worst_context_kld": 2.591646807757937
+            "worst_context_kld": 2.401359476169069
           },
           "deployed": {
             "contexts": 3,
-            "max_kld": 38.50217819213867,
-            "mean_kld": 2.4596415123241706,
+            "max_kld": 46.5369873046875,
+            "mean_kld": 2.2710320648727085,
             "mean_ref_top1_prob": 0.698265131049136,
-            "median_context_kld": 2.3877085507229574,
-            "median_context_p99": 14.297967910766602,
-            "p90_context_kld": 3.087839781739251,
+            "median_context_kld": 2.138339814316542,
+            "median_context_p99": 12.894909858703613,
+            "p90_context_kld": 2.853262484980115,
             "positions": 6141,
-            "top1_agreement": 0.4673505943657385,
-            "worst_context_id": 931,
-            "worst_context_kld": 3.087839781739251
+            "top1_agreement": 0.4921022634750041,
+            "worst_context_id": 967,
+            "worst_context_kld": 2.853262484980115
           },
           "mxfp8": {
             "contexts": 3,
@@ -222,7 +222,7 @@
         },
         "key": "wikipedia_fr",
         "label": "wikipedia_fr",
-        "relative_to_run": 1.3523405961202293
+        "relative_to_run": 1.2458933299878305
       },
       {
         "cells": {
@@ -241,29 +241,29 @@
           },
           "bxq": {
             "contexts": 7,
-            "max_kld": 27.351600646972656,
-            "mean_kld": 1.8432502642142068,
+            "max_kld": 27.25080680847168,
+            "mean_kld": 1.834970510312959,
             "mean_ref_top1_prob": 0.7784749537441713,
-            "median_context_kld": 1.8458886430239592,
-            "median_context_p99": 11.044820785522461,
-            "p90_context_kld": 2.3559821675205432,
+            "median_context_kld": 1.8415601742815824,
+            "median_context_p99": 11.829607963562012,
+            "p90_context_kld": 2.1152573524147185,
             "positions": 14329,
-            "top1_agreement": 0.5538418591667248,
-            "worst_context_id": 941,
-            "worst_context_kld": 2.3559821675205432
+            "top1_agreement": 0.5639611975713588,
+            "worst_context_id": 953,
+            "worst_context_kld": 2.1152573524147185
           },
           "deployed": {
             "contexts": 7,
-            "max_kld": 29.357194900512695,
-            "mean_kld": 2.14317143353844,
+            "max_kld": 28.285417556762695,
+            "mean_kld": 2.1918876152881976,
             "mean_ref_top1_prob": 0.7784749537441713,
-            "median_context_kld": 2.139621759398296,
-            "median_context_p99": 12.934392929077148,
-            "p90_context_kld": 2.538131887870744,
+            "median_context_kld": 2.070896143583137,
+            "median_context_p99": 13.104333877563477,
+            "p90_context_kld": 2.8790162790291833,
             "positions": 14329,
-            "top1_agreement": 0.5266243282852956,
+            "top1_agreement": 0.5168539325842697,
             "worst_context_id": 953,
-            "worst_context_kld": 2.538131887870744
+            "worst_context_kld": 2.8790162790291833
           },
           "mxfp8": {
             "contexts": 7,
@@ -294,7 +294,7 @@
         },
         "key": "wikipedia_de",
         "label": "wikipedia_de",
-        "relative_to_run": 1.178341526396078
+        "relative_to_run": 1.2024745058469979
       },
       {
         "cells": {
@@ -313,29 +313,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 42.54874801635742,
-            "mean_kld": 1.798070948314533,
+            "max_kld": 41.96346664428711,
+            "mean_kld": 1.832551607899988,
             "mean_ref_top1_prob": 0.7743913512965481,
-            "median_context_kld": 1.7649751579712167,
-            "median_context_p99": 14.137372970581055,
-            "p90_context_kld": 2.7188406299445234,
+            "median_context_kld": 1.8701310658748034,
+            "median_context_p99": 14.359136581420898,
+            "p90_context_kld": 2.840875715129075,
             "positions": 196512,
-            "top1_agreement": 0.6251323074417847,
-            "worst_context_id": 571,
-            "worst_context_kld": 3.539333405062736
+            "top1_agreement": 0.6200181159420289,
+            "worst_context_id": 484,
+            "worst_context_kld": 3.525342154276217
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 38.28725814819336,
-            "mean_kld": 2.0767278693667035,
+            "max_kld": 44.010807037353516,
+            "mean_kld": 2.072932495110033,
             "mean_ref_top1_prob": 0.7743913512965481,
-            "median_context_kld": 2.0825514215339904,
-            "median_context_p99": 15.602153778076172,
-            "p90_context_kld": 3.0638998637147656,
+            "median_context_kld": 2.1021449397322147,
+            "median_context_p99": 15.483391761779785,
+            "p90_context_kld": 3.1956461064032,
             "positions": 196512,
-            "top1_agreement": 0.605067375020355,
+            "top1_agreement": 0.6049401563263312,
             "worst_context_id": 520,
-            "worst_context_kld": 4.1572436834165725
+            "worst_context_kld": 3.9490839144441505
           },
           "mxfp8": {
             "contexts": 96,
@@ -366,7 +366,151 @@
         },
         "key": "wildchat",
         "label": "wildchat",
-        "relative_to_run": 1.1418100527117467
+        "relative_to_run": 1.1372154577295144
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 39,
+            "max_kld": 44.86530303955078,
+            "mean_kld": 1.4989135420459287,
+            "mean_ref_top1_prob": 0.6181785541207915,
+            "median_context_kld": 1.5176459682812242,
+            "median_context_p99": 10.4518461227417,
+            "p90_context_kld": 2.0562577177214045,
+            "positions": 79833,
+            "top1_agreement": 0.5653927573810329,
+            "worst_context_id": 894,
+            "worst_context_kld": 2.545149489019375
+          },
+          "bxq": {
+            "contexts": 39,
+            "max_kld": 34.798072814941406,
+            "mean_kld": 1.7719551250418042,
+            "mean_ref_top1_prob": 0.6181785541207915,
+            "median_context_kld": 1.9004661327607555,
+            "median_context_p99": 10.832013130187988,
+            "p90_context_kld": 2.4125418709857684,
+            "positions": 79833,
+            "top1_agreement": 0.5090877206167875,
+            "worst_context_id": 838,
+            "worst_context_kld": 2.4818499093282402
+          },
+          "deployed": {
+            "contexts": 39,
+            "max_kld": 38.0394401550293,
+            "mean_kld": 2.067425442977532,
+            "mean_ref_top1_prob": 0.6181785541207915,
+            "median_context_kld": 2.1079394514301444,
+            "median_context_p99": 12.09659194946289,
+            "p90_context_kld": 2.864356055391028,
+            "positions": 79833,
+            "top1_agreement": 0.47963874588202876,
+            "worst_context_id": 838,
+            "worst_context_kld": 3.051148267918921
+          },
+          "mxfp8": {
+            "contexts": 39,
+            "max_kld": 25.854270935058594,
+            "mean_kld": 0.4807795312693682,
+            "mean_ref_top1_prob": 0.6181785541207915,
+            "median_context_kld": 0.47554080354894257,
+            "median_context_p99": 5.874552249908447,
+            "p90_context_kld": 0.8254337277877202,
+            "positions": 79833,
+            "top1_agreement": 0.7659990229604299,
+            "worst_context_id": 904,
+            "worst_context_kld": 1.1074666466827954
+          },
+          "nvfp4": {
+            "contexts": 39,
+            "max_kld": 36.620304107666016,
+            "mean_kld": 0.7304099307874191,
+            "mean_ref_top1_prob": 0.6181785541207915,
+            "median_context_kld": 0.747591067729384,
+            "median_context_p99": 7.896267414093018,
+            "p90_context_kld": 1.1061889984972806,
+            "positions": 79833,
+            "top1_agreement": 0.7005749502085604,
+            "worst_context_id": 904,
+            "worst_context_kld": 1.445864427497245
+          }
+        },
+        "key": "wikipedia_zh",
+        "label": "wikipedia_zh",
+        "relative_to_run": 1.1341942764675215
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 96,
+            "max_kld": 36.54373550415039,
+            "mean_kld": 1.5680853907689565,
+            "mean_ref_top1_prob": 0.6868049425612657,
+            "median_context_kld": 1.5512256124312427,
+            "median_context_p99": 12.015377044677734,
+            "p90_context_kld": 2.152355938027465,
+            "positions": 196512,
+            "top1_agreement": 0.5967014737013516,
+            "worst_context_id": 36,
+            "worst_context_kld": 2.582525214020106
+          },
+          "bxq": {
+            "contexts": 96,
+            "max_kld": 39.378456115722656,
+            "mean_kld": 1.7222270669528827,
+            "mean_ref_top1_prob": 0.6868049425612657,
+            "median_context_kld": 1.7462169184915457,
+            "median_context_p99": 11.825867652893066,
+            "p90_context_kld": 2.449975450141369,
+            "positions": 196512,
+            "top1_agreement": 0.5567039163002768,
+            "worst_context_id": 114,
+            "worst_context_kld": 2.796517373867177
+          },
+          "deployed": {
+            "contexts": 96,
+            "max_kld": 38.341392517089844,
+            "mean_kld": 2.028686595381578,
+            "mean_ref_top1_prob": 0.6868049425612657,
+            "median_context_kld": 2.0537626696728117,
+            "median_context_p99": 13.110386848449707,
+            "p90_context_kld": 2.7314164504667997,
+            "positions": 196512,
+            "top1_agreement": 0.5299116593388699,
+            "worst_context_id": 4,
+            "worst_context_kld": 3.506572107144654
+          },
+          "mxfp8": {
+            "contexts": 96,
+            "max_kld": 37.196868896484375,
+            "mean_kld": 0.4974304449139752,
+            "mean_ref_top1_prob": 0.6868049425612657,
+            "median_context_kld": 0.4826045789500282,
+            "median_context_p99": 6.104454040527344,
+            "p90_context_kld": 0.8085046390465469,
+            "positions": 196512,
+            "top1_agreement": 0.7871122374206155,
+            "worst_context_id": 80,
+            "worst_context_kld": 1.7253968548024476
+          },
+          "nvfp4": {
+            "contexts": 96,
+            "max_kld": 33.947750091552734,
+            "mean_kld": 0.7195422517772921,
+            "mean_ref_top1_prob": 0.6868049425612657,
+            "median_context_kld": 0.7051433644363765,
+            "median_context_p99": 7.863751411437988,
+            "p90_context_kld": 1.1927853316023667,
+            "positions": 196512,
+            "top1_agreement": 0.7338941133365902,
+            "worst_context_id": 53,
+            "worst_context_kld": 1.7009350418331064
+          }
+        },
+        "key": "wikipedia_en",
+        "label": "wikipedia_en",
+        "relative_to_run": 1.1129420570128747
       },
       {
         "cells": {
@@ -385,29 +529,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 40.41075897216797,
-            "mean_kld": 1.8291251072838597,
+            "max_kld": 44.15528106689453,
+            "mean_kld": 1.787203256659065,
             "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 1.8503912588432168,
-            "median_context_p99": 12.141514778137207,
-            "p90_context_kld": 2.4894801961682105,
+            "median_context_kld": 1.8031902684070498,
+            "median_context_p99": 11.896342277526855,
+            "p90_context_kld": 2.5275857576081804,
             "positions": 196512,
-            "top1_agreement": 0.5527550480377789,
-            "worst_context_id": 155,
-            "worst_context_kld": 3.0918058855236774
+            "top1_agreement": 0.5591617814688161,
+            "worst_context_id": 208,
+            "worst_context_kld": 3.163597909257638
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 41.034908294677734,
-            "mean_kld": 2.0747903506467162,
+            "max_kld": 43.10159683227539,
+            "mean_kld": 2.025307995636494,
             "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 2.1466476891456874,
-            "median_context_p99": 13.480792999267578,
-            "p90_context_kld": 2.8736326533513354,
+            "median_context_kld": 2.0841908944410252,
+            "median_context_p99": 13.38927936553955,
+            "p90_context_kld": 2.762181235851715,
             "positions": 196512,
-            "top1_agreement": 0.5339063263312164,
-            "worst_context_id": 155,
-            "worst_context_kld": 3.3824833858111054
+            "top1_agreement": 0.5387609916951637,
+            "worst_context_id": 208,
+            "worst_context_kld": 3.514232290307843
           },
           "mxfp8": {
             "contexts": 96,
@@ -438,7 +582,7 @@
         },
         "key": "scientific_papers",
         "label": "scientific_papers",
-        "relative_to_run": 1.1407447815298883
+        "relative_to_run": 1.111088549547169
       },
       {
         "cells": {
@@ -457,29 +601,29 @@
           },
           "bxq": {
             "contexts": 23,
-            "max_kld": 40.05707931518555,
-            "mean_kld": 1.7610534151467816,
+            "max_kld": 43.00507736206055,
+            "mean_kld": 1.707301661391642,
             "mean_ref_top1_prob": 0.6670678899954019,
-            "median_context_kld": 1.8086489990267063,
-            "median_context_p99": 11.264810562133789,
-            "p90_context_kld": 2.657902781678106,
+            "median_context_kld": 1.729092766784372,
+            "median_context_p99": 11.210553169250488,
+            "p90_context_kld": 2.4765564234721924,
             "positions": 47081,
-            "top1_agreement": 0.5491174783883095,
+            "top1_agreement": 0.5551708757248147,
             "worst_context_id": 304,
-            "worst_context_kld": 3.25593660154531
+            "worst_context_kld": 3.1671384824877022
           },
           "deployed": {
             "contexts": 23,
-            "max_kld": 42.641807556152344,
-            "mean_kld": 2.0666366538064116,
+            "max_kld": 35.000213623046875,
+            "mean_kld": 1.9460739920823162,
             "mean_ref_top1_prob": 0.6670678899954019,
-            "median_context_kld": 1.991214396928685,
-            "median_context_p99": 12.774897575378418,
-            "p90_context_kld": 3.14055686713693,
+            "median_context_kld": 2.0447724181031313,
+            "median_context_p99": 13.449270248413086,
+            "p90_context_kld": 2.7793235065102526,
             "positions": 47081,
-            "top1_agreement": 0.5225250100889955,
-            "worst_context_id": 310,
-            "worst_context_kld": 3.3839143853574014
+            "top1_agreement": 0.5389860028461587,
+            "worst_context_id": 337,
+            "worst_context_kld": 3.570137918407335
           },
           "mxfp8": {
             "contexts": 23,
@@ -510,151 +654,7 @@
         },
         "key": "open_news",
         "label": "open_news",
-        "relative_to_run": 1.136261780576151
-      },
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 96,
-            "max_kld": 36.54373550415039,
-            "mean_kld": 1.5680853907689565,
-            "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 1.5512256124312427,
-            "median_context_p99": 12.015377044677734,
-            "p90_context_kld": 2.152355938027465,
-            "positions": 196512,
-            "top1_agreement": 0.5967014737013516,
-            "worst_context_id": 36,
-            "worst_context_kld": 2.582525214020106
-          },
-          "bxq": {
-            "contexts": 96,
-            "max_kld": 41.13975143432617,
-            "mean_kld": 1.736663274181469,
-            "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 1.818968936428913,
-            "median_context_p99": 12.054352760314941,
-            "p90_context_kld": 2.379043205794732,
-            "positions": 196512,
-            "top1_agreement": 0.5558998941540466,
-            "worst_context_id": 45,
-            "worst_context_kld": 2.891395312851532
-          },
-          "deployed": {
-            "contexts": 96,
-            "max_kld": 46.643943786621094,
-            "mean_kld": 2.023094138776235,
-            "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 2.0884355776743306,
-            "median_context_p99": 13.014652252197266,
-            "p90_context_kld": 2.7440561136428148,
-            "positions": 196512,
-            "top1_agreement": 0.5285020762090865,
-            "worst_context_id": 4,
-            "worst_context_kld": 3.3786182444198056
-          },
-          "mxfp8": {
-            "contexts": 96,
-            "max_kld": 37.196868896484375,
-            "mean_kld": 0.4974304449139752,
-            "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 0.4826045789500282,
-            "median_context_p99": 6.104454040527344,
-            "p90_context_kld": 0.8085046390465469,
-            "positions": 196512,
-            "top1_agreement": 0.7871122374206155,
-            "worst_context_id": 80,
-            "worst_context_kld": 1.7253968548024476
-          },
-          "nvfp4": {
-            "contexts": 96,
-            "max_kld": 33.947750091552734,
-            "mean_kld": 0.7195422517772921,
-            "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 0.7051433644363765,
-            "median_context_p99": 7.863751411437988,
-            "p90_context_kld": 1.1927853316023667,
-            "positions": 196512,
-            "top1_agreement": 0.7338941133365902,
-            "worst_context_id": 53,
-            "worst_context_kld": 1.7009350418331064
-          }
-        },
-        "key": "wikipedia_en",
-        "label": "wikipedia_en",
-        "relative_to_run": 1.112321580169937
-      },
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 39,
-            "max_kld": 44.86530303955078,
-            "mean_kld": 1.4989135420459287,
-            "mean_ref_top1_prob": 0.6181785541207915,
-            "median_context_kld": 1.5176459682812242,
-            "median_context_p99": 10.4518461227417,
-            "p90_context_kld": 2.0562577177214045,
-            "positions": 79833,
-            "top1_agreement": 0.5653927573810329,
-            "worst_context_id": 894,
-            "worst_context_kld": 2.545149489019375
-          },
-          "bxq": {
-            "contexts": 39,
-            "max_kld": 42.90846252441406,
-            "mean_kld": 1.7958318514122826,
-            "mean_ref_top1_prob": 0.6181785541207915,
-            "median_context_kld": 1.885194908223622,
-            "median_context_p99": 10.767505645751953,
-            "p90_context_kld": 2.4585992730885,
-            "positions": 79833,
-            "top1_agreement": 0.5110919043503314,
-            "worst_context_id": 904,
-            "worst_context_kld": 3.1248215177600276
-          },
-          "deployed": {
-            "contexts": 39,
-            "max_kld": 31.211334228515625,
-            "mean_kld": 1.9891904709791275,
-            "mean_ref_top1_prob": 0.6181785541207915,
-            "median_context_kld": 2.0918872848284886,
-            "median_context_p99": 11.928448677062988,
-            "p90_context_kld": 2.7211136485306593,
-            "positions": 79833,
-            "top1_agreement": 0.4875176931845227,
-            "worst_context_id": 916,
-            "worst_context_kld": 2.93961666449336
-          },
-          "mxfp8": {
-            "contexts": 39,
-            "max_kld": 25.854270935058594,
-            "mean_kld": 0.4807795312693682,
-            "mean_ref_top1_prob": 0.6181785541207915,
-            "median_context_kld": 0.47554080354894257,
-            "median_context_p99": 5.874552249908447,
-            "p90_context_kld": 0.8254337277877202,
-            "positions": 79833,
-            "top1_agreement": 0.7659990229604299,
-            "worst_context_id": 904,
-            "worst_context_kld": 1.1074666466827954
-          },
-          "nvfp4": {
-            "contexts": 39,
-            "max_kld": 36.620304107666016,
-            "mean_kld": 0.7304099307874191,
-            "mean_ref_top1_prob": 0.6181785541207915,
-            "median_context_kld": 0.747591067729384,
-            "median_context_p99": 7.896267414093018,
-            "p90_context_kld": 1.1061889984972806,
-            "positions": 79833,
-            "top1_agreement": 0.7005749502085604,
-            "worst_context_id": 904,
-            "worst_context_kld": 1.445864427497245
-          }
-        },
-        "key": "wikipedia_zh",
-        "label": "wikipedia_zh",
-        "relative_to_run": 1.093680934332049
+        "relative_to_run": 1.0676205958959224
       },
       {
         "cells": {
@@ -673,29 +673,29 @@
           },
           "bxq": {
             "contexts": 7,
-            "max_kld": 27.038440704345703,
-            "mean_kld": 1.595668133407234,
+            "max_kld": 26.193904876708984,
+            "mean_kld": 1.6661806227520861,
             "mean_ref_top1_prob": 0.6964561335831647,
-            "median_context_kld": 1.6504266919963024,
-            "median_context_p99": 10.824897766113281,
-            "p90_context_kld": 1.9814749104151195,
+            "median_context_kld": 1.6014363500002338,
+            "median_context_p99": 11.498273849487305,
+            "p90_context_kld": 2.012908883131005,
             "positions": 14329,
-            "top1_agreement": 0.5761044036569195,
+            "top1_agreement": 0.566264219415172,
             "worst_context_id": 968,
-            "worst_context_kld": 1.9814749104151195
+            "worst_context_kld": 2.012908883131005
           },
           "deployed": {
             "contexts": 7,
-            "max_kld": 23.746353149414062,
-            "mean_kld": 1.8962839492114514,
+            "max_kld": 27.235666275024414,
+            "mean_kld": 1.937432643860255,
             "mean_ref_top1_prob": 0.6964561335831647,
-            "median_context_kld": 1.8275499984532189,
-            "median_context_p99": 12.621654510498047,
-            "p90_context_kld": 2.69385492139005,
+            "median_context_kld": 1.9339302284016013,
+            "median_context_p99": 13.26597785949707,
+            "p90_context_kld": 2.474590089885823,
             "positions": 14329,
-            "top1_agreement": 0.5553772070626003,
-            "worst_context_id": 968,
-            "worst_context_kld": 2.69385492139005
+            "top1_agreement": 0.546583850931677,
+            "worst_context_id": 932,
+            "worst_context_kld": 2.474590089885823
           },
           "mxfp8": {
             "contexts": 7,
@@ -726,7 +726,7 @@
         },
         "key": "wikipedia_ja",
         "label": "wikipedia_ja",
-        "relative_to_run": 1.0425998071022378
+        "relative_to_run": 1.0628799327064873
       },
       {
         "cells": {
@@ -745,29 +745,29 @@
           },
           "bxq": {
             "contexts": 26,
-            "max_kld": 30.751243591308594,
-            "mean_kld": 1.5551117815921176,
+            "max_kld": 31.572242736816406,
+            "mean_kld": 1.5861372701441745,
             "mean_ref_top1_prob": 0.6452037802048386,
-            "median_context_kld": 1.5219009448575713,
-            "median_context_p99": 10.262850761413574,
-            "p90_context_kld": 2.2438624677555334,
+            "median_context_kld": 1.5233652582278852,
+            "median_context_p99": 10.139749526977539,
+            "p90_context_kld": 2.4009817014978534,
             "positions": 53222,
-            "top1_agreement": 0.5566119273984442,
+            "top1_agreement": 0.5576829130810567,
             "worst_context_id": 339,
-            "worst_context_kld": 2.7539297546599255
+            "worst_context_kld": 2.831962915742372
           },
           "deployed": {
             "contexts": 26,
-            "max_kld": 38.074745178222656,
-            "mean_kld": 1.795426876451551,
+            "max_kld": 37.007591247558594,
+            "mean_kld": 1.8328970269387121,
             "mean_ref_top1_prob": 0.6452037802048386,
-            "median_context_kld": 1.8161399239846208,
-            "median_context_p99": 12.137351036071777,
-            "p90_context_kld": 2.6750341700629416,
+            "median_context_kld": 1.8116546152001858,
+            "median_context_p99": 11.446545600891113,
+            "p90_context_kld": 2.6563850069971284,
             "positions": 53222,
-            "top1_agreement": 0.5383488031265267,
+            "top1_agreement": 0.5303633835631881,
             "worst_context_id": 339,
-            "worst_context_kld": 3.235694817899114
+            "worst_context_kld": 4.061378358077808
           },
           "mxfp8": {
             "contexts": 26,
@@ -798,7 +798,79 @@
         },
         "key": "public_domain_review",
         "label": "public_domain_review",
-        "relative_to_run": 0.9871473709583284
+        "relative_to_run": 1.0055314567059896
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 7,
+            "max_kld": 30.324600219726562,
+            "mean_kld": 1.3557814874016374,
+            "mean_ref_top1_prob": 0.6783229950499561,
+            "median_context_kld": 1.1565023463317496,
+            "median_context_p99": 11.258171081542969,
+            "p90_context_kld": 2.8760561197873624,
+            "positions": 14329,
+            "top1_agreement": 0.6365412799218368,
+            "worst_context_id": 942,
+            "worst_context_kld": 2.8760561197873624
+          },
+          "bxq": {
+            "contexts": 7,
+            "max_kld": 26.802766799926758,
+            "mean_kld": 1.366573651235621,
+            "mean_ref_top1_prob": 0.6783229950499561,
+            "median_context_kld": 1.6788156536905139,
+            "median_context_p99": 10.924899101257324,
+            "p90_context_kld": 2.2242239348285806,
+            "positions": 14329,
+            "top1_agreement": 0.6100914229883453,
+            "worst_context_id": 972,
+            "worst_context_kld": 2.2242239348285806
+          },
+          "deployed": {
+            "contexts": 7,
+            "max_kld": 25.41287612915039,
+            "mean_kld": 1.7258839744647188,
+            "mean_ref_top1_prob": 0.6783229950499561,
+            "median_context_kld": 2.24875410781004,
+            "median_context_p99": 13.35194206237793,
+            "p90_context_kld": 2.653788978496691,
+            "positions": 14329,
+            "top1_agreement": 0.5816874869146487,
+            "worst_context_id": 942,
+            "worst_context_kld": 2.653788978496691
+          },
+          "mxfp8": {
+            "contexts": 7,
+            "max_kld": 28.36335563659668,
+            "mean_kld": 0.4384885917166796,
+            "mean_ref_top1_prob": 0.6783229950499561,
+            "median_context_kld": 0.28109426868333354,
+            "median_context_p99": 4.122045516967773,
+            "p90_context_kld": 1.2080682259526418,
+            "positions": 14329,
+            "top1_agreement": 0.8062670109568009,
+            "worst_context_id": 972,
+            "worst_context_kld": 1.2080682259526418
+          },
+          "nvfp4": {
+            "contexts": 7,
+            "max_kld": 28.87299919128418,
+            "mean_kld": 0.52814768643654,
+            "mean_ref_top1_prob": 0.6783229950499561,
+            "median_context_kld": 0.3326322147542188,
+            "median_context_p99": 4.5705671310424805,
+            "p90_context_kld": 1.0986080994352987,
+            "positions": 14329,
+            "top1_agreement": 0.7693488729150674,
+            "worst_context_id": 972,
+            "worst_context_kld": 1.0986080994352987
+          }
+        },
+        "key": "wikipedia_es",
+        "label": "wikipedia_es",
+        "relative_to_run": 0.9468238539551412
       },
       {
         "cells": {
@@ -817,29 +889,29 @@
           },
           "bxq": {
             "contexts": 33,
-            "max_kld": 32.46535110473633,
-            "mean_kld": 1.5230550719494145,
+            "max_kld": 29.982336044311523,
+            "mean_kld": 1.5394993128451908,
             "mean_ref_top1_prob": 0.6446389525396158,
-            "median_context_kld": 1.4728623104069254,
-            "median_context_p99": 9.21851634979248,
-            "p90_context_kld": 2.37556835422248,
+            "median_context_kld": 1.498481160053086,
+            "median_context_p99": 8.978272438049316,
+            "p90_context_kld": 2.3378235146521353,
             "positions": 67551,
-            "top1_agreement": 0.559606815591183,
+            "top1_agreement": 0.5551805302660212,
             "worst_context_id": 925,
-            "worst_context_kld": 2.5572512492636292
+            "worst_context_kld": 2.368994362874638
           },
           "deployed": {
             "contexts": 33,
-            "max_kld": 39.30211639404297,
-            "mean_kld": 1.6837435207264078,
+            "max_kld": 43.24997329711914,
+            "mean_kld": 1.70928190402215,
             "mean_ref_top1_prob": 0.6446389525396158,
-            "median_context_kld": 1.6837463906299435,
-            "median_context_p99": 9.897921562194824,
-            "p90_context_kld": 2.244692611545905,
+            "median_context_kld": 1.6604002016192938,
+            "median_context_p99": 10.152688026428223,
+            "p90_context_kld": 2.3281872730597724,
             "positions": 67551,
-            "top1_agreement": 0.54341164453524,
+            "top1_agreement": 0.5401104350786813,
             "worst_context_id": 925,
-            "worst_context_kld": 2.942326232320723
+            "worst_context_kld": 3.1738207166754964
           },
           "mxfp8": {
             "contexts": 33,
@@ -870,7 +942,7 @@
         },
         "key": "wikisource_zh",
         "label": "wikisource_zh",
-        "relative_to_run": 0.9257425137458916
+        "relative_to_run": 0.9377159205409366
       },
       {
         "cells": {
@@ -889,29 +961,29 @@
           },
           "bxq": {
             "contexts": 72,
-            "max_kld": 32.1933479309082,
-            "mean_kld": 1.4946107758018452,
+            "max_kld": 42.736114501953125,
+            "mean_kld": 1.5024764906905146,
             "mean_ref_top1_prob": 0.6303461857865943,
-            "median_context_kld": 1.500009053620101,
-            "median_context_p99": 9.486432075500488,
-            "p90_context_kld": 2.3340939446212148,
+            "median_context_kld": 1.524894262356297,
+            "median_context_p99": 9.397890090942383,
+            "p90_context_kld": 2.2545776405591753,
             "positions": 147384,
-            "top1_agreement": 0.5584595342778049,
-            "worst_context_id": 388,
-            "worst_context_kld": 2.7210277162281766
+            "top1_agreement": 0.560467893394127,
+            "worst_context_id": 439,
+            "worst_context_kld": 3.1497801563410865
           },
           "deployed": {
             "contexts": 72,
-            "max_kld": 38.19701385498047,
-            "mean_kld": 1.6581158258948987,
+            "max_kld": 38.11989212036133,
+            "mean_kld": 1.7055569548461247,
             "mean_ref_top1_prob": 0.6303461857865943,
-            "median_context_kld": 1.7261176263246027,
-            "median_context_p99": 10.789175033569336,
-            "p90_context_kld": 2.4446691744187006,
+            "median_context_kld": 1.7174336067742633,
+            "median_context_p99": 10.889728546142578,
+            "p90_context_kld": 2.5845404803262886,
             "positions": 147384,
-            "top1_agreement": 0.5436546707919448,
-            "worst_context_id": 388,
-            "worst_context_kld": 3.081658722460763
+            "top1_agreement": 0.5379891982847528,
+            "worst_context_id": 439,
+            "worst_context_kld": 3.1001899896399703
           },
           "mxfp8": {
             "contexts": 72,
@@ -942,7 +1014,7 @@
         },
         "key": "public_domain_books",
         "label": "public_domain_books",
-        "relative_to_run": 0.9116520977515372
+        "relative_to_run": 0.93567240499365
       },
       {
         "cells": {
@@ -961,29 +1033,29 @@
           },
           "bxq": {
             "contexts": 23,
-            "max_kld": 29.142873764038086,
-            "mean_kld": 1.4135229026365408,
+            "max_kld": 32.18777847290039,
+            "mean_kld": 1.3472795071741162,
             "mean_ref_top1_prob": 0.7590483138412703,
-            "median_context_kld": 1.4962088109376634,
-            "median_context_p99": 10.456581115722656,
-            "p90_context_kld": 2.075816721653685,
+            "median_context_kld": 1.3778201581111544,
+            "median_context_p99": 9.770347595214844,
+            "p90_context_kld": 2.0772022397851937,
             "positions": 47081,
-            "top1_agreement": 0.6318260019965591,
-            "worst_context_id": 257,
-            "worst_context_kld": 2.9318732747040346
+            "top1_agreement": 0.6348208406788302,
+            "worst_context_id": 281,
+            "worst_context_kld": 2.5506611948721116
           },
           "deployed": {
             "contexts": 23,
-            "max_kld": 35.14911651611328,
-            "mean_kld": 1.6310768855710862,
+            "max_kld": 28.261322021484375,
+            "mean_kld": 1.6435152613497837,
             "mean_ref_top1_prob": 0.7590483138412703,
-            "median_context_kld": 1.7073620463209902,
-            "median_context_p99": 11.311198234558105,
-            "p90_context_kld": 2.3911826341192004,
+            "median_context_kld": 1.8677532599153965,
+            "median_context_p99": 11.690622329711914,
+            "p90_context_kld": 2.478815819713035,
             "positions": 47081,
-            "top1_agreement": 0.6096089717720524,
+            "top1_agreement": 0.6109895711645887,
             "worst_context_id": 257,
-            "worst_context_kld": 3.4397168545750416
+            "worst_context_kld": 3.2918247335656012
           },
           "mxfp8": {
             "contexts": 23,
@@ -997,96 +1069,24 @@
             "top1_agreement": 0.8099870436057008,
             "worst_context_id": 326,
             "worst_context_kld": 1.3840455316931175
-          },
-          "nvfp4": {
-            "contexts": 23,
-            "max_kld": 27.018136978149414,
-            "mean_kld": 0.5421799561781261,
-            "mean_ref_top1_prob": 0.7590483138412703,
-            "median_context_kld": 0.476610005990993,
-            "median_context_p99": 5.588261604309082,
-            "p90_context_kld": 0.9304680026896384,
-            "positions": 47081,
-            "top1_agreement": 0.7833945753063869,
-            "worst_context_id": 257,
-            "worst_context_kld": 1.2764761217314473
-          }
-        },
-        "key": "regulations",
-        "label": "regulations",
-        "relative_to_run": 0.8967857619490438
-      },
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 7,
-            "max_kld": 30.324600219726562,
-            "mean_kld": 1.3557814874016374,
-            "mean_ref_top1_prob": 0.6783229950499561,
-            "median_context_kld": 1.1565023463317496,
-            "median_context_p99": 11.258171081542969,
-            "p90_context_kld": 2.8760561197873624,
-            "positions": 14329,
-            "top1_agreement": 0.6365412799218368,
-            "worst_context_id": 942,
-            "worst_context_kld": 2.8760561197873624
-          },
-          "bxq": {
-            "contexts": 7,
-            "max_kld": 31.311445236206055,
-            "mean_kld": 1.2539216444090706,
-            "mean_ref_top1_prob": 0.6783229950499561,
-            "median_context_kld": 1.1608309682978981,
-            "median_context_p99": 8.832525253295898,
-            "p90_context_kld": 2.1513075835982343,
-            "positions": 14329,
-            "top1_agreement": 0.6273291925465838,
-            "worst_context_id": 972,
-            "worst_context_kld": 2.1513075835982343
-          },
-          "deployed": {
-            "contexts": 7,
-            "max_kld": 34.88376998901367,
-            "mean_kld": 1.5977484929865589,
-            "mean_ref_top1_prob": 0.6783229950499561,
-            "median_context_kld": 1.3127510703777805,
-            "median_context_p99": 10.077347755432129,
-            "p90_context_kld": 2.78642329790802,
-            "positions": 14329,
-            "top1_agreement": 0.5910391513713448,
-            "worst_context_id": 954,
-            "worst_context_kld": 2.78642329790802
-          },
-          "mxfp8": {
-            "contexts": 7,
-            "max_kld": 28.36335563659668,
-            "mean_kld": 0.4384885917166796,
-            "mean_ref_top1_prob": 0.6783229950499561,
-            "median_context_kld": 0.28109426868333354,
-            "median_context_p99": 4.122045516967773,
-            "p90_context_kld": 1.2080682259526418,
-            "positions": 14329,
-            "top1_agreement": 0.8062670109568009,
-            "worst_context_id": 972,
-            "worst_context_kld": 1.2080682259526418
-          },
-          "nvfp4": {
-            "contexts": 7,
-            "max_kld": 28.87299919128418,
-            "mean_kld": 0.52814768643654,
-            "mean_ref_top1_prob": 0.6783229950499561,
-            "median_context_kld": 0.3326322147542188,
-            "median_context_p99": 4.5705671310424805,
-            "p90_context_kld": 1.0986080994352987,
-            "positions": 14329,
-            "top1_agreement": 0.7693488729150674,
-            "worst_context_id": 972,
-            "worst_context_kld": 1.0986080994352987
+          },
+          "nvfp4": {
+            "contexts": 23,
+            "max_kld": 27.018136978149414,
+            "mean_kld": 0.5421799561781261,
+            "mean_ref_top1_prob": 0.7590483138412703,
+            "median_context_kld": 0.476610005990993,
+            "median_context_p99": 5.588261604309082,
+            "p90_context_kld": 0.9304680026896384,
+            "positions": 47081,
+            "top1_agreement": 0.7833945753063869,
+            "worst_context_id": 257,
+            "worst_context_kld": 1.2764761217314473
           }
         },
-        "key": "wikipedia_es",
-        "label": "wikipedia_es",
-        "relative_to_run": 0.87846140936772
+        "key": "regulations",
+        "label": "regulations",
+        "relative_to_run": 0.9016361915452181
       },
       {
         "cells": {
@@ -1105,29 +1105,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 41.494056701660156,
-            "mean_kld": 1.3686865325159163,
+            "max_kld": 38.396728515625,
+            "mean_kld": 1.3358294237073802,
             "mean_ref_top1_prob": 0.7233971787516119,
-            "median_context_kld": 1.2778304976400205,
-            "median_context_p99": 10.17981243133545,
-            "p90_context_kld": 2.134711669512087,
+            "median_context_kld": 1.2826492898134843,
+            "median_context_p99": 9.845369338989258,
+            "p90_context_kld": 2.00187886403533,
             "positions": 196512,
-            "top1_agreement": 0.6207152743852793,
+            "top1_agreement": 0.6250712424686533,
             "worst_context_id": 708,
-            "worst_context_kld": 2.9879818448351
+            "worst_context_kld": 2.7549330833305397
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 34.1971549987793,
-            "mean_kld": 1.560781639985526,
+            "max_kld": 45.10734176635742,
+            "mean_kld": 1.546925506292042,
             "mean_ref_top1_prob": 0.7233971787516119,
-            "median_context_kld": 1.473631995802428,
-            "median_context_p99": 11.064117431640625,
-            "p90_context_kld": 2.313850639680717,
+            "median_context_kld": 1.5185987562235763,
+            "median_context_p99": 11.175138473510742,
+            "p90_context_kld": 2.279972149569506,
             "positions": 196512,
-            "top1_agreement": 0.6012406367041199,
+            "top1_agreement": 0.6033880882592412,
             "worst_context_id": 708,
-            "worst_context_kld": 3.254033478573557
+            "worst_context_kld": 3.2894278163579322
           },
           "mxfp8": {
             "contexts": 96,
@@ -1158,79 +1158,7 @@
         },
         "key": "libretexts",
         "label": "libretexts",
-        "relative_to_run": 0.8581365873261262
-      },
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 44,
-            "max_kld": 31.8831787109375,
-            "mean_kld": 1.0957652045871973,
-            "mean_ref_top1_prob": 0.7986141030003556,
-            "median_context_kld": 1.113559921679193,
-            "median_context_p99": 10.494965553283691,
-            "p90_context_kld": 1.6013505308745482,
-            "positions": 90068,
-            "top1_agreement": 0.7165141892792113,
-            "worst_context_id": 658,
-            "worst_context_kld": 2.033105903094215
-          },
-          "bxq": {
-            "contexts": 44,
-            "max_kld": 34.63332748413086,
-            "mean_kld": 1.196611976685152,
-            "mean_ref_top1_prob": 0.7986141030003556,
-            "median_context_kld": 1.210198833030204,
-            "median_context_p99": 10.698575973510742,
-            "p90_context_kld": 1.866644513191281,
-            "positions": 90068,
-            "top1_agreement": 0.689623395656615,
-            "worst_context_id": 680,
-            "worst_context_kld": 2.4041545637647666
-          },
-          "deployed": {
-            "contexts": 44,
-            "max_kld": 31.547298431396484,
-            "mean_kld": 1.4436261843316183,
-            "mean_ref_top1_prob": 0.7986141030003556,
-            "median_context_kld": 1.5403832245691995,
-            "median_context_p99": 11.939769744873047,
-            "p90_context_kld": 2.1837179062227996,
-            "positions": 90068,
-            "top1_agreement": 0.6608118310609762,
-            "worst_context_id": 628,
-            "worst_context_kld": 2.8726303894076324
-          },
-          "mxfp8": {
-            "contexts": 44,
-            "max_kld": 25.849998474121094,
-            "mean_kld": 0.32388194098508744,
-            "mean_ref_top1_prob": 0.7986141030003556,
-            "median_context_kld": 0.29406276775443013,
-            "median_context_p99": 4.61008358001709,
-            "p90_context_kld": 0.5193967982707784,
-            "positions": 90068,
-            "top1_agreement": 0.8578518452724608,
-            "worst_context_id": 590,
-            "worst_context_kld": 0.7979057760773107
-          },
-          "nvfp4": {
-            "contexts": 44,
-            "max_kld": 26.16546058654785,
-            "mean_kld": 0.47263660588120515,
-            "mean_ref_top1_prob": 0.7986141030003556,
-            "median_context_kld": 0.4749053284319568,
-            "median_context_p99": 6.230027198791504,
-            "p90_context_kld": 0.7040379286701839,
-            "positions": 90068,
-            "top1_agreement": 0.8229670915308434,
-            "worst_context_id": 590,
-            "worst_context_kld": 1.0825279485134078
-          }
-        },
-        "key": "stackv2",
-        "label": "stackv2",
-        "relative_to_run": 0.7937231035139933
+        "relative_to_run": 0.8486468333441731
       },
       {
         "cells": {
@@ -1249,29 +1177,29 @@
           },
           "bxq": {
             "contexts": 36,
-            "max_kld": 28.927227020263672,
-            "mean_kld": 1.2778636800520669,
+            "max_kld": 33.88340759277344,
+            "mean_kld": 1.2661847644547204,
             "mean_ref_top1_prob": 0.8326448152069219,
-            "median_context_kld": 1.1591657174536962,
-            "median_context_p99": 10.449883460998535,
-            "p90_context_kld": 1.9698587425136318,
+            "median_context_kld": 1.0651777554342132,
+            "median_context_p99": 10.165213584899902,
+            "p90_context_kld": 2.288437004973564,
             "positions": 73692,
-            "top1_agreement": 0.6831949193942355,
+            "top1_agreement": 0.6826114096509798,
             "worst_context_id": 1016,
-            "worst_context_kld": 2.813886272014663
+            "worst_context_kld": 2.661403536306777
           },
           "deployed": {
             "contexts": 36,
-            "max_kld": 32.71179962158203,
-            "mean_kld": 1.421633946558572,
+            "max_kld": 31.666025161743164,
+            "mean_kld": 1.4835057973768624,
             "mean_ref_top1_prob": 0.8326448152069219,
-            "median_context_kld": 1.1716843978003344,
-            "median_context_p99": 11.04438591003418,
-            "p90_context_kld": 2.2876916350565337,
+            "median_context_kld": 1.3026068426019195,
+            "median_context_p99": 11.720060348510742,
+            "p90_context_kld": 2.4576422275537153,
             "positions": 73692,
-            "top1_agreement": 0.6721218042664061,
+            "top1_agreement": 0.6591353199804592,
             "worst_context_id": 1016,
-            "worst_context_kld": 3.1927021256291948
+            "worst_context_kld": 3.1780537988024644
           },
           "mxfp8": {
             "contexts": 36,
@@ -1302,7 +1230,7 @@
         },
         "key": "starcoder_structured",
         "label": "starcoder_structured",
-        "relative_to_run": 0.7816315056974008
+        "relative_to_run": 0.8138546375186066
       },
       {
         "cells": {
@@ -1321,29 +1249,29 @@
           },
           "bxq": {
             "contexts": 52,
-            "max_kld": 43.123695373535156,
-            "mean_kld": 1.1808067984364206,
+            "max_kld": 35.210819244384766,
+            "mean_kld": 1.1990024557136765,
             "mean_ref_top1_prob": 0.8794735901070214,
-            "median_context_kld": 1.091582201404463,
-            "median_context_p99": 11.706655502319336,
-            "p90_context_kld": 1.9767263845151137,
+            "median_context_kld": 1.1524328843660978,
+            "median_context_p99": 11.834178924560547,
+            "p90_context_kld": 1.9367562142488,
             "positions": 106444,
-            "top1_agreement": 0.7440907895231296,
-            "worst_context_id": 631,
-            "worst_context_kld": 2.481150413090197
+            "top1_agreement": 0.7408684378640412,
+            "worst_context_id": 635,
+            "worst_context_kld": 2.496798226059531
           },
           "deployed": {
             "contexts": 52,
-            "max_kld": 38.950897216796875,
-            "mean_kld": 1.3498761812382902,
+            "max_kld": 41.40315246582031,
+            "mean_kld": 1.4158959782631781,
             "mean_ref_top1_prob": 0.8794735901070214,
-            "median_context_kld": 1.2660082146230165,
-            "median_context_p99": 13.014894485473633,
-            "p90_context_kld": 2.0756156860047312,
+            "median_context_kld": 1.3886546807947975,
+            "median_context_p99": 13.138496398925781,
+            "p90_context_kld": 2.124109961913988,
             "positions": 106444,
-            "top1_agreement": 0.7285521025139979,
+            "top1_agreement": 0.721637668633272,
             "worst_context_id": 635,
-            "worst_context_kld": 2.544323761465564
+            "worst_context_kld": 2.825086072791175
           },
           "mxfp8": {
             "contexts": 52,
@@ -1374,7 +1302,79 @@
         },
         "key": "github_code",
         "label": "github_code",
-        "relative_to_run": 0.7421782200688829
+        "relative_to_run": 0.7767637377561921
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 44,
+            "max_kld": 31.8831787109375,
+            "mean_kld": 1.0957652045871973,
+            "mean_ref_top1_prob": 0.7986141030003556,
+            "median_context_kld": 1.113559921679193,
+            "median_context_p99": 10.494965553283691,
+            "p90_context_kld": 1.6013505308745482,
+            "positions": 90068,
+            "top1_agreement": 0.7165141892792113,
+            "worst_context_id": 658,
+            "worst_context_kld": 2.033105903094215
+          },
+          "bxq": {
+            "contexts": 44,
+            "max_kld": 30.687049865722656,
+            "mean_kld": 1.1873244913889716,
+            "mean_ref_top1_prob": 0.7986141030003556,
+            "median_context_kld": 1.1496710022759318,
+            "median_context_p99": 9.833146095275879,
+            "p90_context_kld": 1.762736219529926,
+            "positions": 90068,
+            "top1_agreement": 0.6883132744148865,
+            "worst_context_id": 656,
+            "worst_context_kld": 2.2664268642779404
+          },
+          "deployed": {
+            "contexts": 44,
+            "max_kld": 39.51999282836914,
+            "mean_kld": 1.3872222137307078,
+            "mean_ref_top1_prob": 0.7986141030003556,
+            "median_context_kld": 1.3179673332330244,
+            "median_context_p99": 11.684328079223633,
+            "p90_context_kld": 2.0111891791915695,
+            "positions": 90068,
+            "top1_agreement": 0.6693831327441488,
+            "worst_context_id": 614,
+            "worst_context_kld": 2.377006486507535
+          },
+          "mxfp8": {
+            "contexts": 44,
+            "max_kld": 25.849998474121094,
+            "mean_kld": 0.32388194098508744,
+            "mean_ref_top1_prob": 0.7986141030003556,
+            "median_context_kld": 0.29406276775443013,
+            "median_context_p99": 4.61008358001709,
+            "p90_context_kld": 0.5193967982707784,
+            "positions": 90068,
+            "top1_agreement": 0.8578518452724608,
+            "worst_context_id": 590,
+            "worst_context_kld": 0.7979057760773107
+          },
+          "nvfp4": {
+            "contexts": 44,
+            "max_kld": 26.16546058654785,
+            "mean_kld": 0.47263660588120515,
+            "mean_ref_top1_prob": 0.7986141030003556,
+            "median_context_kld": 0.4749053284319568,
+            "median_context_p99": 6.230027198791504,
+            "p90_context_kld": 0.7040379286701839,
+            "positions": 90068,
+            "top1_agreement": 0.8229670915308434,
+            "worst_context_id": 590,
+            "worst_context_kld": 1.0825279485134078
+          }
+        },
+        "key": "stackv2",
+        "label": "stackv2",
+        "relative_to_run": 0.7610332456468044
       }
     ],
     "stratum": [
@@ -1395,29 +1395,29 @@
           },
           "bxq": {
             "contexts": 36,
-            "max_kld": 32.78110122680664,
-            "mean_kld": 2.010609236108725,
+            "max_kld": 32.99455261230469,
+            "mean_kld": 2.1263077593205724,
             "mean_ref_top1_prob": 0.7072548605424939,
-            "median_context_kld": 1.8578505931293814,
-            "median_context_p99": 12.251582145690918,
-            "p90_context_kld": 3.3174707252638598,
+            "median_context_kld": 1.9892246005107965,
+            "median_context_p99": 12.25793170928955,
+            "p90_context_kld": 3.513730331786002,
             "positions": 73692,
-            "top1_agreement": 0.5405335721652282,
-            "worst_context_id": 969,
-            "worst_context_kld": 3.804737185271593
+            "top1_agreement": 0.5307496064701731,
+            "worst_context_id": 975,
+            "worst_context_kld": 4.436682889352712
           },
           "deployed": {
             "contexts": 36,
-            "max_kld": 41.130943298339844,
-            "mean_kld": 2.3637873771391753,
+            "max_kld": 49.14295196533203,
+            "mean_kld": 2.364139314385358,
             "mean_ref_top1_prob": 0.7072548605424939,
-            "median_context_kld": 2.3372776174560403,
-            "median_context_p99": 13.823667526245117,
-            "p90_context_kld": 4.12772207818787,
+            "median_context_kld": 2.3165460351273297,
+            "median_context_p99": 14.349602699279785,
+            "p90_context_kld": 3.541496895815821,
             "positions": 73692,
-            "top1_agreement": 0.5128100743635673,
+            "top1_agreement": 0.5113445150084134,
             "worst_context_id": 946,
-            "worst_context_kld": 4.479432370259336
+            "worst_context_kld": 4.5225937464682
           },
           "mxfp8": {
             "contexts": 36,
@@ -1448,7 +1448,7 @@
         },
         "key": "other_multilingual",
         "label": "Other multilingual content",
-        "relative_to_run": 1.2996388354501642
+        "relative_to_run": 1.2969721777661531
       },
       {
         "cells": {
@@ -1467,29 +1467,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 42.54874801635742,
-            "mean_kld": 1.798070948314533,
+            "max_kld": 41.96346664428711,
+            "mean_kld": 1.832551607899988,
             "mean_ref_top1_prob": 0.7743913512965481,
-            "median_context_kld": 1.7649751579712167,
-            "median_context_p99": 14.137372970581055,
-            "p90_context_kld": 2.7188406299445234,
+            "median_context_kld": 1.8701310658748034,
+            "median_context_p99": 14.359136581420898,
+            "p90_context_kld": 2.840875715129075,
             "positions": 196512,
-            "top1_agreement": 0.6251323074417847,
-            "worst_context_id": 571,
-            "worst_context_kld": 3.539333405062736
+            "top1_agreement": 0.6200181159420289,
+            "worst_context_id": 484,
+            "worst_context_kld": 3.525342154276217
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 38.28725814819336,
-            "mean_kld": 2.0767278693667035,
+            "max_kld": 44.010807037353516,
+            "mean_kld": 2.072932495110033,
             "mean_ref_top1_prob": 0.7743913512965481,
-            "median_context_kld": 2.0825514215339904,
-            "median_context_p99": 15.602153778076172,
-            "p90_context_kld": 3.0638998637147656,
+            "median_context_kld": 2.1021449397322147,
+            "median_context_p99": 15.483391761779785,
+            "p90_context_kld": 3.1956461064032,
             "positions": 196512,
-            "top1_agreement": 0.605067375020355,
+            "top1_agreement": 0.6049401563263312,
             "worst_context_id": 520,
-            "worst_context_kld": 4.1572436834165725
+            "worst_context_kld": 3.9490839144441505
           },
           "mxfp8": {
             "contexts": 96,
@@ -1520,79 +1520,7 @@
         },
         "key": "dialogue_instruction",
         "label": "Natural dialogue, instruction following, and assistance",
-        "relative_to_run": 1.1418100527117467
-      },
-      {
-        "cells": {
-          "all_quantized": {
-            "contexts": 96,
-            "max_kld": 40.053466796875,
-            "mean_kld": 1.6850800175124105,
-            "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 1.6989074666767194,
-            "median_context_p99": 12.346853256225586,
-            "p90_context_kld": 2.368493804166054,
-            "positions": 196512,
-            "top1_agreement": 0.5902896515225533,
-            "worst_context_id": 208,
-            "worst_context_kld": 2.906335476682348
-          },
-          "bxq": {
-            "contexts": 96,
-            "max_kld": 40.41075897216797,
-            "mean_kld": 1.8291251072838597,
-            "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 1.8503912588432168,
-            "median_context_p99": 12.141514778137207,
-            "p90_context_kld": 2.4894801961682105,
-            "positions": 196512,
-            "top1_agreement": 0.5527550480377789,
-            "worst_context_id": 155,
-            "worst_context_kld": 3.0918058855236774
-          },
-          "deployed": {
-            "contexts": 96,
-            "max_kld": 41.034908294677734,
-            "mean_kld": 2.0747903506467162,
-            "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 2.1466476891456874,
-            "median_context_p99": 13.480792999267578,
-            "p90_context_kld": 2.8736326533513354,
-            "positions": 196512,
-            "top1_agreement": 0.5339063263312164,
-            "worst_context_id": 155,
-            "worst_context_kld": 3.3824833858111054
-          },
-          "mxfp8": {
-            "contexts": 96,
-            "max_kld": 31.088207244873047,
-            "mean_kld": 0.5014125614115543,
-            "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 0.47762323952768193,
-            "median_context_p99": 6.297145366668701,
-            "p90_context_kld": 0.8113779869139484,
-            "positions": 196512,
-            "top1_agreement": 0.787890815828041,
-            "worst_context_id": 196,
-            "worst_context_kld": 1.4113257386916236
-          },
-          "nvfp4": {
-            "contexts": 96,
-            "max_kld": 30.191476821899414,
-            "mean_kld": 0.7033966684846105,
-            "mean_ref_top1_prob": 0.6988670956880845,
-            "median_context_kld": 0.6896445865857285,
-            "median_context_p99": 7.739993095397949,
-            "p90_context_kld": 1.0432936284602716,
-            "positions": 196512,
-            "top1_agreement": 0.7418427373391956,
-            "worst_context_id": 194,
-            "worst_context_kld": 1.590631930451227
-          }
-        },
-        "key": "scientific_technical",
-        "label": "Scientific and technical exposition",
-        "relative_to_run": 1.1407447815298883
+        "relative_to_run": 1.1372154577295144
       },
       {
         "cells": {
@@ -1611,29 +1539,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 41.13975143432617,
-            "mean_kld": 1.736663274181469,
+            "max_kld": 39.378456115722656,
+            "mean_kld": 1.7222270669528827,
             "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 1.818968936428913,
-            "median_context_p99": 12.054352760314941,
-            "p90_context_kld": 2.379043205794732,
+            "median_context_kld": 1.7462169184915457,
+            "median_context_p99": 11.825867652893066,
+            "p90_context_kld": 2.449975450141369,
             "positions": 196512,
-            "top1_agreement": 0.5558998941540466,
-            "worst_context_id": 45,
-            "worst_context_kld": 2.891395312851532
+            "top1_agreement": 0.5567039163002768,
+            "worst_context_id": 114,
+            "worst_context_kld": 2.796517373867177
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 46.643943786621094,
-            "mean_kld": 2.023094138776235,
+            "max_kld": 38.341392517089844,
+            "mean_kld": 2.028686595381578,
             "mean_ref_top1_prob": 0.6868049425612657,
-            "median_context_kld": 2.0884355776743306,
-            "median_context_p99": 13.014652252197266,
-            "p90_context_kld": 2.7440561136428148,
+            "median_context_kld": 2.0537626696728117,
+            "median_context_p99": 13.110386848449707,
+            "p90_context_kld": 2.7314164504667997,
             "positions": 196512,
-            "top1_agreement": 0.5285020762090865,
+            "top1_agreement": 0.5299116593388699,
             "worst_context_id": 4,
-            "worst_context_kld": 3.3786182444198056
+            "worst_context_kld": 3.506572107144654
           },
           "mxfp8": {
             "contexts": 96,
@@ -1664,7 +1592,79 @@
         },
         "key": "encyclopedic_reference",
         "label": "Encyclopedic and factual reference",
-        "relative_to_run": 1.112321580169937
+        "relative_to_run": 1.1129420570128747
+      },
+      {
+        "cells": {
+          "all_quantized": {
+            "contexts": 96,
+            "max_kld": 40.053466796875,
+            "mean_kld": 1.6850800175124105,
+            "mean_ref_top1_prob": 0.6988670956880845,
+            "median_context_kld": 1.6989074666767194,
+            "median_context_p99": 12.346853256225586,
+            "p90_context_kld": 2.368493804166054,
+            "positions": 196512,
+            "top1_agreement": 0.5902896515225533,
+            "worst_context_id": 208,
+            "worst_context_kld": 2.906335476682348
+          },
+          "bxq": {
+            "contexts": 96,
+            "max_kld": 44.15528106689453,
+            "mean_kld": 1.787203256659065,
+            "mean_ref_top1_prob": 0.6988670956880845,
+            "median_context_kld": 1.8031902684070498,
+            "median_context_p99": 11.896342277526855,
+            "p90_context_kld": 2.5275857576081804,
+            "positions": 196512,
+            "top1_agreement": 0.5591617814688161,
+            "worst_context_id": 208,
+            "worst_context_kld": 3.163597909257638
+          },
+          "deployed": {
+            "contexts": 96,
+            "max_kld": 43.10159683227539,
+            "mean_kld": 2.025307995636494,
+            "mean_ref_top1_prob": 0.6988670956880845,
+            "median_context_kld": 2.0841908944410252,
+            "median_context_p99": 13.38927936553955,
+            "p90_context_kld": 2.762181235851715,
+            "positions": 196512,
+            "top1_agreement": 0.5387609916951637,
+            "worst_context_id": 208,
+            "worst_context_kld": 3.514232290307843
+          },
+          "mxfp8": {
+            "contexts": 96,
+            "max_kld": 31.088207244873047,
+            "mean_kld": 0.5014125614115543,
+            "mean_ref_top1_prob": 0.6988670956880845,
+            "median_context_kld": 0.47762323952768193,
+            "median_context_p99": 6.297145366668701,
+            "p90_context_kld": 0.8113779869139484,
+            "positions": 196512,
+            "top1_agreement": 0.787890815828041,
+            "worst_context_id": 196,
+            "worst_context_kld": 1.4113257386916236
+          },
+          "nvfp4": {
+            "contexts": 96,
+            "max_kld": 30.191476821899414,
+            "mean_kld": 0.7033966684846105,
+            "mean_ref_top1_prob": 0.6988670956880845,
+            "median_context_kld": 0.6896445865857285,
+            "median_context_p99": 7.739993095397949,
+            "p90_context_kld": 1.0432936284602716,
+            "positions": 196512,
+            "top1_agreement": 0.7418427373391956,
+            "worst_context_id": 194,
+            "worst_context_kld": 1.590631930451227
+          }
+        },
+        "key": "scientific_technical",
+        "label": "Scientific and technical exposition",
+        "relative_to_run": 1.111088549547169
       },
       {
         "cells": {
@@ -1683,29 +1683,29 @@
           },
           "bxq": {
             "contexts": 72,
-            "max_kld": 42.90846252441406,
-            "mean_kld": 1.6708091608251348,
+            "max_kld": 34.798072814941406,
+            "mean_kld": 1.665412877785023,
             "mean_ref_top1_prob": 0.6303062367294193,
-            "median_context_kld": 1.7225041677944846,
-            "median_context_p99": 9.77382755279541,
-            "p90_context_kld": 2.424303819353178,
+            "median_context_kld": 1.7605853694755507,
+            "median_context_p99": 10.072793960571289,
+            "p90_context_kld": 2.3378235146521353,
             "positions": 147384,
-            "top1_agreement": 0.5333279053357216,
-            "worst_context_id": 904,
-            "worst_context_kld": 3.1248215177600276
+            "top1_agreement": 0.5302135917060197,
+            "worst_context_id": 838,
+            "worst_context_kld": 2.4818499093282402
           },
           "deployed": {
             "contexts": 72,
-            "max_kld": 39.30211639404297,
-            "mean_kld": 1.8491939521132974,
+            "max_kld": 43.24997329711914,
+            "mean_kld": 1.9032763209563155,
             "mean_ref_top1_prob": 0.6303062367294193,
-            "median_context_kld": 1.8757585070465967,
-            "median_context_p99": 10.719711303710938,
-            "p90_context_kld": 2.6011211656892907,
+            "median_context_kld": 1.9972974845502194,
+            "median_context_p99": 11.016493797302246,
+            "p90_context_kld": 2.78199031600221,
             "positions": 147384,
-            "top1_agreement": 0.5131357542202681,
+            "top1_agreement": 0.5073549367638278,
             "worst_context_id": 925,
-            "worst_context_kld": 2.942326232320723
+            "worst_context_kld": 3.1738207166754964
           },
           "mxfp8": {
             "contexts": 72,
@@ -1736,7 +1736,7 @@
         },
         "key": "chinese",
         "label": "Chinese across several content types",
-        "relative_to_run": 1.01670915823006
+        "relative_to_run": 1.044141696667837
       },
       {
         "cells": {
@@ -1755,29 +1755,29 @@
           },
           "bxq": {
             "contexts": 72,
-            "max_kld": 40.05707931518555,
-            "mean_kld": 1.5756689115334925,
+            "max_kld": 43.00507736206055,
+            "mean_kld": 1.548540776399458,
             "mean_ref_top1_prob": 0.6885551524106842,
-            "median_context_kld": 1.5722352390911682,
-            "median_context_p99": 10.71493911743164,
-            "p90_context_kld": 2.3747487983014195,
+            "median_context_kld": 1.5455610482044142,
+            "median_context_p99": 10.553234100341797,
+            "p90_context_kld": 2.4009817014978534,
             "positions": 147384,
-            "top1_agreement": 0.5782445855723823,
+            "top1_agreement": 0.5815217391304348,
             "worst_context_id": 304,
-            "worst_context_kld": 3.25593660154531
+            "worst_context_kld": 3.1671384824877022
           },
           "deployed": {
             "contexts": 72,
-            "max_kld": 42.641807556152344,
-            "mean_kld": 1.8295626415753163,
+            "max_kld": 37.007591247558594,
+            "mean_kld": 1.8085538267964556,
             "mean_ref_top1_prob": 0.6885551524106842,
-            "median_context_kld": 1.8727219985929093,
-            "median_context_p99": 12.137351036071777,
-            "p90_context_kld": 2.6750341700629416,
+            "median_context_kld": 1.8304506825500275,
+            "median_context_p99": 11.705071449279785,
+            "p90_context_kld": 2.6872941921846776,
             "positions": 147384,
-            "top1_agreement": 0.5560576453346361,
-            "worst_context_id": 257,
-            "worst_context_kld": 3.4397168545750416
+            "top1_agreement": 0.5588734190956955,
+            "worst_context_id": 339,
+            "worst_context_kld": 4.061378358077808
           },
           "mxfp8": {
             "contexts": 72,
@@ -1808,7 +1808,7 @@
         },
         "key": "news_history_legal_essays",
         "label": "News, history, economics, legal analysis, and essays",
-        "relative_to_run": 1.005915626708278
+        "relative_to_run": 0.9921767220208605
       },
       {
         "cells": {
@@ -1827,29 +1827,29 @@
           },
           "bxq": {
             "contexts": 72,
-            "max_kld": 32.1933479309082,
-            "mean_kld": 1.4946107758018452,
+            "max_kld": 42.736114501953125,
+            "mean_kld": 1.5024764906905146,
             "mean_ref_top1_prob": 0.6303461857865943,
-            "median_context_kld": 1.500009053620101,
-            "median_context_p99": 9.486432075500488,
-            "p90_context_kld": 2.3340939446212148,
+            "median_context_kld": 1.524894262356297,
+            "median_context_p99": 9.397890090942383,
+            "p90_context_kld": 2.2545776405591753,
             "positions": 147384,
-            "top1_agreement": 0.5584595342778049,
-            "worst_context_id": 388,
-            "worst_context_kld": 2.7210277162281766
+            "top1_agreement": 0.560467893394127,
+            "worst_context_id": 439,
+            "worst_context_kld": 3.1497801563410865
           },
           "deployed": {
             "contexts": 72,
-            "max_kld": 38.19701385498047,
-            "mean_kld": 1.6581158258948987,
+            "max_kld": 38.11989212036133,
+            "mean_kld": 1.7055569548461247,
             "mean_ref_top1_prob": 0.6303461857865943,
-            "median_context_kld": 1.7261176263246027,
-            "median_context_p99": 10.789175033569336,
-            "p90_context_kld": 2.4446691744187006,
+            "median_context_kld": 1.7174336067742633,
+            "median_context_p99": 10.889728546142578,
+            "p90_context_kld": 2.5845404803262886,
             "positions": 147384,
-            "top1_agreement": 0.5436546707919448,
-            "worst_context_id": 388,
-            "worst_context_kld": 3.081658722460763
+            "top1_agreement": 0.5379891982847528,
+            "worst_context_id": 439,
+            "worst_context_kld": 3.1001899896399703
           },
           "mxfp8": {
             "contexts": 72,
@@ -1880,7 +1880,7 @@
         },
         "key": "literary_narrative",
         "label": "Literary, narrative, and creative writing",
-        "relative_to_run": 0.9116520977515372
+        "relative_to_run": 0.93567240499365
       },
       {
         "cells": {
@@ -1899,29 +1899,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 41.494056701660156,
-            "mean_kld": 1.3686865325159163,
+            "max_kld": 38.396728515625,
+            "mean_kld": 1.3358294237073802,
             "mean_ref_top1_prob": 0.7233971787516119,
-            "median_context_kld": 1.2778304976400205,
-            "median_context_p99": 10.17981243133545,
-            "p90_context_kld": 2.134711669512087,
+            "median_context_kld": 1.2826492898134843,
+            "median_context_p99": 9.845369338989258,
+            "p90_context_kld": 2.00187886403533,
             "positions": 196512,
-            "top1_agreement": 0.6207152743852793,
+            "top1_agreement": 0.6250712424686533,
             "worst_context_id": 708,
-            "worst_context_kld": 2.9879818448351
+            "worst_context_kld": 2.7549330833305397
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 34.1971549987793,
-            "mean_kld": 1.560781639985526,
+            "max_kld": 45.10734176635742,
+            "mean_kld": 1.546925506292042,
             "mean_ref_top1_prob": 0.7233971787516119,
-            "median_context_kld": 1.473631995802428,
-            "median_context_p99": 11.064117431640625,
-            "p90_context_kld": 2.313850639680717,
+            "median_context_kld": 1.5185987562235763,
+            "median_context_p99": 11.175138473510742,
+            "p90_context_kld": 2.279972149569506,
             "positions": 196512,
-            "top1_agreement": 0.6012406367041199,
+            "top1_agreement": 0.6033880882592412,
             "worst_context_id": 708,
-            "worst_context_kld": 3.254033478573557
+            "worst_context_kld": 3.2894278163579322
           },
           "mxfp8": {
             "contexts": 96,
@@ -1952,7 +1952,7 @@
         },
         "key": "worked_math_reasoning",
         "label": "Worked mathematics, science, and formal reasoning",
-        "relative_to_run": 0.8581365873261262
+        "relative_to_run": 0.8486468333441731
       },
       {
         "cells": {
@@ -1971,29 +1971,29 @@
           },
           "bxq": {
             "contexts": 36,
-            "max_kld": 28.927227020263672,
-            "mean_kld": 1.2778636800520669,
+            "max_kld": 33.88340759277344,
+            "mean_kld": 1.2661847644547204,
             "mean_ref_top1_prob": 0.8326448152069219,
-            "median_context_kld": 1.1591657174536962,
-            "median_context_p99": 10.449883460998535,
-            "p90_context_kld": 1.9698587425136318,
+            "median_context_kld": 1.0651777554342132,
+            "median_context_p99": 10.165213584899902,
+            "p90_context_kld": 2.288437004973564,
             "positions": 73692,
-            "top1_agreement": 0.6831949193942355,
+            "top1_agreement": 0.6826114096509798,
             "worst_context_id": 1016,
-            "worst_context_kld": 2.813886272014663
+            "worst_context_kld": 2.661403536306777
           },
           "deployed": {
             "contexts": 36,
-            "max_kld": 32.71179962158203,
-            "mean_kld": 1.421633946558572,
+            "max_kld": 31.666025161743164,
+            "mean_kld": 1.4835057973768624,
             "mean_ref_top1_prob": 0.8326448152069219,
-            "median_context_kld": 1.1716843978003344,
-            "median_context_p99": 11.04438591003418,
-            "p90_context_kld": 2.2876916350565337,
+            "median_context_kld": 1.3026068426019195,
+            "median_context_p99": 11.720060348510742,
+            "p90_context_kld": 2.4576422275537153,
             "positions": 73692,
-            "top1_agreement": 0.6721218042664061,
+            "top1_agreement": 0.6591353199804592,
             "worst_context_id": 1016,
-            "worst_context_kld": 3.1927021256291948
+            "worst_context_kld": 3.1780537988024644
           },
           "mxfp8": {
             "contexts": 36,
@@ -2024,7 +2024,7 @@
         },
         "key": "structured_data_tools",
         "label": "Structured data, tool calls, APIs, JSON, and tables",
-        "relative_to_run": 0.7816315056974008
+        "relative_to_run": 0.8138546375186066
       },
       {
         "cells": {
@@ -2043,29 +2043,29 @@
           },
           "bxq": {
             "contexts": 96,
-            "max_kld": 43.123695373535156,
-            "mean_kld": 1.1880508384670891,
+            "max_kld": 35.210819244384766,
+            "mean_kld": 1.1936500553981868,
             "mean_ref_top1_prob": 0.8424129918497996,
-            "median_context_kld": 1.147498215812061,
-            "median_context_p99": 10.951738357543945,
-            "p90_context_kld": 1.8728795462892707,
+            "median_context_kld": 1.1496776715632782,
+            "median_context_p99": 10.919404983520508,
+            "p90_context_kld": 1.8727022038840826,
             "positions": 196512,
-            "top1_agreement": 0.7191265673343104,
-            "worst_context_id": 631,
-            "worst_context_kld": 2.481150413090197
+            "top1_agreement": 0.716780654616512,
+            "worst_context_id": 635,
+            "worst_context_kld": 2.496798226059531
           },
           "deployed": {
             "contexts": 96,
-            "max_kld": 38.950897216796875,
-            "mean_kld": 1.3928449326560655,
+            "max_kld": 41.40315246582031,
+            "mean_kld": 1.4027538361857956,
             "mean_ref_top1_prob": 0.8424129918497996,
-            "median_context_kld": 1.4079201443575817,
-            "median_context_p99": 12.25846004486084,
-            "p90_context_kld": 2.1199874049731187,
+            "median_context_kld": 1.3496713696438576,
+            "median_context_p99": 12.287495613098145,
+            "p90_context_kld": 2.0891086081149934,
             "positions": 196512,
-            "top1_agreement": 0.6975044780980296,
-            "worst_context_id": 628,
-            "worst_context_kld": 2.8726303894076324
+            "top1_agreement": 0.6976876730174238,
+            "worst_context_id": 635,
+            "worst_context_kld": 2.825086072791175
           },
           "mxfp8": {
             "contexts": 96,
@@ -2096,7 +2096,7 @@
         },
         "key": "code_docs_issues",
         "label": "Source code, tests, technical documentation, and issue discussions",
-        "relative_to_run": 0.7658029583145585
+        "relative_to_run": 0.7695539288727226
       }
     ]
   },
@@ -2116,29 +2116,29 @@
     },
     "bxq": {
       "contexts": 768,
-      "max_kld": 43.123695373535156,
-      "mean_kld": 1.5886988350554396,
+      "max_kld": 44.15528106689453,
+      "mean_kld": 1.5851210894611856,
       "mean_ref_top1_prob": 0.7206552024685455,
-      "median_context_kld": 1.593714205632617,
-      "median_context_p99": 11.235530853271484,
-      "p90_context_kld": 2.4154826366798985,
+      "median_context_kld": 1.5678742616561478,
+      "median_context_p99": 11.206637382507324,
+      "p90_context_kld": 2.4125418709857684,
       "positions": 1572096,
-      "top1_agreement": 0.5981314118221788,
-      "worst_context_id": 969,
-      "worst_context_kld": 3.804737185271593
+      "top1_agreement": 0.5983623137518319,
+      "worst_context_id": 975,
+      "worst_context_kld": 4.436682889352712
     },
     "deployed": {
       "contexts": 768,
-      "max_kld": 46.643943786621094,
-      "mean_kld": 1.818803280313192,
+      "max_kld": 49.14295196533203,
+      "mean_kld": 1.8228142090582435,
       "mean_ref_top1_prob": 0.7206552024685455,
-      "median_context_kld": 1.820053068387497,
-      "median_context_p99": 12.465417861938477,
-      "p90_context_kld": 2.727569613311982,
+      "median_context_kld": 1.8277322289077729,
+      "median_context_p99": 12.440461158752441,
+      "p90_context_kld": 2.7397062899684954,
       "positions": 1572096,
-      "top1_agreement": 0.5775257999511481,
+      "top1_agreement": 0.577097709045758,
       "worst_context_id": 946,
-      "worst_context_kld": 4.479432370259336
+      "worst_context_kld": 4.5225937464682
     },
     "mxfp8": {
       "contexts": 768,

Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.md

@@ -1,6 +1,6 @@
 # Fidelity by domain - gemma-4-26B-A4B-it / Neural-ICE-Gemma-4-26B-A4B-it-NVFP4
 
-768 contexts, 1572096 scored positions, mean 1.81880328, reference top-1 72.1%, top-1 agreement 57.7526%.
+768 contexts, 1572096 scored positions, mean 1.82281421, reference top-1 72.1%, top-1 agreement 57.7098%.
 
 `x run` is the domain's mean divided by the run's mean, so 1.00 is a domain that degrades exactly as much as the model as a whole. The `deployed` cell is QxQ; `bxq` is the teacher-ID counterfactual on the same student weights. `Median ctx` and `p90 ctx` are the spread of per-context means within the domain.
 
@@ -8,16 +8,16 @@
 
 | Domain | Contexts | Ref top-1 | Mean KLD | x run | Median ctx | p90 ctx | Worst ctx |
 |---|---|---|---|---|---|---|---|
-| Other multilingual content | 36 | 70.7% | 2.36378738 | 1.30 | 2.33727762 | 4.12772208 | 4.479432 |
-| Natural dialogue, instruction following, and assistance | 96 | 77.4% | 2.07672787 | 1.14 | 2.08255142 | 3.06389986 | 4.157244 |
-| Scientific and technical exposition | 96 | 69.9% | 2.07479035 | 1.14 | 2.14664769 | 2.87363265 | 3.382483 |
-| Encyclopedic and factual reference | 96 | 68.7% | 2.02309414 | 1.11 | 2.08843558 | 2.74405611 | 3.378618 |
-| Chinese across several content types | 72 | 63.0% | 1.84919395 | 1.02 | 1.87575851 | 2.60112117 | 2.942326 |
-| News, history, economics, legal analysis, and essays | 72 | 68.9% | 1.82956264 | 1.01 | 1.87272200 | 2.67503417 | 3.439717 |
-| Literary, narrative, and creative writing | 72 | 63.0% | 1.65811583 | 0.91 | 1.72611763 | 2.44466917 | 3.081659 |
-| Worked mathematics, science, and formal reasoning | 96 | 72.3% | 1.56078164 | 0.86 | 1.47363200 | 2.31385064 | 3.254033 |
-| Structured data, tool calls, APIs, JSON, and tables | 36 | 83.3% | 1.42163395 | 0.78 | 1.17168440 | 2.28769164 | 3.192702 |
-| Source code, tests, technical documentation, and issue discussions | 96 | 84.2% | 1.39284493 | 0.77 | 1.40792014 | 2.11998740 | 2.872630 |
+| Other multilingual content | 36 | 70.7% | 2.36413931 | 1.30 | 2.31654604 | 3.54149690 | 4.522594 |
+| Natural dialogue, instruction following, and assistance | 96 | 77.4% | 2.07293250 | 1.14 | 2.10214494 | 3.19564611 | 3.949084 |
+| Encyclopedic and factual reference | 96 | 68.7% | 2.02868660 | 1.11 | 2.05376267 | 2.73141645 | 3.506572 |
+| Scientific and technical exposition | 96 | 69.9% | 2.02530800 | 1.11 | 2.08419089 | 2.76218124 | 3.514232 |
+| Chinese across several content types | 72 | 63.0% | 1.90327632 | 1.04 | 1.99729748 | 2.78199032 | 3.173821 |
+| News, history, economics, legal analysis, and essays | 72 | 68.9% | 1.80855383 | 0.99 | 1.83045068 | 2.68729419 | 4.061378 |
+| Literary, narrative, and creative writing | 72 | 63.0% | 1.70555695 | 0.94 | 1.71743361 | 2.58454048 | 3.100190 |
+| Worked mathematics, science, and formal reasoning | 96 | 72.3% | 1.54692551 | 0.85 | 1.51859876 | 2.27997215 | 3.289428 |
+| Structured data, tool calls, APIs, JSON, and tables | 36 | 83.3% | 1.48350580 | 0.81 | 1.30260684 | 2.45764223 | 3.178054 |
+| Source code, tests, technical documentation, and issue discussions | 96 | 84.2% | 1.40275384 | 0.77 | 1.34967137 | 2.08910861 | 2.825086 |
 
 ## The same domains across the ladder
 
@@ -25,44 +25,44 @@ Each column is a separate measurement of the same contexts, so a domain that mov
 
 | Domain | deployed | bxq | mxfp8 | nvfp4 | all_quantized |
 |---|---|---|---|---|---|
-| Other multilingual content | 2.36378738 | 2.01060924 | 0.68445047 | 0.89229785 | 1.79519304 |
-| Natural dialogue, instruction following, and assistance | 2.07672787 | 1.79807095 | 0.59552252 | 0.82821448 | 1.66159067 |
-| Scientific and technical exposition | 2.07479035 | 1.82912511 | 0.50141256 | 0.70339667 | 1.68508002 |
-| Encyclopedic and factual reference | 2.02309414 | 1.73666327 | 0.49743044 | 0.71954225 | 1.56808539 |
-| Chinese across several content types | 1.84919395 | 1.67080916 | 0.44706156 | 0.68167667 | 1.40434265 |
-| News, history, economics, legal analysis, and essays | 1.82956264 | 1.57566891 | 0.44719073 | 0.64140586 | 1.41631587 |
-| Literary, narrative, and creative writing | 1.65811583 | 1.49461078 | 0.39173107 | 0.62338708 | 1.35504621 |
-| Worked mathematics, science, and formal reasoning | 1.56078164 | 1.36868653 | 0.37927146 | 0.52795210 | 1.23575505 |
-| Structured data, tool calls, APIs, JSON, and tables | 1.42163395 | 1.27786368 | 0.47775227 | 0.61546437 | 1.16853193 |
-| Source code, tests, technical documentation, and issue discussions | 1.39284493 | 1.18805084 | 0.36354693 | 0.52781126 | 1.09767677 |
+| Other multilingual content | 2.36413931 | 2.12630776 | 0.68445047 | 0.89229785 | 1.79519304 |
+| Natural dialogue, instruction following, and assistance | 2.07293250 | 1.83255161 | 0.59552252 | 0.82821448 | 1.66159067 |
+| Encyclopedic and factual reference | 2.02868660 | 1.72222707 | 0.49743044 | 0.71954225 | 1.56808539 |
+| Scientific and technical exposition | 2.02530800 | 1.78720326 | 0.50141256 | 0.70339667 | 1.68508002 |
+| Chinese across several content types | 1.90327632 | 1.66541288 | 0.44706156 | 0.68167667 | 1.40434265 |
+| News, history, economics, legal analysis, and essays | 1.80855383 | 1.54854078 | 0.44719073 | 0.64140586 | 1.41631587 |
+| Literary, narrative, and creative writing | 1.70555695 | 1.50247649 | 0.39173107 | 0.62338708 | 1.35504621 |
+| Worked mathematics, science, and formal reasoning | 1.54692551 | 1.33582942 | 0.37927146 | 0.52795210 | 1.23575505 |
+| Structured data, tool calls, APIs, JSON, and tables | 1.48350580 | 1.26618476 | 0.47775227 | 0.61546437 | 1.16853193 |
+| Source code, tests, technical documentation, and issue discussions | 1.40275384 | 1.19365006 | 0.36354693 | 0.52781126 | 1.09767677 |
 
 ## Ranked by source dataset
 
 | Source | Contexts | Ref top-1 | Mean KLD | x run | Median ctx | p90 ctx | Worst ctx |
 |---|---|---|---|---|---|---|---|
-| wikipedia_ru | 6 | 67.4% | 3.42883253 | 1.89 | 3.58539667 | 4.24686247 | 4.246862 |
-| wikipedia_cs | 6 | 70.9% | 2.94733312 | 1.62 | 2.91655752 | 4.47943237 | 4.479432 |
-| wikipedia_fr | 3 | 69.8% | 2.45964151 | 1.35 | 2.38770855 | 3.08783978 | 3.087840 |
-| wikipedia_de | 7 | 77.8% | 2.14317143 | 1.18 | 2.13962176 | 2.53813189 | 2.538132 |
-| wildchat | 96 | 77.4% | 2.07672787 | 1.14 | 2.08255142 | 3.06389986 | 4.157244 |
-| scientific_papers | 96 | 69.9% | 2.07479035 | 1.14 | 2.14664769 | 2.87363265 | 3.382483 |
-| open_news | 23 | 66.7% | 2.06663665 | 1.14 | 1.99121440 | 3.14055687 | 3.383914 |
-| wikipedia_en | 96 | 68.7% | 2.02309414 | 1.11 | 2.08843558 | 2.74405611 | 3.378618 |
-| wikipedia_zh | 39 | 61.8% | 1.98919047 | 1.09 | 2.09188728 | 2.72111365 | 2.939617 |
-| wikipedia_ja | 7 | 69.6% | 1.89628395 | 1.04 | 1.82755000 | 2.69385492 | 2.693855 |
-| public_domain_review | 26 | 64.5% | 1.79542688 | 0.99 | 1.81613992 | 2.67503417 | 3.235695 |
-| wikisource_zh | 33 | 64.5% | 1.68374352 | 0.93 | 1.68374639 | 2.24469261 | 2.942326 |
-| public_domain_books | 72 | 63.0% | 1.65811583 | 0.91 | 1.72611763 | 2.44466917 | 3.081659 |
-| regulations | 23 | 75.9% | 1.63107689 | 0.90 | 1.70736205 | 2.39118263 | 3.439717 |
-| wikipedia_es | 7 | 67.8% | 1.59774849 | 0.88 | 1.31275107 | 2.78642330 | 2.786423 |
-| libretexts | 96 | 72.3% | 1.56078164 | 0.86 | 1.47363200 | 2.31385064 | 3.254033 |
-| stackv2 | 44 | 79.9% | 1.44362618 | 0.79 | 1.54038322 | 2.18371791 | 2.872630 |
-| starcoder_structured | 36 | 83.3% | 1.42163395 | 0.78 | 1.17168440 | 2.28769164 | 3.192702 |
-| github_code | 52 | 87.9% | 1.34987618 | 0.74 | 1.26600821 | 2.07561569 | 2.544324 |
+| wikipedia_cs | 6 | 70.9% | 3.20466405 | 1.76 | 3.43752576 | 4.52259375 | 4.522594 |
+| wikipedia_ru | 6 | 67.4% | 3.01358420 | 1.65 | 3.34436380 | 4.17155865 | 4.171559 |
+| wikipedia_fr | 3 | 69.8% | 2.27103206 | 1.25 | 2.13833981 | 2.85326248 | 2.853262 |
+| wikipedia_de | 7 | 77.8% | 2.19188762 | 1.20 | 2.07089614 | 2.87901628 | 2.879016 |
+| wildchat | 96 | 77.4% | 2.07293250 | 1.14 | 2.10214494 | 3.19564611 | 3.949084 |
+| wikipedia_zh | 39 | 61.8% | 2.06742544 | 1.13 | 2.10793945 | 2.86435606 | 3.051148 |
+| wikipedia_en | 96 | 68.7% | 2.02868660 | 1.11 | 2.05376267 | 2.73141645 | 3.506572 |
+| scientific_papers | 96 | 69.9% | 2.02530800 | 1.11 | 2.08419089 | 2.76218124 | 3.514232 |
+| open_news | 23 | 66.7% | 1.94607399 | 1.07 | 2.04477242 | 2.77932351 | 3.570138 |
+| wikipedia_ja | 7 | 69.6% | 1.93743264 | 1.06 | 1.93393023 | 2.47459009 | 2.474590 |
+| public_domain_review | 26 | 64.5% | 1.83289703 | 1.01 | 1.81165462 | 2.65638501 | 4.061378 |
+| wikipedia_es | 7 | 67.8% | 1.72588397 | 0.95 | 2.24875411 | 2.65378898 | 2.653789 |
+| wikisource_zh | 33 | 64.5% | 1.70928190 | 0.94 | 1.66040020 | 2.32818727 | 3.173821 |
+| public_domain_books | 72 | 63.0% | 1.70555695 | 0.94 | 1.71743361 | 2.58454048 | 3.100190 |
+| regulations | 23 | 75.9% | 1.64351526 | 0.90 | 1.86775326 | 2.47881582 | 3.291825 |
+| libretexts | 96 | 72.3% | 1.54692551 | 0.85 | 1.51859876 | 2.27997215 | 3.289428 |
+| starcoder_structured | 36 | 83.3% | 1.48350580 | 0.81 | 1.30260684 | 2.45764223 | 3.178054 |
+| github_code | 52 | 87.9% | 1.41589598 | 0.78 | 1.38865468 | 2.12410996 | 2.825086 |
+| stackv2 | 44 | 79.9% | 1.38722221 | 0.76 | 1.31796733 | 2.01118918 | 2.377006 |
 
 ## Reading
 
-- Weakest domain: **Other multilingual content** at 2.36378738, 1.30x the run mean of 1.81880328 over 36 context(s).
-- Strongest domain: **Source code, tests, technical documentation, and issue discussions** at 1.39284493, 0.77x the run mean. The spread across domains is 1.7x.
+- Weakest domain: **Other multilingual content** at 2.36413931, 1.30x the run mean of 1.82281421 over 36 context(s).
+- Strongest domain: **Source code, tests, technical documentation, and issue discussions** at 1.40275384, 0.77x the run mean. The spread across domains is 1.7x.
 - Every cell is weakest on the same domain (Other multilingual content), so the ranking is a property of the model and the text rather than of one format.
 

QXQ.md

@@ -1,6 +1,6 @@
 # QxQ and BxQ — Routed-Model Distribution Fidelity
 
-**Applies to:** laws version 12, Law 14.
+**Applies to:** laws version 14, Laws 14 and 17.
 **Read first:** [`LAWS.md`](LAWS.md) for the laws, [`README.md`](README.md) for the
 harness.
 
@@ -106,8 +106,9 @@ completion order. No allowlist entry overrides this.
 ## 5. Certification fails closed
 
 A kernel's `_supports_batch_invariance()` is a claim about itself. It is not
-evidence, and at least one kernel's claim is false in a way that silently
-destroys the measurement (§6).
+evidence. `CutlassExpertsFp4` declared True and still produced NaN on W4A4
+checkpoints; the NaN was uninitialized per-expert activation scales, not the
+kernel, but the lesson stands: a self-declaration is not a probe.
 
 `inspect_model_moe_backends` in `vllm/v1/sample/kld.py` therefore grants
 `certified_for_exact_repeat` only when all three of these hold:
@@ -119,7 +120,11 @@ destroys the measurement (§6).
 3. The run is not expert-parallel.
 
 The allowlist currently holds `MarlinExperts`, `BatchedMarlinExperts`,
-`TritonExperts`, and `Nvfp4QuantizationEmulationTritonExperts`.
+`TritonExperts`, `Nvfp4QuantizationEmulationTritonExperts`, and
+`CutlassExpertsFp4`. CUTLASS earned its place after the SM120 bitwise
+permutation test (24/24, `atol=0`) and after the W4A4 NaN was shown to be a
+loader gap, not a kernel defect. The run's own zero-tolerance exact-repeat
+control remains the binding gate.
 
 The default is refusal. An unprobed backend — including one that arrives with a
 future vLLM bump — is uncertified until somebody probes it, and an uncertified
@@ -130,13 +135,31 @@ producing garbage that the harness dutifully writes down as a fidelity result.
 DeepGEMM, FlashInfer MoE, AITER, XPU, CPU, and every expert-parallel path remain
 uncertified.
 
-## 6. Case study: a kernel that declares batch invariance and returns NaN
+## 6. Case study: uninitialized scales, not a broken kernel
 
 `CutlassExpertsFp4` self-declares batch invariance. Under
-`VLLM_BATCH_INVARIANT=1`, on W4A4 NVFP4 checkpoints, it takes finite inputs to
-NaN. The scorer then refuses a non-finite KLD and the engine dies, which is how
-this surfaced: two NVFP4 candidates failing with `KLD is not finite`, no report
-produced, and nothing to indicate the kernel rather than the checkpoint.
+`VLLM_BATCH_INVARIANT=1`, on two W4A4 NVFP4 checkpoints, it took finite inputs
+to NaN. The scorer refused a non-finite KLD and the engine died. The first
+reading was that the kernel's claim was false. That was wrong.
+
+The kernel is bitwise batch-invariant on this hardware. The CUDA grouped GEMM
+asserts at compile time that it uses `PersistentTileSchedulerSm100Group` "for
+batch invariance", and `tests/v1/determinism/test_cutlass_batch_invariance.py`
+passed 24/24 on SM120 at `atol=0, rtol=0` across both activations, both expert
+counts (40, 64), both top-k values, and all three shape cases.
+
+The NaN was uninitialized memory. NVFP4 allocated per-expert activation scales
+with `torch.empty`. Checkpoints that omitted `input_scale` keys for some experts
+in some layers — `bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4` (14 layers short)
+and `Neural-ICE/Gemma-4-26B-A4B-it-NVFP4` (16 layers short) — left those slots
+holding whatever was on the GPU. CUTLASS consumes one scale per expert
+(`a1_gscale` / `a2_gscale` of length `e`); a NaN slot damages that expert. The
+unsloth export, with every slot written, never went NaN.
+
+Loading does not catch this. Strict all-parameters-loaded tracking is off by
+default for quantized models, and any module with `process_weights_after_loading`
+has every parameter force-marked as loaded, so a partially filled tensor is
+invisible to it by construction.
 
 Localizing it took two purpose-built tools.
 
@@ -159,21 +182,21 @@ scoring harness does and what a generation-only probe does not.
 With both, the failure is exactly reproducible: `moe.experts` at layer 0, row 814
 of `context-0002`, and layer 14, row 170 of `context-0004`.
 
-The `--moe-backend` override then isolated the kernel:
+The `--moe-backend` override then isolated the failure to the native path on
+the two incomplete checkpoints, not to batch invariance itself:
 
 | `VLLM_BATCH_INVARIANT` | Backend | Result |
 | --- | --- | --- |
 | 0 | `FLASHINFER_CUTLASS` (auto) | pass |
-| 1 | `VLLM_CUTLASS` (auto) | **NaN** |
-| 1 | `marlin` (forced) | pass |
-| 1 | `emulation` (forced) | pass |
+| 1 | `VLLM_CUTLASS` (auto) | **NaN** on incomplete exports |
+| 1 | `marlin` (forced) | pass (drops activation scales) |
+| 1 | `emulation` (forced) | pass (collapses per-expert scales) |
 
-The bug is specific to the CUTLASS FP4 experts under batch invariance, not to
-NVFP4, not to the checkpoints, and not to batch invariance itself. It is why §5
-fails closed: this kernel's self-declaration was the only thing standing between
-a broken forward pass and a published fidelity number.
+The fail-closed allowlist was the right reaction to a NaN that looked like a
+kernel defect. Once the scales were the cause, keeping CUTLASS off the list
+was the thing standing between native BxQ and a published number.
 
-## 7. W4A4 NVFP4 scores on faithful emulation
+## 7. W4A4 NVFP4 scores on native CUTLASS
 
 A W4A16 export and a W4A4 export of the same model differ only in whether
 activations are also quantized, and that difference decides which kernels can
@@ -182,27 +205,108 @@ so `_quantizes_activations_to_fp4` in `examples/offline_inference/score_mode_kld
 reads the checkpoint: any entry in `quantization_config.config_groups` whose
 `input_activations` declares `num_bits: 4` and `type: "float"` makes it W4A4.
 
-For those checkpoints the scorer pins `moe_backend="emulation"`, which
-quantize-dequantizes both activation stages in the checkpoint's own scheme and is
-batch invariant by construction.
-
-Two details matter and both were bugs first:
+For those checkpoints the scorer pins `moe_backend="cutlass"`. vLLM CUTLASS
+is the only native W4A4 MoE path that keeps a per-expert activation-scale
+vector. FlashInfer collapses every expert to one scalar via
+`amax_for_moe_activation_quant(...).repeat(num_experts)` — the same defect as
+emulation. Marlin drops activation scales and would score W4A4 as W4A16.
+
+Unwritten slots are now a NaN sentinel, filled from the maximum of the present
+per-expert scales before CUTLASS fuses them into the weight alphas. Too large
+wastes quantization range; too small overflows e4m3. A layer with no finite
+positive slot is refused rather than invented. The fill is recorded on the
+layer at fill time as `uncalibrated_experts_filled_from_layer_max` and Law 17
+discloses it. A complete export such as unsloth records an empty substitution
+list: native CUTLASS used the checkpoint's own scales.
+
+Both halves of the model are walked, because a dense NVFP4 projection can omit a
+shard's scale exactly as an expert can. `inspect_model_moe_backends` covers the
+routed experts and `inspect_model_nvfp4_dense_scales` covers everything else,
+reported separately so the denominators stay meaningful — a fill on 3 of 200
+dense layers is a different claim than 3 of 30 experts. The fill records a scan
+marker on every layer it visits, filled or not, because counting NVFP4 layers
+after load is guesswork: the dense paths delete or overwrite the very parameter
+they were named for. Without the dense walk a dense W4A4 candidate reported no
+substitution, and Law 17 read that silence as a clean bill of health rather than
+as an absence of evidence, which is why it now returns `NOT_APPLICABLE` when
+nothing was inspected at all.
+
+| Path | Per-expert activation scales | Exact repeat |
+| --- | --- | --- |
+| vLLM CUTLASS FP4 | honoured (per-expert vector) | certified on SM120 |
+| FlashInfer FP4 | collapsed to one scalar for the layer | uncertified |
+| Marlin | dropped entirely (scores W4A16) | certified |
+| Emulation | collapsed to a layer maximum | certified |
 
 **The pin belongs on the student only.** It is applied to `student_kwargs`, not
 the shared `llm_kwargs`. Applied to the latter it propagates to the unquantized
 BF16 teacher, which has no such scheme, and reference engine initialization fails.
 
-**Marlin is refused here, not merely not preferred.** Marlin is batch invariant
-and would run, but its MoE path drops activation scales — it would score a W4A4
-checkpoint as though it were W4A16 and report a fidelity the checkpoint never
-delivers. A wrong number that passes every law is worse than a refusal.
-
-The honest consequence, stated in the published card: an NVFP4 result measures
-the quantization scheme rather than a native FP4 kernel's own rounding. Read the
-family's leaderboard with that in mind. Two NVFP4 exports in the
-gemma-4-26B-A4B-it family carry byte-identical QDQ diagnostics (0.66652247 and
-1.43642041) and yet score QxQ 1.16299506 and 1.75580190 — a 0.6 nat spread that
-lives entirely in the activation scheme and kernel path, not in the weights.
+A weight-only W4A16 NVFP4 result still measures the quantization scheme rather than
+a native FP4 kernel's own rounding, because dense Marlin is not batch invariant
+and those linear layers use deterministic emulation. That limit is stated on the
+published card.
+
+### History: the emulation collapse we published through
+
+Before the loader gap was understood, W4A4 scoring was pinned to emulation.
+The emulation branch of `convert_to_nvfp4_moe_kernel_format` collapses the
+per-expert activation scales to one scalar per layer —
+`a13_scale = 1.0 / a13_scale.max()` — and vLLM's own comment says taking the
+largest global scale "likely results in overflowing the FP8 range for other
+experts." Since `a2_scale` holds `1 / w2_input_global_scale`, `.max()` applies
+the *smallest* per-expert scale. Measured on
+`unsloth/gemma-4-26B-A4B-it-NVFP4`:
+
+| Parameter | Slots | Distinct values | Worst spread | Collapsed |
+| --- | --- | --- | --- | --- |
+| `w13_input_global_scale` | 256 | 1 | 1.0x | no |
+| `w2_input_global_scale` | 128 | 86–99 | 150.5x | yes, all 30 layers |
+
+Four NVFP4 exports carrying byte-identical QDQ diagnostics (0.66652247 and
+1.43642041) scored QxQ between 1.16299506 and 1.82545539 on that path, and that
+0.6 nat spread was written up as living "entirely in the activation scheme and
+kernel path." That reading was unsupported. The spread was substantially an
+artifact of the measurement. It stays here because it is the exact shape of
+mistake this document exists to prevent: a real, reproducible, bitwise-exact
+number that is nonetheless measuring the harness rather than the checkpoint.
+
+Law 17 still exists for the remaining real substitutions — an uncalibrated
+expert filled from the layer maximum is one — so a substituted result ranks
+only against candidates measured the same way, and is never withdrawn for
+having a disclosed fill.
+
+### How far to discount a filled result
+
+The disclosure states a spread. It does not say what that spread costs, and the
+first native-CUTLASS campaign answered the question by accident.
+
+The two filled candidates landed on top of each other:
+
+| Candidate | Layers filled | Worst spread | QxQ | BxQ |
+| --- | --- | --- | --- | --- |
+| `Neural-ICE/Gemma-4-26B-A4B-it-NVFP4` | 15 of 30 | 230.6x | 1.82281421 | 1.58512109 |
+| `bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4` | 12 of 30 | 240.3x | 1.82307292 | 1.59361302 |
+
+0.00026 nats apart on QxQ, from different publishers, with different fill
+footprints, and with one quantizing its LM head while the other does not. They
+are not the same checkpoint: distinct `student_weights_sha256`, and distinct
+per-position KLD digests in `kld_evidence`, so the two runs did not produce
+bitwise-identical output either.
+
+Two independent quantizations do not agree to four decimal places on their own.
+The reading that fits is that once a tensor's per-expert scales are replaced by
+one layer maximum on half the layers, the fill sets the number and the
+checkpoint's own choices stop being visible in it. On the same suite the clean
+NVFP4 exports separate normally — 1.13846019 for unsloth against 1.77968754 for
+RedHatAI — so the collapse is not the suite failing to discriminate.
+
+This is n=2 and therefore a strong hint rather than a proof; a fill-direction
+sweep on one checkpoint would settle it, and has not been run. Treat a filled
+QxQ as an upper bound on that family of exports rather than a measurement of the
+particular one, which is what the separate comparability group already enforces.
+It is also the argument against ever promoting the fill out of the harness: see
+§13.
 
 ## 8. Checkpoint defects the pipeline had to fix, not tolerate
 
@@ -232,23 +336,52 @@ module probe is what separated them, naming `layers.0.router.proj` rather than
 `moe.experts`. The fix dequantizes those tensors into BF16 at load time in
 `Gemma4Model.load_weights`, with tests in `tests/kernels/moe/test_gemma4router.py`
 covering the transposed packed values and the refusal of an indivisible group
-count. That checkpoint now scores 1.17816288 and passes all sixteen laws.
+count. That checkpoint now scores 1.17816288 and passes all seventeen laws.
 
 ## 9. The runtime the numbers are bound to
 
 Scoring pins `VLLM_BATCH_INVARIANT=1`, disables DeepGEMM
 (`VLLM_MOE_USE_DEEP_GEMM=0`) and FlashInfer autotune, sets `NCCL_DETERMINISTIC=1`
 and `CUBLAS_WORKSPACE_CONFIG=:4096:8`, enforces eager execution, disables prefix
-caching, and holds `max_num_seqs=1`.
+caching, holds `max_num_seqs=1`, and pins `kv_cache_dtype=bfloat16`.
 
 None of that is optional and none of it is a performance setting. Each one closes
 a path by which two runs of the same tokens could diverge.
 
+The KV cache pin closes the widest such path found so far, and it was open for
+the whole campaign. Left at `auto`, vLLM resolves the KV cache dtype from a
+scheme declared in the candidate's own config, so
+`unsloth/gemma-4-26B-A4B-it-NVFP4` — which declares an 8-bit float KV cache —
+had its attention read and written through a quantized cache while every
+candidate it was ranked against used an unquantized one. That is a difference in
+how the measurement was taken, not a property of the checkpoint being measured.
+The cache is never quantized: scoring holds 4096 tokens, so there is no memory
+pressure that quantizing it could relieve, and nothing to weigh against the loss
+of comparability.
+
+The rescore that followed measured what it had cost, and doubled as the cleanest
+determinism evidence in this document. unsloth is the only one of the nine
+candidates whose inspection reported a declared KV cache scheme, and it is the
+only one whose number moved: 1.16570700 to 1.13846019, so its declared FP8 cache
+had been costing it 0.02724681 nats, or 2.3% of its KLD, and enough to place it
+above `gemma-4-26B-A4B-it-W4A16` at 1.15926068 when it belongs below. Five of the
+remaining eight are quoted at their pre-pin values elsewhere in this document —
+0.69415039, 1.17816288 with its BxQ and delta, 1.77968754, 1.82281421, and
+1.82307292 — and every one came back identical to eight decimal places on a
+different `vllm_commit`. So the defect was worth a rescore and was not worth a
+panic, the blast radius was exactly the set the mechanism predicted, and the
+harness reproduced everything outside it. `assert_unquantized_kv_cache` reads the dtype the engine actually
+resolved and refuses both a quantized value and `auto`, because the failure being
+prevented is precisely a value nobody checked.
+
 Because that runtime *is* part of the result, its identity is bound into the
 comparability key: `vllm_commit`, `vllm_dirty_digest`,
-`compiled_extensions_sha256`, `torch`, `driver`, and `gpu_names`, alongside the
-suite and geometry. Two candidates are ranked against each other only when all of
-it matches. A consequence worth internalizing before committing to this
+`compiled_extensions_sha256`, `torch`, `driver`, `gpu_names`, and
+`kv_cache_dtype`, alongside the suite and geometry. Two candidates are ranked
+against each other only when all of it matches. `kv_cache_dtype` is in that list
+because of the unsloth case above: the key's one job is to bound a ranking to
+runs that ran alike, and it had nothing to say about a candidate whose attention
+ran at a different precision than its neighbours'. A consequence worth internalizing before committing to this
 repository: **any commit changes `vllm_commit`, and a dirty tree changes
 `vllm_dirty_digest`, so the next scoring run treats every prior report as stale
 and rescores the family.** That is correct behaviour, not a bug, and it is why
@@ -338,13 +471,32 @@ suite, geometry, runtime, or laws version, and not against any published
 elsewhere. The comparability key is printed with every leaderboard group for
 exactly this reason.
 
-**An NVFP4 result measures the scheme, not a native FP4 kernel.** See §7. The
-faithful-emulation path is a deliberate substitution, disclosed rather than hidden.
+**A W4A4 NVFP4 MoE result is a native CUTLASS number.** See §7. FlashInfer and
+emulation collapse per-expert activation scales; those paths are not how a
+W4A4 candidate is scored. A W4A16 dense NVFP4 result still measures the scheme
+rather than a native FP4 kernel, because dense Marlin is not batch invariant.
 
 **The QDQ ladder is diagnostic, never a candidate.** Those cells round weights on
 synthetic BF16 checkpoints and route naturally. They are not QxQ or BxQ, they are
 not rankable against deployed candidates, and the published tables separate them.
 
+**A scheme label names the narrowest group, not the whole model.** A checkpoint
+may quantize attention at one width and its experts at another, and may declare a
+KV cache scheme. `unsloth/gemma-4-26B-A4B-it-NVFP4` is `format:
+"mixed-precision"`: FP8 W8A8 attention, NVFP4 W4A4 experts and dense MLP, and a
+declared FP8 KV cache. Its QxQ of 1.13846019 against 1.77968754 for a complete
+all-`Linear` NVFP4 export is therefore mostly the 8-bit attention, not a better
+NVFP4 export, and it lands between the all-FP8 candidate at 0.69415039 and the
+all-NVFP4 ones exactly where a hybrid should. The declared KV cache is a separate
+matter and is no longer in that number: §9 pins an unquantized cache, which is
+worth 0.02724681 of the 0.64 separating it from the complete export, so the
+attention width still carries the result. This is not a comparability failure — the key deliberately excludes
+the candidate's scheme, because ranking schemes against one reference is the
+point — but it was a labelling one until `scheme_mix` and `kv_cache_scheme` were
+added to the inspection. Component coverage cannot substitute: it counts how many
+weights are quantized, not at what width, so a hybrid and a uniform export both
+read `all`.
+
 **A flip rate is not an error rate.** Two experts disagreeing on a token is not
 by itself a wrong answer; the delta is what quantifies the cost. AutoRound's
 93.07% flip rate with a +0.275 delta is the point — high disagreement, bounded
@@ -354,3 +506,47 @@ consequence.
 says two runs agree bit for bit. It says nothing about whether the kernel computes
 the right thing, which is what the zero baseline (Law 1) and the reference binding
 (Laws 12 and 16) are for.
+
+## 13. The fill is a harness policy, not a vLLM fix
+
+Half of the loader change here — allocating consumed NVFP4 scales as a NaN
+sentinel instead of `torch.empty`, so an unwritten slot is detectable rather
+than arbitrary — is not ours to contribute. Three open upstream PRs already do
+it on the same files by the same mechanism: #54444 on the ModelOpt linear
+methods and fused experts, #45320 on the ModelOpt per-expert scales, #52501 on
+the linear per-block `weight_scale`. A fourth, #55073, is actively reworking the
+same compressed-tensors and ModelOpt scale code.
+
+Where we differ is the policy after detection, and the difference is deliberate
+on both sides. All three upstream PRs **reject**: they raise at load time naming
+the parameter and the affected experts. #45320 states the position outright —
+"this remains fail-fast only, it does not guess missing calibration statistics
+or add an imputation policy." Filling from the layer maximum is exactly the
+imputation policy they declined.
+
+They are right for a serving engine, and the evidence for that is in §7's
+convergence table rather than in any argument from principle. A user served a
+filled checkpoint gets a model whose numerics are substantially set by an
+invented scale, with nothing on the surface to say so. Refusing to load is the
+better failure.
+
+The harness can do what the engine should not, because it discloses. Law 17
+records the fill on the report, the substituted parameters enter the
+comparability key, and a filled candidate ranks only against others measured the
+same way. That is the whole justification, and it does not transfer to a serving
+path that has no comparability key to put anything in.
+
+Two consequences follow.
+
+**Those two candidates will stop loading on stock vLLM.** When any of the
+rejecting PRs lands, `Neural-ICE/Gemma-4-26B-A4B-it-NVFP4` and
+`bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4` will refuse at load. Their
+published numbers stay valid for what they are and become unreproducible without
+this fill, so the fill has to be maintained as a standing, disclosed divergence
+rather than treated as a fix awaiting merge.
+
+**The contribution worth making is evidence, not code.** #45320, #54444, and
+#55073 all report that no end-to-end evaluation was run: no Blackwell hardware,
+or no affected checkpoint, or both. This harness has SM120, two affected
+checkpoints, and measured numbers for what the missing scales cost. That is the
+gap in those PRs, and it is not a competing patch.

README.md

@@ -11,7 +11,7 @@ tags:
 
 Mean KL divergence against on-disk size
 
-Scored under the distribution-fidelity laws, version 12. Read [`LAWS.md`](LAWS.md) first: these numbers are comparable only within this artifact's token suite, geometry, and runtime identity, and not against any number produced elsewhere.
+Scored under the distribution-fidelity laws, version 13. Read [`LAWS.md`](LAWS.md) first: these numbers are comparable only within this artifact's token suite, geometry, and runtime identity, and not against any number produced elsewhere.
 
 Each candidate directory holds its one-pager (`report.md`), its raw report, its compliance receipt, and its Law 14 attribution where one was produced. `reference/` carries the reusable teacher tensors and head, so a third party can score a new candidate without loading the reference checkpoint. `checksums.txt` covers every file here.
 
@@ -21,13 +21,13 @@ Each candidate directory holds its one-pager (`report.md`), its raw report, its
 |---|---|---|---|---|---|---|---|
 | RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic | fp8_per_channel | 26.68 GiB | 0.69415039 | 0.49783177 | +0.19631862 | 50.6828% | certified |
 | cyankiwi/gemma-4-26B-A4B-it-AWQ-4bit | int4_g32_sym | 16.01 GiB | 1.02626820 | 0.84088786 | +0.18538034 | 62.7369% | certified |
+| unsloth/gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.75 GiB | 1.13846019 | 0.95586495 | +0.18259525 | 65.3084% | certified |
 | dhruvil237/gemma-4-26B-A4B-it-W4A16 | int4_g64_sym | 14.58 GiB | 1.15926068 | 1.01252351 | +0.14673717 | 66.7300% | certified |
-| unsloth/gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.75 GiB | 1.16299506 | 1.00117013 | +0.16182492 | 66.4622% | certified |
 | Intel/gemma-4-26B-A4B-it-int4-AutoRound | int4_g128_sym | 14.31 GiB | 1.17816288 | 0.90308200 | +0.27508088 | 93.0727% | certified |
 | lcu0312/gemma-4-26B-A4B-it-AWQ-4bit | int4_g32_sym | 16.01 GiB | 1.23587221 | 1.08009636 | +0.15577585 | 66.1806% | certified |
-| RedHatAI/gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.30 GiB | 1.75580190 | 1.53130858 | +0.22449332 | 74.3937% | certified |
-| Neural-ICE/Gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.29 GiB | 1.81880328 | 1.58869884 | +0.23010445 | 74.4082% | certified |
-| bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.29 GiB | 1.82545539 | 1.58674213 | +0.23871325 | 74.4254% | certified |
+| RedHatAI/gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.30 GiB | 1.77968754 | 1.54324011 | +0.23644744 | 74.7339% | certified |
+| Neural-ICE/Gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.29 GiB | 1.82281421 | 1.58512109 | +0.23769312 | 74.4778% | certified |
+| bg-digitalservices/Gemma-4-26B-A4B-it-NVFP4 | nvfp4 | 15.29 GiB | 1.82307292 | 1.59361302 | +0.22945990 | 74.5241% | certified |
 
 ## QxQ fidelity against size
 

baselines/self-kld.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "305f8014578ad62130ab0ab7f2e584eb96a4f5dc330ae7fc8bc0097a35737ca2",
+  "capture_manifest_sha256": "66daab5f0c830939040abbe57e06064e0be33e863dea4e88b7b92d0c0159ff60",
   "confidence_buckets": [
     {
       "frac": 0.06888128969223253,
@@ -65,6 +65,7 @@
     }
   ],
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "max_kld": 0.0,
   "mean_kld": 0.0,
   "mean_kld_reverse": 0.0,
@@ -72,6 +73,11 @@
   "model_runner_v2": false,
   "num_positions": 2047,
   "num_rows": 1,
+  "nvfp4_dense_scale_inspection": {
+    "layers_filled": 0,
+    "layers_scanned": 0,
+    "workers": 1
+  },
   "p90_kld": 0.0,
   "p99_kld": 0.0,
   "per_context": [
@@ -87,6 +93,7 @@
       "top1_agreement": 1.0
     }
   ],
+  "quantization_substitutions": [],
   "reference_weights_sha256": "201ebceb92c5b166cc028c3008607ba28c478c7699b1d382e7be7ede81f2ab75",
   "runtime_binding": {
     "compiled_extensions": {
@@ -118,7 +125,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -221,12 +228,12 @@
     }
   },
   "timings": {
-    "capture_forward": 1.5388168222270906,
-    "capture_write": 0.5069590341299772,
-    "replay_probe": 3.009788952767849,
-    "score_forward": 0.47462915861979127,
-    "student_load": 19.50744666904211,
-    "teacher_load": 22.79794552316889
+    "capture_forward": 1.5579553740099072,
+    "capture_write": 0.506651840172708,
+    "replay_probe": 2.922064954880625,
+    "score_forward": 0.4500223221257329,
+    "student_load": 19.129065116867423,
+    "teacher_load": 22.79266786808148
   },
   "top1_agreement": 1.0,
   "topk_agreement": {

checksums.txt

@@ -1,11 +1,11 @@
-8787862b89f3803c10bcf864e349d7fefbd3d11c10f6fb77515cea588660a546  Gemma-4-26B-A4B-it-NVFP4/attribution.json
-5917457214cc198d7221bad6f531afddaccbdda64adf75e41c773b9832c24ed9  Gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
+0042ec48f341f100aea0279d3651b7ab3b14bf76ac4776679742a3835a37bcdf  Gemma-4-26B-A4B-it-NVFP4/attribution.json
+de2224cf09ff44eb055340123fabce5074e2d66db73210e760d291e5d88798e9  Gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
 6341a42f62589320b0701dc2677041ecd349847e0a7f32a50d3ff2770725d970  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4-qdq-manifest.json
-d09dcb71ea79fc96741b88e6e819263bba34da8f7b5192925db91e4e04390e08  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+1d754405fcbd41b98b1b6e7eef7c92a8c0da89ed33405671306b60c4eb776075  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
 c07ded1a130f6d28ddf3debad2c10d7e27a22bb53bae1ea132cf1c0d238c5b14  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4-qdq-manifest.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00002.safetensors
@@ -774,23 +774,23 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  Gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00767.safetensors
-325539f7f93755ae547d87cd914c5d590f2b7a4aea5ff5eae4fb89b335a7fb99  Gemma-4-26B-A4B-it-NVFP4/compliance.json
-fcb5f52395551026b3ddf5d8d0f7b54c2b3b751f45f79ee21f22166aa24f18d3  Gemma-4-26B-A4B-it-NVFP4/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  Gemma-4-26B-A4B-it-NVFP4/manifest.json
+4adeda55f1033130496c5611ae77c9e148dd9df00251a38123c337c4dd7a8de5  Gemma-4-26B-A4B-it-NVFP4/compliance.json
+0de1e655b8ad0ef41a439d00abf5d7ec7b7df42dd26950f9759d2ba73f71abe3  Gemma-4-26B-A4B-it-NVFP4/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  Gemma-4-26B-A4B-it-NVFP4/manifest.json
 c175fd2c40b96586dfb956c6fd44315dbe3538252646845597a986bd0d9a2965  Gemma-4-26B-A4B-it-NVFP4/provenance.json
-13f0b02a4807e26ae950aa648b5b1da1651e200895436cdc30fc6e449a46b492  Gemma-4-26B-A4B-it-NVFP4/report.json
-8aa4ea8c67cf87597408ca85c852aa782530831bb46696cf95a7700052bbd2c5  Gemma-4-26B-A4B-it-NVFP4/report.md
-6374093b56e06003618718473d57fa152d2e0b87367fd52c95b3d5fd3eb36d69  Gemma-4-26B-A4B-it-NVFP4/strata.json
-63a912ab3c25f650e144567c2fd20197e3f0a2c51f68945625d395301734e982  Gemma-4-26B-A4B-it-NVFP4/strata.md
-458ad2149d9fc4f27f3c9623ff31553cfd998bfe4e7ba028cb589bbc8a7596ff  LAWS.md
-b25b690322d7775e1e2f8cc70709b7adc041bd90926dad67bae4163453157c30  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution.json
-a3793cc728ba661bbc4275f6d3e091563a0f62cbc22d77041a91c224ef55e302  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
+57907fb1940697ba4895d09cf6c74e19162e1d3288f849d0fd58d45f8b758cb1  Gemma-4-26B-A4B-it-NVFP4/report.json
+6249dc19b25885f5127a7efc22bf740991713879eefd7ae628ea8033457b6682  Gemma-4-26B-A4B-it-NVFP4/report.md
+078b53087fbcda0f91dd0242561d8cfb179b6637aca4340d37668e26e89c9ada  Gemma-4-26B-A4B-it-NVFP4/strata.json
+19f37ca798f388d09479a655a792237792329c75c4f1d8d2489cad1f97239c63  Gemma-4-26B-A4B-it-NVFP4/strata.md
+8b7485af16fc5855b3d28171665f2e83afa319a512a73fa6c3a483b30e665e79  LAWS.md
+f0196f3e2a3386b9512431ca26a228e81d45f9054f5a2d19df098dea8ad501e8  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution.json
+57b34f584306df73aac44802ff1b14624299dbae6cd0fef42bc152970d04a4b1  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
 6341a42f62589320b0701dc2677041ecd349847e0a7f32a50d3ff2770725d970  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4-qdq-manifest.json
-d09dcb71ea79fc96741b88e6e819263bba34da8f7b5192925db91e4e04390e08  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+1d754405fcbd41b98b1b6e7eef7c92a8c0da89ed33405671306b60c4eb776075  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
 c07ded1a130f6d28ddf3debad2c10d7e27a22bb53bae1ea132cf1c0d238c5b14  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4-qdq-manifest.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00002.safetensors
@@ -1559,54 +1559,54 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  Neural-ICE-Gem
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00767.safetensors
-ce98aa64ec807c31d17a1b2b9bac9a626592652fecdeb98a2624f0fc3e493855  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/compliance.json
-d5ae43f7b6d9fafa8c865255c71f12f66ee62c47598fa677d83c7189631e1841  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/manifest.json
+5e539e7d033d934255debe56f281868378c651d97d70c2676ee8dd9253168aef  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/compliance.json
+9dafddd768995f6a83660ad2c41f238a12ec37dfd48709c3af2b23cdd1110b40  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/manifest.json
 dbfd855639d775b58d0de2c805c0af7c8909843cc2c86b0a94449125cc07c071  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/provenance.json
-64210e7384c8f18107f97cd519ad52061668b411ae897837095b5d68b98754bd  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.json
-15d3ed592a9469094d7a86ff47b782c18401bdcea5c26780fd3cf29a3d237e78  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.md
-c073caad8ba8793c6055b1c6583a37ba53ff43906806760460268483e0b4622f  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.json
-06358b003f525cfe8c4423cf55219e62a7162a205adcd229ba68448e8395823b  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.md
-adb4d43016f9081dc87b0f3a709e17bee53f69556898a3f9741df89c1f5c97c0  QXQ.md
-27b81b58042dced2b8b6ad1c7b4af8b853be81c99ce875cbe9dbaf5218e9f1ce  README.md
-29e7d497becf1682129a5c7771fdd79bd590303be21e12bdcec94950d0b0257a  baselines/self-kld.json
-156789c86d7b14b0bfbb0e6eef4275f8dd0b1f80b1a86c320a64dac3fa8c4e86  bxq-vs-size.png
-71ce75b3f6aa2b4a8a78846d9316b63b3dba8455ae0d79550dd6a7de1c65bab7  environment/gpu-clocks.txt
-20e46a5fb6e5d744dafe29066534032bf58a9a76d5d5e5c753ef40503794bce4  environment/gpu-smi-query.txt
-98d0669894325c5f2d55a4e64883b60cecafdde2a6a9b25a85b7cb12d85c644d  environment/gpu-smi.txt
+89e2cc4a0a86b3fe5a58530497bb5856e28e5740f7acec66699c182d1b70a18a  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.json
+2b9400cfbceac7262c056536f298889dd8da93e08a7df7aa53271b7197eca49c  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/report.md
+df99b57e3995a8f70cc40c7879396a2cd2795e78cace47fa52278029cf69157e  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.json
+c19e552cdf48aee0bc2d1f1ada5e99da41258e22e7e4157794fcf0450b61a00f  Neural-ICE-Gemma-4-26B-A4B-it-NVFP4/strata.md
+9b772f41e3d6af894e9d80f2bf68645100c21c6229f50cf9ca12a2f1cb71d434  QXQ.md
+ae389ce700843af78236aad23895546e701823b42b8ab331da862efe9c04bea2  README.md
+4aada9174334098222847b4346d9e428fe2afb882d8a796d00d5747198665a67  baselines/self-kld.json
+2e882dd8cf756ef5547a715cf8e35ce2b7d5db1163d59c675147525cc6312ad5  bxq-vs-size.png
+99a3267b0f7a1636e85fc6cec5b5438e9c19e5a22d99d242e3a0d39ab61f38d7  environment/gpu-clocks.txt
+c84041618eb70f0f966b3d748610afb94213564bdfcc10e6fda2a80dedd7727c  environment/gpu-smi-query.txt
+ccfcf8608ae6dd2ab0d2bb5ecd6a93ecd65f38b5927a47e3d84f7d87daa8441f  environment/gpu-smi.txt
 0948e1e93ea671e5d584fdcd893e989c0fd669a2b8f6fa3e0aedb32cd95d09f6  environment/gpu-topology.txt
-9d0f16bd4ce3c2530dcb26421cde32977c72871845084bd91e3d7f7ab81e4518  environment/host-cpu.txt
-bf6acfb409bb077fcbd526faf927ac5a76670c80a44c7579ad2bfe556490bb0f  environment/host-memory.txt
-e30b0aff01bc937699db5a35b61b53966dc7e995c69a1cb198081b23ef2cb0bd  environment/host-mounts.txt
+094fb13fe7aa06fe35d96aaf4508a76b4c66b1404f26ad72d75f91fc16c0dcce  environment/host-cpu.txt
+4d5a23786cc7b208d58046b3eaae0087435885240d86b193cfe4231ab3bb2b59  environment/host-memory.txt
+d64a3c158ef5a0f22017cf82691f1d04ee469d0ffcc97e819f3b614bdc20b6d0  environment/host-mounts.txt
 1869a8ac025334480b6b464255abd14d55fa69d2cae23a7740ecf273727b8d45  environment/host-os.txt
 66ba81f841a1f5eb8c7ee29c87dba7daa1c3588cd791aa1f754952ea81ed0626  environment/host-uname.txt
 f4045294424e8624a17a0e91be63846a1fd1d7a494fa6e946acb85322020fdd9  environment/models/Gemma-4-26B-A4B-it-NVFP4.txt
 880bcc4d1d3bf503033ea35f30c56496519b9389f5b1e0226a6f723f1006e307  environment/models/gemma-4-26B-A4B-it-AWQ-4bit.txt
 dd4892431f129f9db10dff7febf58fa2ba18b823bace0100028177790e5c70b3  environment/models/gemma-4-26B-A4B-it-FP8-dynamic.txt
-b4add09aa4a37c5c641a86471ed4c55cc7365d622b3f0f97a0283f55fe8c38a1  environment/models/gemma-4-26B-A4B-it-NVFP4.txt
-f41f4a64202beac98f0f8edcff63fb0f06dee8be425d7a373f82ebb99d7b1a76  environment/models/gemma-4-26B-A4B-it-W4A16.txt
+b3d9a52901f5c06ff878587fae86a8a83eed8e31e8bfc519afe30c0f895e8a42  environment/models/gemma-4-26B-A4B-it-NVFP4.txt
+d410331f950c2579b1f0b3d99a5ab4b6f88f319e559b0472f4b4a23e05e5b7e1  environment/models/gemma-4-26B-A4B-it-W4A16.txt
 0578fbfefe1f39e27d6de807ada9d5d2c8086e1b11f0a1389a46afa225acb3c8  environment/models/gemma-4-26B-A4B-it-int4-AutoRound.txt
 73c04bea441febbc0eb093cd2e711684756f351685ea7eed6bed4ffd0180dd48  environment/models/gemma-4-26B-A4B-it.txt
-ffce623595dbe8d7c618cbf6da7b493fa5c367e7ce22d9e0d5bd468e39381691  environment/pip-freeze.txt
-3693fe8300bbf8e34fc33bbc156d3416b21a93798a44e2728fa2bc243748b5dd  environment/repo-git-describe.txt
-e9432f2e4000ee15b03923d1cb7055521c7545babc5029667059d02c0cba1acd  environment/repo-git-log.txt
+9944754d06c1f1ba0ede60cf741414f7e3cb9edc3f4b72b0558adc4f96d47a64  environment/pip-freeze.txt
+4a86f2ec35346da468e93c28537163eae96570aa29bc45c3b1feb818f7c3861a  environment/repo-git-describe.txt
+436c9b0de3d6ef3c032dea87cce5a550acacc056e5534f2950e97611c3dc189d  environment/repo-git-log.txt
 c2a25b1c9a46dd870f8f58312dac0aca13d7d0a0d8f18e1f34d7b628cf94a35b  environment/repo-git-status.txt
 e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855  environment/runtime.err
-2f4046f6118864524c755f534997fa94da14004ce97c57a2292cc34eadb1cf8a  environment/runtime.json
-7ff9d2c86969f1b499866277d1cbb30a4bcb4c3469497992007728a2cfe12c57  environment/summary.md
+07fb9046fcfceff740452ec8acf5b6f0f5c411f1347728466f0a21ee3d56421b  environment/runtime.json
+19ac82efe06498f376f0cfd0ebaaa7fc3858391eee811432df3a59508ce0e986  environment/summary.md
 1a85dc0961f460a894135050a328eb11783f46888d03b020f957ab0fdff09ae9  environment/toolchain-gcc.txt
 1b68796b137798804d656d009a5a89bee4afaa91548a0c7f5c34ba91985c04d1  environment/toolchain-ldd.txt
 579b25f79a97a2b5df173c5530ef6cd925c4d7cbe935981be17b61f81324ce44  environment/toolchain-nvcc.txt
 e23ef1779f1deeb02df894c0df6264615e00b353fef264311cae9e47b9746bfc  excluded-candidates.json
-87aef3ae05cbc5b8bbcb3a7a6b01d7a1a90ab01da7f82263b6ae45d859157d95  gemma-4-26B-A4B-it-AWQ-4bit/attribution.json
+e7b01b1bbfe9795f15f9d1d312296b57d44239cd15021f5a3f82a02bd6a73069  gemma-4-26B-A4B-it-AWQ-4bit/attribution.json
 ee03c1af0c53ff2527c9c05c9605aaf282b9900f4c71dc256b5ecfea81579ebf  gemma-4-26B-A4B-it-AWQ-4bit/attribution/bxq.json
 95a546cafffed2f8b4d57e9036a44fd716cb337890f45615bf905deb16fa6932  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym-qdq-manifest.json
-844aa0a080fd4d90714b7ebc0db5f8c87d15fa92f857230f4028cd1b6efee488  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym.json
+0a038edef1cf7c76da9a661a268d37c305fd1ee1c27d51b01d3065527e946623  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym.json
 29f86272efb12e607bce7289ad94b2bb0ea76ca741f731232b1f1d6957dea165  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym-qdq-manifest.json
-f2bcc004757e01983b25123b6ef3c210704f5c24987a395a45e54a9542db4dc9  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing-manifest.json
+fa0f4f46f83915fb01f1e97ea0b5dd8381c54c44a813ee9e4f8785d6b3bed1ad  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00002.safetensors
@@ -2375,23 +2375,23 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00767.safetensors
-dbc99b42222444eae461c1dcd41ac6a67038b7e41494badf0b208e90ecf4ebd6  gemma-4-26B-A4B-it-AWQ-4bit/compliance.json
-158df15ba76a60d154ec3cbb757bc46d34c2161fae58e48244a998f53478a1d1  gemma-4-26B-A4B-it-AWQ-4bit/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  gemma-4-26B-A4B-it-AWQ-4bit/manifest.json
+396071fa90919bff9263ed8f94218097d29b05b47f8f4ac9c9ba71a9b62c8ac6  gemma-4-26B-A4B-it-AWQ-4bit/compliance.json
+84703048a0b5988e4b4900302d6dd572b006b4b276f8c0b77158dc14510ffe13  gemma-4-26B-A4B-it-AWQ-4bit/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  gemma-4-26B-A4B-it-AWQ-4bit/manifest.json
 8ef16987cdccd9a222111fabd0bdc837c41b97bbe1c0aa01a7641ff537e674a5  gemma-4-26B-A4B-it-AWQ-4bit/provenance.json
-a31fa6da274f2da9ad09c98b457cb057a78d7045b6f29ee1b7f33418fce0e6a7  gemma-4-26B-A4B-it-AWQ-4bit/report.json
-896410c1d321723440be730a3731533f082404b919fa59516ea09c62061b9bf2  gemma-4-26B-A4B-it-AWQ-4bit/report.md
+3755274c70924d33953cb899fcbaa880fcd2ce858d4e0cbddb1ad827b5757062  gemma-4-26B-A4B-it-AWQ-4bit/report.json
+467a197ad0ab89f89755be92f347da80e93fd7d383479e82ae7d930f16fcd9aa  gemma-4-26B-A4B-it-AWQ-4bit/report.md
 762f792812a73305aa2a2fef6e8b1b17d0df825b154f83bfc53a0cc1759eb7cf  gemma-4-26B-A4B-it-AWQ-4bit/strata.json
 1b5dd5e052247057fdbc01dbe2a5a11d0b53cdca9e5a82a84a5dc28b20e01d6f  gemma-4-26B-A4B-it-AWQ-4bit/strata.md
-c365b5f92c560d75ed8d65485007151dc30b98b7fa476378b0aec5ed80db5a0e  gemma-4-26B-A4B-it-FP8-dynamic/attribution.json
+325924097077dde910c9c6a6a7d10fc312fdd7c78c37d770ad776b9796a8e019  gemma-4-26B-A4B-it-FP8-dynamic/attribution.json
 56e14de2cfd9d3db15819c509d49dcf63f778a9db767c1e029f9876869cfb208  gemma-4-26B-A4B-it-FP8-dynamic/attribution/bxq.json
 fc4518c69a1f476f8609339c8b06b832801b09e4100a0385f29c0b83af02f605  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-fp8_per_channel-qdq-manifest.json
-9b5170d26dda334cf0f4c4bf39044cc0eb7e255dbd59f9faa4884dc20cee7a0a  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-fp8_per_channel.json
+33cd6ceb3e27344ef58f72ef706a0e8519509e4a722ea118c6d60139871660cd  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-fp8_per_channel.json
 6b5d02b900ddb6089910611736deb7218e77d04b697cbd9378c3b36a4861a269  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-fp8_per_channel-qdq-manifest.json
-d2dc6725deb3dad3cf9bd09c9572747d838cd8678bdda57bcaad6a1ccbf63aba  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-fp8_per_channel.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing-manifest.json
+a2494c417781a4d700592b5749e1bb15c28e59b151beba8955b9297d5ea1b9e0  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-fp8_per_channel.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  gemma-4-26B-A4B-it-FP8-dynamic/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00002.safetensors
@@ -3160,22 +3160,22 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  gemma-4-26B-A4B-it-FP8-dynamic/attribution/routing/routing_00767.safetensors
-49fdd1d9b54095b91fdefbdc4292dff1812cb5e1d454e239fb3cf7014db27992  gemma-4-26B-A4B-it-FP8-dynamic/compliance.json
-9ef5acba49a153774d41a412d086d25a660663f68b632cf28f0990ef19b6a5be  gemma-4-26B-A4B-it-FP8-dynamic/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  gemma-4-26B-A4B-it-FP8-dynamic/manifest.json
+c738840c5b61255054afc6276a7231ee036df82d8bd7bb6e1d90fbb36a7b662e  gemma-4-26B-A4B-it-FP8-dynamic/compliance.json
+879e9f5d62c8dd3017ecf17104391b26e525c137701e8cb5dbb797d6ba429116  gemma-4-26B-A4B-it-FP8-dynamic/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  gemma-4-26B-A4B-it-FP8-dynamic/manifest.json
 012c69a12da1e3df6777194a5f39440a3db002c13b963e1348f2741ff2bfcced  gemma-4-26B-A4B-it-FP8-dynamic/provenance.json
-a2d4c10b82d5a04a5a50c1b189da1cf9a3613e1c630344609d98a9b9ca2b5112  gemma-4-26B-A4B-it-FP8-dynamic/report.json
-50c2db86093102f5ab7a86df64a559cfaf6db4a567d78366d70ed4272e770ba0  gemma-4-26B-A4B-it-FP8-dynamic/report.md
+e47357985d8de92d062ccd241b20003f7320bc6e35c6003db52948ff36444efc  gemma-4-26B-A4B-it-FP8-dynamic/report.json
+150281a5269c4bcb304dade944a2fa777f68cbab7350bcbc9dec06cd3def0cd5  gemma-4-26B-A4B-it-FP8-dynamic/report.md
 6161d1875ed85688f779be905f306c57b4ab1068bd8502b59b19e8311f8d7412  gemma-4-26B-A4B-it-FP8-dynamic/strata.json
 5e1fde7a4ba4abb8a10e05345668ae91c806eb2e77822aa2bed4c2455d8465bf  gemma-4-26B-A4B-it-FP8-dynamic/strata.md
-3d0f6c097c9b8680ac4d075d8e163947d56d5497de9226117755420b158a8b3f  gemma-4-26B-A4B-it-NVFP4/attribution.json
-7385709eb3087845cda21d177fce2e691f2aa1209542b800a81ba13227b32527  gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
+c6b65dc27ae1cc154f74e3e80378b5549ded352aa393a6ddfefc8bc056351e0e  gemma-4-26B-A4B-it-NVFP4/attribution.json
+3c709d1f1ebdaaeadee7eebf111f18ea4811ffa421cb4de9ed3818e0ee5e8ab2  gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
 6341a42f62589320b0701dc2677041ecd349847e0a7f32a50d3ff2770725d970  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4-qdq-manifest.json
-d09dcb71ea79fc96741b88e6e819263bba34da8f7b5192925db91e4e04390e08  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+1d754405fcbd41b98b1b6e7eef7c92a8c0da89ed33405671306b60c4eb776075  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
 c07ded1a130f6d28ddf3debad2c10d7e27a22bb53bae1ea132cf1c0d238c5b14  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4-qdq-manifest.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00002.safetensors
@@ -3944,23 +3944,23 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00767.safetensors
-fa7ad69a2931f5bb5dd5fc50c513a2b7c81ebd7f95c101187fdf95d3ce7f8b16  gemma-4-26B-A4B-it-NVFP4/compliance.json
-a8aa96d4fd4a7b77341a9943ce3422f438f388bebee9fc3d174efa6f81d7ae67  gemma-4-26B-A4B-it-NVFP4/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  gemma-4-26B-A4B-it-NVFP4/manifest.json
+4674ab8146f79f5b63fd8a2f8c3d954dc391860cf23b76cb194b9342835b574f  gemma-4-26B-A4B-it-NVFP4/compliance.json
+3d92bc9b8698bc0c35f0606818d701191e378220ba456db9b0e24fe5a360910b  gemma-4-26B-A4B-it-NVFP4/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  gemma-4-26B-A4B-it-NVFP4/manifest.json
 b912c1094667e6d6d42b62b2f9b4ada71ce7e157824a4edbb97411063ea25240  gemma-4-26B-A4B-it-NVFP4/provenance.json
-79d2cc74a06ce450c5f3b7baff77b5f6623377fb4ac5d45a23a2e3bf22bc7aaa  gemma-4-26B-A4B-it-NVFP4/report.json
-21d07e6bc9917af756f359762d0e65ee6d2e257fc025b5a45a8a9869741eb651  gemma-4-26B-A4B-it-NVFP4/report.md
-20abd21bd5fe16388bbc1db76af442785cb16587ad08ca687b531e29c472373c  gemma-4-26B-A4B-it-NVFP4/strata.json
-bf6a0b26d72fecf79c175c1a3a446fdff30b0da389a85af3d3f7cf6903a7bd01  gemma-4-26B-A4B-it-NVFP4/strata.md
-1f8326d6b68fdc9d90e7f2698e93d30825676748a1e38df36bef35fc517cabda  gemma-4-26B-A4B-it-W4A16/attribution.json
+f7ab6593cf691d5d9683543e132c4445cbd9ee754bad66e36608a11b67133310  gemma-4-26B-A4B-it-NVFP4/report.json
+e7bda14667c2d0399cc1ecfac5924d786f32265cd3f28df8d3bc642619e369e2  gemma-4-26B-A4B-it-NVFP4/report.md
+573aa840d5a7d059a6427403c7b8b7d9467e49552b6919ba328da3d0003b6f95  gemma-4-26B-A4B-it-NVFP4/strata.json
+dcb01d759e80ac29e8f6590e8b5d31d463c52161bc0ef89048bea6780c3ad683  gemma-4-26B-A4B-it-NVFP4/strata.md
+0f340dbe97d8252a1a278644d838cdedb96de3c40dafdd385a667c0850f49f52  gemma-4-26B-A4B-it-W4A16/attribution.json
 1c6200099837db161a106135dd4295f0f7f3e42f900acd005c55530ae3025623  gemma-4-26B-A4B-it-W4A16/attribution/bxq.json
 b7eb22c6bf18a2c214aa59f318917f19237a5a921348e023333a35dcbc9f1183  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-int4_g64_sym-qdq-manifest.json
-575237f254006bda98e3f3a851834f3597fae7b110904adf6858e533a1111942  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-int4_g64_sym.json
+ad95131b500732fa89774ecb5f6c07e9b8f007884246cde42871663c8be25ff6  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-int4_g64_sym.json
 3cd87d5de0eedffc935e1377aee71c134cf4aabf41d7adafe950b810a37e7d18  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g64_sym-qdq-manifest.json
-b257d209031a46233acd70a901226044d03d110601b810ee80a33a1966875d38  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g64_sym.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing-manifest.json
+8ce0aac455f4b8ac7b51c0378a1295dc6e62c1a9447149304f416944d7065153  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g64_sym.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  gemma-4-26B-A4B-it-W4A16/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00002.safetensors
@@ -4729,25 +4729,25 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  gemma-4-26B-A4B-it-W4A16/attribution/routing/routing_00767.safetensors
-eaf40f5f541bc8f4792b1fc2f15f234670a3aeb4ee0fe0383553836e120cc28f  gemma-4-26B-A4B-it-W4A16/compliance.json
-4d0b7f7ff671ebd35a9000f37ee03b851d03f9d30ca9f67be79ada8004fd3d16  gemma-4-26B-A4B-it-W4A16/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  gemma-4-26B-A4B-it-W4A16/manifest.json
+a79be1ba7b8d6566c0f3b2613ce342dc26d4f3fe595cfe1041f09fe36e40149b  gemma-4-26B-A4B-it-W4A16/compliance.json
+c2df6223894c4d9ef2e9fb3aba532f2c00aa959dd5596101230b7450b6ad5edc  gemma-4-26B-A4B-it-W4A16/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  gemma-4-26B-A4B-it-W4A16/manifest.json
 22d3392fd18f290e3ca668a07b7ce4b421bd0eb318f4d6a9960a3ece5127c3d2  gemma-4-26B-A4B-it-W4A16/provenance.json
-14fd0c9ece7a4a735b38a98b515a539a9110eb34fae8a784165f59f8f2110c90  gemma-4-26B-A4B-it-W4A16/report.json
-fe5eb21b8e1683c2ef45761673ac59a991a1bd1d54c57076e972a35b86c0c2d2  gemma-4-26B-A4B-it-W4A16/report.md
+9ab847df40eeb008f191f3db226b0b3737f6cbff1865c353a892163db6b61960  gemma-4-26B-A4B-it-W4A16/report.json
+a240cab8d9fc46c997b74da03d6d8f5368adba543312962fa3fdcf36ff6c2845  gemma-4-26B-A4B-it-W4A16/report.md
 c9b912a31dee082614c269634cf9320632748f214263287f74b506f8e252c577  gemma-4-26B-A4B-it-W4A16/strata.json
 d4f5e797a15a062993c5da2256169b54f434b0c43d4ec28ea2546b13d1437fca  gemma-4-26B-A4B-it-W4A16/strata.md
-11518f2a284d6537a2f206f157c864697b7ed1954a6446be2de0357dc611b1b4  gemma-4-26B-A4B-it-int4-AutoRound/attribution.json
+b2dd7a0867adf02316ada2c2535f0ac4dab4bea6c9738cab6cf316b0c2b1866a  gemma-4-26B-A4B-it-int4-AutoRound/attribution.json
 315a9c83b09bc4395b12015769a25f8a490287ac7b06bf393f2c21b006d1e675  gemma-4-26B-A4B-it-int4-AutoRound/attribution/bxq.json
 665cb06f0466573371227fdeabae4d3a2aa14d3a070101c666c8124cf9d96300  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g128_sym-qdq-manifest.json
-f653b1f1e6aa54116aca287921c9081750e65f6f546da195fda1066b8b1d93a2  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g128_sym.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+22bc083a4ab61f17c5d25a45e708d6abb44eec8c8e6fd7cfd878629fe7b1ed43  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g128_sym.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
 c2d03984b1aa5571d81a05d4a7b46d04b495e168a4cc0493a31d5d56376ab6e4  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-experts-attention-dense_mlp-int4_g128_sym-qdq-manifest.json
-e327a00b9690ca57b110b540dcb1143b44a367fc15ee88812eba7d646d7283a0  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-experts-attention-dense_mlp-int4_g128_sym.json
+4027a9083df1e135c4c6001d721c5fae687310b15e376d090b8d15119351076b  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-experts-attention-dense_mlp-int4_g128_sym.json
 8f9469249982bdabf76fbfcc345dc5bfd0ed3a5be41befb3d8c3ff22bbd07e3d  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-int4_g128_sym-qdq-manifest.json
-71ba8c2246e1cc6cb50d7b88a5ac2c0c96d8c748f13f713b692af64282d91e4d  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-int4_g128_sym.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing-manifest.json
+5ddecbf00c7f70de7237b150bea4e00f37f4fd8c230308b61858e117f75c776b  gemma-4-26B-A4B-it-int4-AutoRound/attribution/gemma-4-26B-A4B-it-qdq-router-int4_g128_sym.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00002.safetensors
@@ -5516,25 +5516,25 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  gemma-4-26B-A4
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  gemma-4-26B-A4B-it-int4-AutoRound/attribution/routing/routing_00767.safetensors
-ad99660331bfe9d163378b743a28a0c9a8060794b35cb94bef78406b0a1bc589  gemma-4-26B-A4B-it-int4-AutoRound/compliance.json
-a42eca79ec888e74ac2e6afae1b2a2d679daadeaccba3f35be79653bec9a4c99  gemma-4-26B-A4B-it-int4-AutoRound/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  gemma-4-26B-A4B-it-int4-AutoRound/manifest.json
+673d472126684e0919452215d962bebd5b37e3675de602a72b12c04d118115b9  gemma-4-26B-A4B-it-int4-AutoRound/compliance.json
+00a0effafe32a8d4e3e2dc9ce1a6b9428ada966c25c480cbe2a9d1ed61f49e25  gemma-4-26B-A4B-it-int4-AutoRound/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  gemma-4-26B-A4B-it-int4-AutoRound/manifest.json
 a93112dbd2d4a1e80869ea1921154bfcd851902b4480a02e974f616373bb4289  gemma-4-26B-A4B-it-int4-AutoRound/provenance.json
-9bf6ef21d1315be887dc8257a524c953e20f9be720252a2e7c66ef2980ae2ffe  gemma-4-26B-A4B-it-int4-AutoRound/report.json
-7fafe74bf7aed8a52c6390e891ce3c664a5bcd1a8e843065acbe39ea3199d60f  gemma-4-26B-A4B-it-int4-AutoRound/report.md
+08ddb32451bbf08e8061cf5aef74dd2a633bcbce2ac16783f35d3d53c5924717  gemma-4-26B-A4B-it-int4-AutoRound/report.json
+c33f60d51d6876be1734fc8b5ee1d3dbc7636af7737379866b50e0ae75b41161  gemma-4-26B-A4B-it-int4-AutoRound/report.md
 27c181d4d9d3f959bab20b37bfe678666e453b2d047dd664e4f5a6760e44702c  gemma-4-26B-A4B-it-int4-AutoRound/strata.json
 5247bd5a68fb5e24f746bae0042b175295de15b78b73e69b7f0d4ceb73b27b5b  gemma-4-26B-A4B-it-int4-AutoRound/strata.md
-2a2d8e17b597a083afd806dc5dff079f6f50225b506dda259999d13f1fd89ae8  kld-vs-size.json
-7e29df6d3683b3257f6e25453aeddd8c4e57483944b3b7ee3bd01a9b90e316be  kld-vs-size.png
-276044b7f1007e59034e2398955bdf48fb25df4882e4b007fb5a6ebae9f3803b  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution.json
+1bc560bf586b34b38a23064504a1429da353e7eff4bd011deb3a910f735eae62  kld-vs-size.json
+713ad4895d3f9eaaea9bc8bcd071a9713b33629aed47b0db4a6bfa089cd98200  kld-vs-size.png
+0bed2b610b84912a7c3091fbb5511b92b43e7070e77c83db2335b59d5ff52a0c  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution.json
 5af848a9d0afac809893f80c3dba9f4e325cea23243614d94f69a765ba2343c4  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/bxq.json
 95a546cafffed2f8b4d57e9036a44fd716cb337890f45615bf905deb16fa6932  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym-qdq-manifest.json
-844aa0a080fd4d90714b7ebc0db5f8c87d15fa92f857230f4028cd1b6efee488  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym.json
+0a038edef1cf7c76da9a661a268d37c305fd1ee1c27d51b01d3065527e946623  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym.json
 29f86272efb12e607bce7289ad94b2bb0ea76ca741f731232b1f1d6957dea165  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym-qdq-manifest.json
-f2bcc004757e01983b25123b6ef3c210704f5c24987a395a45e54a9542db4dc9  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing-manifest.json
+fa0f4f46f83915fb01f1e97ea0b5dd8381c54c44a813ee9e4f8785d6b3bed1ad  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00002.safetensors
@@ -6303,15 +6303,15 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  lcu0312-gemma-
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing_00767.safetensors
-ba18500c34fd963ff061c0ce2c865337e27c21d85010268715bf7c74f18e0050  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/compliance.json
-c319e78aa95c3f7c1851f1227fa0feccb2736c81934ef66948e8a0cd2ffdb19a  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/manifest.json
+aa1c266327774537b4a4fefdb474865fb63b88752c888755e825c75cebdea668  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/compliance.json
+27d07b95cddbdfe44e51744db0f2fe43cb6515c50ab249d2d2ff681a139a8f14  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/manifest.json
 c67582b0e1581873e3bfddb1cebb6d7dfd66018fd007e61e1aa1f6bfa57b6c19  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/provenance.json
-a87a9a63617ce97b479291f479f9834484707456a4e6dd0df97a8f4376669119  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/report.json
-ef98612ad1516756f4f83760deeb60ec726237746a32b63607604f31c1b1230f  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/report.md
+72739f2a9b36c8352176f12f3c0ed066efae16e15c3be345d539c8fee00046f7  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/report.json
+6efdfe368fe3c3e06cda45023004996765f19582a8a268e254040a2030491ec5  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/report.md
 b8ffb69716b1a32af6b33c185df6245ea7f2ce4f38b1a230c358a17c56f71f70  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/strata.json
 9f4e24428d03d4850f89d1d29ea8250112ce6011bb1388a52f573b9ab9ba738a  lcu0312-gemma-4-26B-A4B-it-AWQ-4bit/strata.md
-7e29df6d3683b3257f6e25453aeddd8c4e57483944b3b7ee3bd01a9b90e316be  qxq-vs-size.png
+713ad4895d3f9eaaea9bc8bcd071a9713b33629aed47b0db4a6bfa089cd98200  qxq-vs-size.png
 a4c04ff540324d1c73202e30b034d1be9d068c8f83e9f0249ec8b08755bcde76  reference/hidden_0.safetensors
 188b6affc744ed627301fa665c37c595d363611bab58ec4f0836eae577c83126  reference/hidden_1.safetensors
 08d7e8b3b4f4aace7f38f76fbaea46ca260d5bb8f7ece1301a9da1ac70b78833  reference/hidden_10.safetensors
@@ -7082,7 +7082,7 @@ f6107bcde93dd7bdee83161f0e085a90deab9ec99bc910bbf232dd35451f371c  reference/hidd
 6bbcfd2f51d8b702d115c4230d02e9cdbc6fd878b4ed788e0eed2ffb3745bac7  reference/hidden_99.safetensors
 3a92448a235b5cd0f5c187ff34835d0d23a0b0e35b59b60f723d602c5a3c79f4  reference/lm_head.safetensors
 733c161bf801ca7ea977e343f0663b579a4b2021bdbec75ce2556e0137ffb3fb  reference/logits_0.safetensors
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  reference/manifest.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  reference/manifest.json
 33ec6c73ab7e340c0b1b218ec490deec1b7f093c9b5e2b81d0010f50ac6af5c0  suite/partitions.json
 cf13ad5411a70ae341179b29965c95211dd89cf6a08f04f391f88ef61ec0fa26  suite/source-registry.json
 b4ddafd5d0e80dc536613824988c4639b64a8ca61b8f9debdc6ec3d5156a0c75  suite/sources.json
@@ -8112,14 +8112,14 @@ a1575fef2be8630e79bcb99ea1e09239a488d7e385473d5a597b3fcfd30d54fb  suite/tokens/c
 4f8b2a26c6fa2fe01215d368d1709413ee79ad12532a5394fc6ce523706bc4ac  suite/tokens/context-1022.json
 e570315115fbb04b5e4328a899a786c69127bfcb842342dda3bef92e71dfd201  suite/tokens/context-1023.json
 36b7275c1c0df25256687dd4a67364c3b2da0e53113c41ba51299465df357b3b  suite/validation/capability-overlap.json
-bff10d742135cf559f6904512484e22d9defbcd4ebcf7bab4bee815ee7891179  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution.json
-5584907096e490db5f3e2f9447c4bb0bd4e58a803b751f3e755f2b6bf6509213  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
+2ed7868029aa93dff5912325569180252ea1f3bab734a8bd222bc3a9ee9474d0  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution.json
+25d53b245b112515e16226f3f3c551af544ff7c73a1dc92c28efeb95efc47e8e  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/bxq.json
 6341a42f62589320b0701dc2677041ecd349847e0a7f32a50d3ff2770725d970  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4-qdq-manifest.json
-d09dcb71ea79fc96741b88e6e819263bba34da8f7b5192925db91e4e04390e08  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
-42f414528e51ef5f10ce276e80d582a9ad70b689e96d2ddb5aa46cfadca2e0e0  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
+1d754405fcbd41b98b1b6e7eef7c92a8c0da89ed33405671306b60c4eb776075  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-dense_mlp-nvfp4.json
+cc30304ecc9bd4d6b58477d834e0f238c4c33bf4fa391a360439b10c47432a52  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json
 c07ded1a130f6d28ddf3debad2c10d7e27a22bb53bae1ea132cf1c0d238c5b14  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4-qdq-manifest.json
-43b796f6127ba10f12994046152be8f1e9ae50566d909cf56eb25b564880b579  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
-65a995be0ed6303769d53ef0e1a4be1cb260707bbabf0ab6228e281be8a4b38e  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
+8f0bc8adcc0456ac2d31b169bc83664f5a0a0666e5f95271db2a0eb1b1ab31a6  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json
+0fcde10d9c35222c9efe3362b3439a790aec49caa70d829bd455bfac62b3ce9a  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing-manifest.json
 b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00000.safetensors
 91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00001.safetensors
 8f60ee2f976b0d66b1f81721901d6a86d51030e7329ef75fa7a6ae11a6f1d98e  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00002.safetensors
@@ -8888,11 +8888,11 @@ b779cd665458525be833699b0410d646c4d3ab86895ff45fb2e2d0e97ccf2298  unsloth-gemma-
 c153d4991edd679f6cd5e01f16aada000771f39de47cbaf8902b2161d94efab1  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00765.safetensors
 a93ece76c557b61ddb06094f8f5add3e6aa0a5b65d7aa1ea7b7b674e9f716cd4  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00766.safetensors
 27010839450156046282d6c6fb18fb00fe5f23934ee0eaa8e9e8f1def971b4ff  unsloth-gemma-4-26B-A4B-it-NVFP4/attribution/routing/routing_00767.safetensors
-815124b43b3ab82a05030c9de7f7c26ef64a861c31b54a41fcb4b860814243af  unsloth-gemma-4-26B-A4B-it-NVFP4/compliance.json
-4fe25f36705fe8106ab1b7c7c9bc7fd95a71e1702d225c59a6e0619f556dabf2  unsloth-gemma-4-26B-A4B-it-NVFP4/inspect.json
-d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a  unsloth-gemma-4-26B-A4B-it-NVFP4/manifest.json
+1b6b29d763bcd373e7dd12fbe289cb3c3929a9fcda2880a268078a7ad68a595f  unsloth-gemma-4-26B-A4B-it-NVFP4/compliance.json
+78ab7741a0ff8f900462ee2e5286d835707f0e02ca6bec8c8117ff28138f17b9  unsloth-gemma-4-26B-A4B-it-NVFP4/inspect.json
+b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef  unsloth-gemma-4-26B-A4B-it-NVFP4/manifest.json
 657a736ad94e7849bc398130fd6df6fe666f877ec1561dcb49ccb13a01703c38  unsloth-gemma-4-26B-A4B-it-NVFP4/provenance.json
-8ebaa87d7ee324b770b513ddb4f37956c71a4afcfdbb87b0056a2688290b333c  unsloth-gemma-4-26B-A4B-it-NVFP4/report.json
-851927d530519f822492d16ca6420556b4d4563d369638eac4f90420d4e7b4ea  unsloth-gemma-4-26B-A4B-it-NVFP4/report.md
-869899e33de175eaec0686226adc7e652415165cf305ec74d5f13923676faa80  unsloth-gemma-4-26B-A4B-it-NVFP4/strata.json
-f1d8f11247eecd344096804bbc7050248dbc24ceff6155bdb702d67bb0c5f21a  unsloth-gemma-4-26B-A4B-it-NVFP4/strata.md
+3ad6acd6d386c3f65a2a4b0ca72bc18a81c0c927bf8c8bd5d78d47907cd8d34d  unsloth-gemma-4-26B-A4B-it-NVFP4/report.json
+328ec27c87ec94b6fc7e947cbd56fe6c8b8f24891c4a1b50dc4537518d2f017d  unsloth-gemma-4-26B-A4B-it-NVFP4/report.md
+723f42e2dd8e50ec4abdc07df9684e49571d675c31ef01afa70ba7179a0dbe16  unsloth-gemma-4-26B-A4B-it-NVFP4/strata.json
+a62157162bd3f999d4e432f30e6572c7f4bfc8b7c367fe3132dd55006978222c  unsloth-gemma-4-26B-A4B-it-NVFP4/strata.md

environment/gpu-clocks.txt

@@ -1,7 +1,7 @@
 ### nvidia-smi --query-gpu=index,name,serial,uuid,driver_version,vbios_version,pstate,clocks.sm,clocks.max.sm,power.limit,persistence_mode,ecc.mode.current --format=csv
 
 index, name, serial, uuid, driver_version, vbios_version, pstate, clocks.current.sm [MHz], clocks.max.sm [MHz], power.limit [W], persistence_mode, ecc.mode.current
-0, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1794225038968, GPU-41235b51-d6ab-afe6-c393-e1c377b81299, 580.173.02, 98.02.81.00.07, P8, 1170 MHz, 4087 MHz, 575.00 W, Disabled, Disabled
+0, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1794225038968, GPU-41235b51-d6ab-afe6-c393-e1c377b81299, 580.173.02, 98.02.81.00.07, P8, 1177 MHz, 4087 MHz, 575.00 W, Disabled, Disabled
 1, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1793025105275, GPU-547cb69a-4bcf-5f0a-c154-5397b1dc5a78, 580.173.02, 98.02.81.00.07, P8, 1177 MHz, 4087 MHz, 575.00 W, Disabled, Disabled
 2, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1792925028706, GPU-0f14730c-9c8d-aa20-9478-324d31766353, 580.173.02, 98.02.81.00.07, P8, 1177 MHz, 4080 MHz, 575.00 W, Disabled, Disabled
 3, NVIDIA RTX PRO 6000 Blackwell Workstation Edition, 1794125045055, GPU-2aea821a-9e57-5a64-9471-83306dddcd97, 580.173.02, 98.02.81.00.07, P8, 1177 MHz, 4087 MHz, 575.00 W, Disabled, Disabled

environment/gpu-smi-query.txt

@@ -3,7 +3,7 @@
 
 ==============NVSMI LOG==============
 
-Timestamp                                              : Tue Sep  8 01:43:22 2026
+Timestamp                                              : Wed Sep  9 02:41:47 2026
 Driver Version                                         : 580.173.02
 CUDA Version                                           : 13.0
 
@@ -49,8 +49,8 @@ GPU 00000000:01:00.0
         ECC Object                                     : 7.16
         Power Management Object                        : N/A
     Inforom BBX Object Flush
-        Latest Timestamp                               : 2026/09/07 20:07:16.694
-        Latest Duration                                : 30306 us
+        Latest Timestamp                               : 2026/09/09 00:25:04.592
+        Latest Duration                                : 66488 us
     GPU Operation Mode
         Current                                        : N/A
         Pending                                        : N/A
@@ -85,10 +85,10 @@ GPU 00000000:01:00.0
         Bridge Chip
             Type                                       : N/A
             Firmware                                   : N/A
-        Replays Since Reset                            : 41090
+        Replays Since Reset                            : 44280
         Replay Number Rollovers                        : 0
-        Tx Throughput                                  : 2259 KB/s
-        Rx Throughput                                  : 1494 KB/s
+        Tx Throughput                                  : 840 KB/s
+        Rx Throughput                                  : 531 KB/s
         Atomic Caps Outbound                           : N/A
         Atomic Caps Inbound                            : FETCHADD_32 FETCHADD_64 SWAP_32 SWAP_64 CAS_32 CAS_64 
     Fan Speed                                          : 30 %
@@ -113,12 +113,12 @@ GPU 00000000:01:00.0
     FB Memory Usage
         Total                                          : 97887 MiB
         Reserved                                       : 640 MiB
-        Used                                           : 4 MiB
-        Free                                           : 97244 MiB
+        Used                                           : 2 MiB
+        Free                                           : 97247 MiB
     BAR1 Memory Usage
         Total                                          : 256 MiB
-        Used                                           : 3 MiB
-        Free                                           : 253 MiB
+        Used                                           : 2 MiB
+        Free                                           : 254 MiB
     Conf Compute Protected Memory Usage
         Total                                          : 0 MiB
         Used                                           : 0 MiB
@@ -186,8 +186,8 @@ GPU 00000000:01:00.0
             Low                                        : 0 bank(s)
             None                                       : 0 bank(s)
     Temperature
-        GPU Current Temp                               : 30 C
-        GPU T.Limit Temp                               : 63 C
+        GPU Current Temp                               : 31 C
+        GPU T.Limit Temp                               : 62 C
         GPU Shutdown T.Limit Temp                      : -5 C
         GPU Slowdown T.Limit Temp                      : -2 C
         GPU Max Operating T.Limit Temp                 : 0 C
@@ -195,8 +195,8 @@ GPU 00000000:01:00.0
         Memory Current Temp                            : N/A
         Memory Max Operating T.Limit Temp              : N/A
     GPU Power Readings
-        Average Power Draw                             : 19.10 W
-        Instantaneous Power Draw                       : 19.86 W
+        Average Power Draw                             : 15.78 W
+        Instantaneous Power Draw                       : 15.97 W
         Current Power Limit                            : 575.00 W
         Requested Power Limit                          : 575.00 W
         Default Power Limit                            : 600.00 W
@@ -298,8 +298,8 @@ GPU 00000000:21:00.0
         ECC Object                                     : 7.16
         Power Management Object                        : N/A
     Inforom BBX Object Flush
-        Latest Timestamp                               : 2026/09/07 20:06:52.409
-        Latest Duration                                : 68770 us
+        Latest Timestamp                               : 2026/09/09 00:24:38.190
+        Latest Duration                                : 26003 us
     GPU Operation Mode
         Current                                        : N/A
         Pending                                        : N/A
@@ -336,8 +336,8 @@ GPU 00000000:21:00.0
             Firmware                                   : N/A
         Replays Since Reset                            : 155
         Replay Number Rollovers                        : 0
-        Tx Throughput                                  : 2265 KB/s
-        Rx Throughput                                  : 1672 KB/s
+        Tx Throughput                                  : 2425 KB/s
+        Rx Throughput                                  : 1534 KB/s
         Atomic Caps Outbound                           : N/A
         Atomic Caps Inbound                            : FETCHADD_32 FETCHADD_64 SWAP_32 SWAP_64 CAS_32 CAS_64 
     Fan Speed                                          : 30 %
@@ -362,12 +362,12 @@ GPU 00000000:21:00.0
     FB Memory Usage
         Total                                          : 97887 MiB
         Reserved                                       : 640 MiB
-        Used                                           : 4 MiB
-        Free                                           : 97244 MiB
+        Used                                           : 2 MiB
+        Free                                           : 97247 MiB
     BAR1 Memory Usage
         Total                                          : 256 MiB
-        Used                                           : 3 MiB
-        Free                                           : 253 MiB
+        Used                                           : 2 MiB
+        Free                                           : 254 MiB
     Conf Compute Protected Memory Usage
         Total                                          : 0 MiB
         Used                                           : 0 MiB
@@ -435,8 +435,8 @@ GPU 00000000:21:00.0
             Low                                        : 0 bank(s)
             None                                       : 0 bank(s)
     Temperature
-        GPU Current Temp                               : 29 C
-        GPU T.Limit Temp                               : 64 C
+        GPU Current Temp                               : 30 C
+        GPU T.Limit Temp                               : 63 C
         GPU Shutdown T.Limit Temp                      : -5 C
         GPU Slowdown T.Limit Temp                      : -2 C
         GPU Max Operating T.Limit Temp                 : 0 C
@@ -444,8 +444,8 @@ GPU 00000000:21:00.0
         Memory Current Temp                            : N/A
         Memory Max Operating T.Limit Temp              : N/A
     GPU Power Readings
-        Average Power Draw                             : 14.75 W
-        Instantaneous Power Draw                       : 16.58 W
+        Average Power Draw                             : 7.58 W
+        Instantaneous Power Draw                       : 7.58 W
         Current Power Limit                            : 575.00 W
         Requested Power Limit                          : 575.00 W
         Default Power Limit                            : 600.00 W
@@ -470,7 +470,7 @@ GPU 00000000:21:00.0
         Graphics                                       : 1177 MHz
         SM                                             : 1177 MHz
         Memory                                         : 405 MHz
-        Video                                          : 600 MHz
+        Video                                          : 592 MHz
     Applications Clocks
         Graphics                                       : 2617 MHz
         Memory                                         : 14001 MHz
@@ -547,8 +547,8 @@ GPU 00000000:41:00.0
         ECC Object                                     : 7.16
         Power Management Object                        : N/A
     Inforom BBX Object Flush
-        Latest Timestamp                               : 2026/09/07 20:07:00.910
-        Latest Duration                                : 66539 us
+        Latest Timestamp                               : 2026/09/09 00:24:46.975
+        Latest Duration                                : 25045 us
     GPU Operation Mode
         Current                                        : N/A
         Pending                                        : N/A
@@ -585,8 +585,8 @@ GPU 00000000:41:00.0
             Firmware                                   : N/A
         Replays Since Reset                            : 3
         Replay Number Rollovers                        : 0
-        Tx Throughput                                  : 2213 KB/s
-        Rx Throughput                                  : 1507 KB/s
+        Tx Throughput                                  : 2319 KB/s
+        Rx Throughput                                  : 1597 KB/s
         Atomic Caps Outbound                           : N/A
         Atomic Caps Inbound                            : FETCHADD_32 FETCHADD_64 SWAP_32 SWAP_64 CAS_32 CAS_64 
     Fan Speed                                          : 30 %
@@ -611,12 +611,12 @@ GPU 00000000:41:00.0
     FB Memory Usage
         Total                                          : 97887 MiB
         Reserved                                       : 640 MiB
-        Used                                           : 4 MiB
-        Free                                           : 97244 MiB
+        Used                                           : 2 MiB
+        Free                                           : 97247 MiB
     BAR1 Memory Usage
         Total                                          : 256 MiB
-        Used                                           : 3 MiB
-        Free                                           : 253 MiB
+        Used                                           : 2 MiB
+        Free                                           : 254 MiB
     Conf Compute Protected Memory Usage
         Total                                          : 0 MiB
         Used                                           : 0 MiB
@@ -684,8 +684,8 @@ GPU 00000000:41:00.0
             Low                                        : 0 bank(s)
             None                                       : 0 bank(s)
     Temperature
-        GPU Current Temp                               : 30 C
-        GPU T.Limit Temp                               : 63 C
+        GPU Current Temp                               : 31 C
+        GPU T.Limit Temp                               : 62 C
         GPU Shutdown T.Limit Temp                      : -5 C
         GPU Slowdown T.Limit Temp                      : -2 C
         GPU Max Operating T.Limit Temp                 : 0 C
@@ -693,8 +693,8 @@ GPU 00000000:41:00.0
         Memory Current Temp                            : N/A
         Memory Max Operating T.Limit Temp              : N/A
     GPU Power Readings
-        Average Power Draw                             : 19.84 W
-        Instantaneous Power Draw                       : 20.94 W
+        Average Power Draw                             : 15.73 W
+        Instantaneous Power Draw                       : 15.73 W
         Current Power Limit                            : 575.00 W
         Requested Power Limit                          : 575.00 W
         Default Power Limit                            : 600.00 W
@@ -796,8 +796,8 @@ GPU 00000000:C1:00.0
         ECC Object                                     : 7.16
         Power Management Object                        : N/A
     Inforom BBX Object Flush
-        Latest Timestamp                               : 2026/09/07 21:19:49.087
-        Latest Duration                                : 66464 us
+        Latest Timestamp                               : 2026/09/09 00:25:02.382
+        Latest Duration                                : 61684 us
     GPU Operation Mode
         Current                                        : N/A
         Pending                                        : N/A
@@ -834,8 +834,8 @@ GPU 00000000:C1:00.0
             Firmware                                   : N/A
         Replays Since Reset                            : 0
         Replay Number Rollovers                        : 0
-        Tx Throughput                                  : 1957 KB/s
-        Rx Throughput                                  : 1494 KB/s
+        Tx Throughput                                  : 2006 KB/s
+        Rx Throughput                                  : 1505 KB/s
         Atomic Caps Outbound                           : N/A
         Atomic Caps Inbound                            : FETCHADD_32 FETCHADD_64 SWAP_32 SWAP_64 CAS_32 CAS_64 
     Fan Speed                                          : 30 %
@@ -860,12 +860,12 @@ GPU 00000000:C1:00.0
     FB Memory Usage
         Total                                          : 97887 MiB
         Reserved                                       : 640 MiB
-        Used                                           : 4 MiB
-        Free                                           : 97244 MiB
+        Used                                           : 2 MiB
+        Free                                           : 97247 MiB
     BAR1 Memory Usage
         Total                                          : 256 MiB
-        Used                                           : 3 MiB
-        Free                                           : 253 MiB
+        Used                                           : 2 MiB
+        Free                                           : 254 MiB
     Conf Compute Protected Memory Usage
         Total                                          : 0 MiB
         Used                                           : 0 MiB
@@ -933,8 +933,8 @@ GPU 00000000:C1:00.0
             Low                                        : 0 bank(s)
             None                                       : 0 bank(s)
     Temperature
-        GPU Current Temp                               : 29 C
-        GPU T.Limit Temp                               : 64 C
+        GPU Current Temp                               : 30 C
+        GPU T.Limit Temp                               : 63 C
         GPU Shutdown T.Limit Temp                      : -5 C
         GPU Slowdown T.Limit Temp                      : -2 C
         GPU Max Operating T.Limit Temp                 : 0 C
@@ -942,8 +942,8 @@ GPU 00000000:C1:00.0
         Memory Current Temp                            : N/A
         Memory Max Operating T.Limit Temp              : N/A
     GPU Power Readings
-        Average Power Draw                             : 18.77 W
-        Instantaneous Power Draw                       : 17.95 W
+        Average Power Draw                             : 15.02 W
+        Instantaneous Power Draw                       : 15.28 W
         Current Power Limit                            : 575.00 W
         Requested Power Limit                          : 575.00 W
         Default Power Limit                            : 600.00 W

environment/gpu-smi.txt

@@ -1,6 +1,6 @@
 ### nvidia-smi
 
-Tue Sep  8 01:43:22 2026       
+Wed Sep  9 02:41:47 2026       
 +-----------------------------------------------------------------------------------------+
 | NVIDIA-SMI 580.173.02             Driver Version: 580.173.02     CUDA Version: 13.0     |
 +-----------------------------------------+------------------------+----------------------+
@@ -9,19 +9,19 @@ Tue Sep  8 01:43:22 2026
 |                                         |                        |               MIG M. |
 |=========================================+========================+======================|
 |   0  NVIDIA RTX PRO 6000 Blac...    Off |   00000000:01:00.0 Off |                  Off |
-| 30%   30C    P8             18W /  575W |       4MiB /  97887MiB |      0%      Default |
+| 30%   31C    P8             15W /  575W |       2MiB /  97887MiB |      0%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 |   1  NVIDIA RTX PRO 6000 Blac...    Off |   00000000:21:00.0 Off |                  Off |
-| 30%   29C    P8             10W /  575W |       4MiB /  97887MiB |      0%      Default |
+| 30%   30C    P8              6W /  575W |       2MiB /  97887MiB |      0%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 |   2  NVIDIA RTX PRO 6000 Blac...    Off |   00000000:41:00.0 Off |                  Off |
-| 30%   30C    P8             17W /  575W |       4MiB /  97887MiB |      0%      Default |
+| 30%   31C    P8             15W /  575W |       2MiB /  97887MiB |      0%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 |   3  NVIDIA RTX PRO 6000 Blac...    Off |   00000000:C1:00.0 Off |                  Off |
-| 30%   29C    P8             16W /  575W |       4MiB /  97887MiB |      0%      Default |
+| 30%   30C    P8             14W /  575W |       2MiB /  97887MiB |      0%      Default |
 |                                         |                        |                  N/A |
 +-----------------------------------------+------------------------+----------------------+
 

environment/host-cpu.txt

@@ -15,7 +15,7 @@ Core(s) per socket:                      24
 Socket(s):                               1
 Stepping:                                1
 Frequency boost:                         enabled
-CPU(s) scaling MHz:                      51%
+CPU(s) scaling MHz:                      44%
 CPU max MHz:                             4816.6992
 CPU min MHz:                             1500.0000
 BogoMIPS:                                8200.02

environment/host-memory.txt

@@ -1,7 +1,7 @@
 ### free -g
 
                total        used        free      shared  buff/cache   available
-Mem:             755          15         163           0         581         739
+Mem:             755          14         137           0         608         740
 Swap:              7           0           7
 
 ### exit=0

environment/host-mounts.txt

@@ -3,11 +3,11 @@
 Filesystem                                  Size  Used Avail Use% Mounted on
 tmpfs                                        76G  5.5M   76G   1% /run
 efivarfs                                    256K  125K  127K  50% /sys/firmware/efi/efivars
-/dev/nvme0n1p2                              3.6T  3.1T  354G  90% /
+/dev/nvme0n1p2                              3.6T  3.1T  348G  91% /
 tmpfs                                       378G     0  378G   0% /dev/shm
 tmpfs                                       5.0M     0  5.0M   0% /run/lock
 /dev/nvme0n1p1                              1.1G  6.2M  1.1G   1% /boot/efi
-/dev/nvme1n1p1                              3.6T  1.8T  1.7T  52% /media/fmodels2
+/dev/nvme1n1p1                              3.6T  1.8T  1.7T  53% /media/fmodels2
 tmpfs                                        76G   20K   76G   1% /run/user/1000
 10.9.99.242:/mnt/GANAS02-T02S01/Ai-Storage   34T   21T   13T  62% /media/netmodels
 

environment/models/gemma-4-26B-A4B-it-NVFP4.txt

@@ -1,3 +1,528 @@
-### /media/fmodels2/cklaus/gemma-4-26B-A4B-it-NVFP4
+### /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4
 
-not a directory
+## file listing (name size)
+chat_template.jinja 18924
+config.json 22602
+generation_config.json 204
+.gitattributes 1570
+inspect.json 1537
+model.safetensors 16909852932
+processor_config.json 1689
+README.md 28303
+recipe.yaml 723
+tokenizer_config.json 2718
+tokenizer.json 32169726
+
+## listing sha256
+d491d0963e563a769c0cbd47861cfd5eafbe9b211a942cdb6a3fb6364e1a8bd4  -
+
+## total bytes / file count
+16942103646	/media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4
+34
+
+## config and tokenizer sha256
+b4258144b95d4b0a3b20ee9b5d83f60863ad627f89599ae11c1a5f75906f07b8  /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4/config.json
+e3527dfff6c0cdc69c9118b000165b0f2afc48665e23d6d9f61632ee44157b6f  /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4/generation_config.json
+04e06452d54505bd60fd2949cf3dcbe0e9f1dece1d025117796995494ad12ec6  /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4/tokenizer_config.json
+adbaa8175acf7609b4359724f40eff359ec4fac1a8647eeb99d4422be708e1cf  /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4/tokenizer.json
+845f1ee48e39fc942fe190da9df6a1c5db229e17a96ea08966ad1c9274e73d1b  /media/fmodels2/unsloth/gemma-4-26B-A4B-it-NVFP4/chat_template.jinja
+
+## config.json
+{
+  "architectures": [
+    "Gemma4ForConditionalGeneration"
+  ],
+  "audio_config": null,
+  "audio_token_id": 258881,
+  "boa_token_id": 256000,
+  "boi_token_id": 255999,
+  "dtype": "bfloat16",
+  "eoa_token_id": 258883,
+  "eoa_token_index": 258883,
+  "eoi_token_id": 258882,
+  "eos_token_id": 106,
+  "head_dim": 256,
+  "image_token_id": 258880,
+  "initializer_range": 0.02,
+  "model_type": "gemma4",
+  "num_attention_heads": 16,
+  "num_key_value_heads": 8,
+  "pad_token_id": 0,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "float-quantized",
+        "input_activations": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": true,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": null,
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "token",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        },
+        "output_activations": null,
+        "targets": [
+          "re:.*self_attn\\.(q|k|v|o)_proj$"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": null,
+          "num_bits": 8,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": null,
+          "strategy": "channel",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      },
+      "group_1": {
+        "format": "nvfp4-pack-quantized",
+        "input_activations": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": "local",
+          "group_size": 16,
+          "num_bits": 4,
+          "observer": "static_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": "torch.float8_e4m3fn",
+          "strategy": "tensor_group",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        },
+        "output_activations": null,
+        "targets": [
+          "re:.*\\.experts\\.\\d+\\.(gate|up|down)_proj$",
+          "re:.*language_model.*\\.mlp\\.(gate|up|down)_proj$"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 16,
+          "num_bits": 4,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "scale_dtype": "torch.float8_e4m3fn",
+          "strategy": "tensor_group",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "mixed-precision",
+    "global_compression_ratio": null,
+    "ignore": [
+      "model.vision_tower.patch_embedder.input_proj",
+      "model.vision_tower.encoder.layers.0.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.0.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.0.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.0.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.0.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.0.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.0.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.1.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.1.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.1.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.1.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.1.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.1.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.1.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.2.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.2.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.2.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.2.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.2.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.2.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.2.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.3.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.3.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.3.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.3.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.3.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.3.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.3.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.4.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.4.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.4.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.4.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.4.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.4.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.4.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.5.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.5.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.5.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.5.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.5.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.5.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.5.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.6.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.6.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.6.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.6.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.6.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.6.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.6.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.7.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.7.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.7.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.7.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.7.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.7.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.7.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.8.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.8.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.8.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.8.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.8.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.8.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.8.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.9.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.9.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.9.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.9.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.9.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.9.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.9.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.10.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.10.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.10.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.10.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.10.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.10.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.10.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.11.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.11.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.11.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.11.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.11.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.11.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.11.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.12.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.12.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.12.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.12.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.12.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.12.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.12.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.13.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.13.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.13.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.13.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.13.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.13.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.13.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.14.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.14.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.14.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.14.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.14.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.14.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.14.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.15.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.15.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.15.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.15.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.15.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.15.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.15.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.16.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.16.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.16.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.16.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.16.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.16.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.16.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.17.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.17.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.17.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.17.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.17.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.17.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.17.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.18.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.18.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.18.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.18.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.18.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.18.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.18.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.19.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.19.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.19.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.19.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.19.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.19.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.19.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.20.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.20.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.20.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.20.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.20.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.20.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.20.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.21.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.21.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.21.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.21.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.21.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.21.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.21.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.22.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.22.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.22.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.22.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.22.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.22.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.22.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.23.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.23.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.23.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.23.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.23.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.23.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.23.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.24.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.24.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.24.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.24.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.24.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.24.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.24.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.25.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.25.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.25.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.25.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.25.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.25.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.25.mlp.down_proj.linear",
+      "model.vision_tower.encoder.layers.26.self_attn.q_proj.linear",
+      "model.vision_tower.encoder.layers.26.self_attn.k_proj.linear",
+      "model.vision_tower.encoder.layers.26.self_attn.v_proj.linear",
+      "model.vision_tower.encoder.layers.26.self_attn.o_proj.linear",
+      "model.vision_tower.encoder.layers.26.mlp.gate_proj.linear",
+      "model.vision_tower.encoder.layers.26.mlp.up_proj.linear",
+      "model.vision_tower.encoder.layers.26.mlp.down_proj.linear",
+      "model.language_model.layers.0.router",
+      "model.language_model.layers.0.router.proj",
+      "model.language_model.layers.1.router",
+      "model.language_model.layers.1.router.proj",
+      "model.language_model.layers.2.router",
+      "model.language_model.layers.2.router.proj",
+      "model.language_model.layers.3.router",
+      "model.language_model.layers.3.router.proj",
+      "model.language_model.layers.4.router",
+      "model.language_model.layers.4.router.proj",
+      "model.language_model.layers.5.router",
+      "model.language_model.layers.5.router.proj",
+      "model.language_model.layers.6.router",
+      "model.language_model.layers.6.router.proj",
+      "model.language_model.layers.7.router",
+      "model.language_model.layers.7.router.proj",
+      "model.language_model.layers.8.router",
+      "model.language_model.layers.8.router.proj",
+      "model.language_model.layers.9.router",
+      "model.language_model.layers.9.router.proj",
+      "model.language_model.layers.10.router",
+      "model.language_model.layers.10.router.proj",
+      "model.language_model.layers.11.router",
+      "model.language_model.layers.11.router.proj",
+      "model.language_model.layers.12.router",
+      "model.language_model.layers.12.router.proj",
+      "model.language_model.layers.13.router",
+      "model.language_model.layers.13.router.proj",
+      "model.language_model.layers.14.router",
+      "model.language_model.layers.14.router.proj",
+      "model.language_model.layers.15.router",
+      "model.language_model.layers.15.router.proj",
+      "model.language_model.layers.16.router",
+      "model.language_model.layers.16.router.proj",
+      "model.language_model.layers.17.router",
+      "model.language_model.layers.17.router.proj",
+      "model.language_model.layers.18.router",
+      "model.language_model.layers.18.router.proj",
+      "model.language_model.layers.19.router",
+      "model.language_model.layers.19.router.proj",
+      "model.language_model.layers.20.router",
+      "model.language_model.layers.20.router.proj",
+      "model.language_model.layers.21.router",
+      "model.language_model.layers.21.router.proj",
+      "model.language_model.layers.22.router",
+      "model.language_model.layers.22.router.proj",
+      "model.language_model.layers.23.router",
+      "model.language_model.layers.23.router.proj",
+      "model.language_model.layers.24.router",
+      "model.language_model.layers.24.router.proj",
+      "model.language_model.layers.25.router",
+      "model.language_model.layers.25.router.proj",
+      "model.language_model.layers.26.router",
+      "model.language_model.layers.26.router.proj",
+      "model.language_model.layers.27.router",
+      "model.language_model.layers.27.router.proj",
+      "model.language_model.layers.28.router",
+      "model.language_model.layers.28.router.proj",
+      "model.language_model.layers.29.router",
+      "model.language_model.layers.29.router.proj",
+      "model.embed_vision.embedding_projection",
+      "lm_head"
+    ],
+    "kv_cache_scheme": {
+      "actorder": null,
+      "block_structure": null,
+      "dynamic": false,
+      "group_size": null,
+      "num_bits": 8,
+      "observer": "static_minmax",
+      "observer_kwargs": {},
+      "scale_dtype": null,
+      "strategy": "tensor",
+      "symmetric": true,
+      "type": "float",
+      "zp_dtype": null
+    },
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.17.2.a20260707"
+  },
+  "text_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "attention_k_eq_v": true,
+    "bos_token_id": 2,
+    "dtype": "bfloat16",
+    "enable_moe_block": true,
+    "eos_token_id": 1,
+    "final_logit_softcapping": 30.0,
+    "global_head_dim": 512,
+    "head_dim": 256,
+    "hidden_activation": "gelu_pytorch_tanh",
+    "hidden_size": 2816,
+    "hidden_size_per_layer_input": 0,
+    "initializer_range": 0.02,
+    "intermediate_size": 2112,
+    "layer_types": [
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "sliding_attention",
+      "full_attention"
+    ],
+    "max_position_embeddings": 262144,
+    "model_type": "gemma4_text",
+    "moe_intermediate_size": 704,
+    "num_attention_heads": 16,
+    "num_experts": 128,
+    "num_global_key_value_heads": 2,
+    "num_hidden_layers": 30,
+    "num_key_value_heads": 8,
+    "num_kv_shared_layers": 0,
+    "pad_token_id": 0,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "full_attention": {
+        "partial_rotary_factor": 0.25,
+        "rope_theta": 1000000.0,
+        "rope_type": "proportional"
+      },
+      "sliding_attention": {
+        "rope_theta": 10000.0,
+        "rope_type": "default"
+      }
+    },
+    "sliding_window": 1024,
+    "tie_word_embeddings": true,
+    "top_k_experts": 8,
+    "use_bidirectional_attention": "vision",
+    "use_cache": true,
+    "use_double_wide_mlp": false,
+    "vocab_size": 262144,
+    "vocab_size_per_layer_input": 262144
+  },
+  "tie_word_embeddings": true,
+  "transformers_version": "5.13.0",
+  "unsloth_fixed": true,
+  "video_token_id": 258884,
+  "vision_config": {
+    "_name_or_path": "",
+    "architectures": null,
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "chunk_size_feed_forward": 0,
+    "default_output_length": 280,
+    "dtype": "bfloat16",
+    "global_head_dim": 72,
+    "head_dim": 72,
+    "hidden_activation": "gelu_pytorch_tanh",
+    "hidden_size": 1152,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 4304,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "max_position_embeddings": 131072,
+    "model_type": "gemma4_vision",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 27,
+    "num_key_value_heads": 16,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "patch_size": 16,
+    "pooling_kernel_size": 3,
+    "position_embedding_size": 10240,
+    "problem_type": null,
+    "return_dict": true,
+    "rms_norm_eps": 1e-06,
+    "rope_parameters": {
+      "rope_theta": 100.0,
+      "rope_type": "default"
+    },
+    "standardize": true,
+    "use_clipped_linears": false
+  },
+  "vision_soft_tokens_per_image": 280
+}

environment/models/gemma-4-26B-A4B-it-W4A16.txt

@@ -1,432 +1,3 @@
 ### /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16
 
-## file listing (name size)
-chat_template.jinja 12045
-config.json 19270
-generation_config.json 203
-.gitattributes 1570
-processor_config.json 1689
-README.md 1284
-recipe.yaml 934
-tokenizer_config.json 2712
-tokenizer.json 32169626
-
-## listing sha256
-0224dd5c4bbfe0b881c212e109312a29dcc818128a4f4780c17f79ecb798c9a6  -
-
-## total bytes / file count
-7945391992	/media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16
-32
-
-## config and tokenizer sha256
-773a83885006e4b425036ea214b15889fff5a04482063c4a2d9a3074f72df9bb  /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16/config.json
-5439541d3bf0ba9ade7c1122dc14703f85f32a520150351b344b14f48c574cc2  /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16/generation_config.json
-db33eb89116392fac778165935900c87495a0b357d43fea009c77be7d96134d0  /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16/tokenizer_config.json
-cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f  /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16/tokenizer.json
-2dfbfc7d538912f4ea11d29d85b4e25d7bc26386e53f57529f1d707c28b5828c  /media/fmodels2/dhruvil237/gemma-4-26B-A4B-it-W4A16/chat_template.jinja
-
-## config.json
-{
-  "architectures": [
-    "Gemma4ForConditionalGeneration"
-  ],
-  "audio_config": null,
-  "audio_token_id": 258881,
-  "boa_token_id": 256000,
-  "boi_token_id": 255999,
-  "dtype": "bfloat16",
-  "eoa_token_id": 258883,
-  "eoa_token_index": 258883,
-  "eoi_token_id": 258882,
-  "eos_token_id": [
-    1,
-    106
-  ],
-  "image_token_id": 258880,
-  "initializer_range": 0.02,
-  "model_type": "gemma4",
-  "quantization_config": {
-    "config_groups": {
-      "group_0": {
-        "format": "pack-quantized",
-        "input_activations": null,
-        "output_activations": null,
-        "targets": [
-          "Linear"
-        ],
-        "weights": {
-          "actorder": "static",
-          "block_structure": null,
-          "dynamic": false,
-          "group_size": 64,
-          "num_bits": 4,
-          "observer": "memoryless_minmax",
-          "observer_kwargs": {},
-          "scale_dtype": null,
-          "strategy": "group",
-          "symmetric": true,
-          "type": "int",
-          "zp_dtype": null
-        }
-      }
-    },
-    "format": "pack-quantized",
-    "global_compression_ratio": null,
-    "ignore": [
-      "model.language_model.layers.0.router.proj",
-      "model.language_model.layers.1.router.proj",
-      "model.language_model.layers.2.router.proj",
-      "model.language_model.layers.3.router.proj",
-      "model.language_model.layers.4.router.proj",
-      "model.language_model.layers.5.router.proj",
-      "model.language_model.layers.6.router.proj",
-      "model.language_model.layers.7.router.proj",
-      "model.language_model.layers.8.router.proj",
-      "model.language_model.layers.9.router.proj",
-      "model.language_model.layers.10.router.proj",
-      "model.language_model.layers.11.router.proj",
-      "model.language_model.layers.12.router.proj",
-      "model.language_model.layers.13.router.proj",
-      "model.language_model.layers.14.router.proj",
-      "model.language_model.layers.15.router.proj",
-      "model.language_model.layers.16.router.proj",
-      "model.language_model.layers.17.router.proj",
-      "model.language_model.layers.18.router.proj",
-      "model.language_model.layers.19.router.proj",
-      "model.language_model.layers.20.router.proj",
-      "model.language_model.layers.21.router.proj",
-      "model.language_model.layers.22.router.proj",
-      "model.language_model.layers.23.router.proj",
-      "model.language_model.layers.24.router.proj",
-      "model.language_model.layers.25.router.proj",
-      "model.language_model.layers.26.router.proj",
-      "model.language_model.layers.27.router.proj",
-      "model.language_model.layers.28.router.proj",
-      "model.language_model.layers.29.router.proj",
-      "model.vision_tower.patch_embedder.input_proj",
-      "model.vision_tower.encoder.layers.0.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.0.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.0.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.0.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.0.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.0.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.0.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.1.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.1.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.1.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.1.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.1.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.1.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.1.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.2.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.2.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.2.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.2.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.2.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.2.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.2.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.3.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.3.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.3.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.3.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.3.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.3.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.3.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.4.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.4.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.4.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.4.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.4.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.4.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.4.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.5.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.5.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.5.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.5.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.5.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.5.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.5.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.6.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.6.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.6.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.6.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.6.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.6.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.6.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.7.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.7.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.7.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.7.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.7.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.7.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.7.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.8.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.8.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.8.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.8.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.8.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.8.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.8.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.9.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.9.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.9.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.9.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.9.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.9.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.9.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.10.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.10.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.10.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.10.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.10.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.10.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.10.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.11.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.11.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.11.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.11.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.11.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.11.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.11.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.12.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.12.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.12.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.12.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.12.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.12.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.12.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.13.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.13.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.13.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.13.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.13.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.13.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.13.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.14.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.14.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.14.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.14.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.14.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.14.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.14.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.15.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.15.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.15.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.15.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.15.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.15.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.15.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.16.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.16.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.16.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.16.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.16.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.16.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.16.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.17.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.17.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.17.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.17.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.17.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.17.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.17.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.18.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.18.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.18.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.18.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.18.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.18.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.18.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.19.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.19.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.19.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.19.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.19.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.19.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.19.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.20.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.20.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.20.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.20.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.20.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.20.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.20.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.21.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.21.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.21.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.21.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.21.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.21.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.21.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.22.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.22.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.22.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.22.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.22.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.22.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.22.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.23.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.23.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.23.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.23.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.23.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.23.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.23.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.24.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.24.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.24.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.24.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.24.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.24.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.24.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.25.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.25.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.25.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.25.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.25.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.25.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.25.mlp.down_proj.linear",
-      "model.vision_tower.encoder.layers.26.self_attn.q_proj.linear",
-      "model.vision_tower.encoder.layers.26.self_attn.k_proj.linear",
-      "model.vision_tower.encoder.layers.26.self_attn.v_proj.linear",
-      "model.vision_tower.encoder.layers.26.self_attn.o_proj.linear",
-      "model.vision_tower.encoder.layers.26.mlp.gate_proj.linear",
-      "model.vision_tower.encoder.layers.26.mlp.up_proj.linear",
-      "model.vision_tower.encoder.layers.26.mlp.down_proj.linear",
-      "model.embed_vision.embedding_projection",
-      "lm_head"
-    ],
-    "kv_cache_scheme": null,
-    "quant_method": "compressed-tensors",
-    "quantization_status": "compressed",
-    "sparsity_config": {},
-    "transform_config": {},
-    "version": "0.15.0"
-  },
-  "text_config": {
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "attention_k_eq_v": true,
-    "bos_token_id": 2,
-    "dtype": "bfloat16",
-    "enable_moe_block": true,
-    "eos_token_id": 1,
-    "final_logit_softcapping": 30.0,
-    "global_head_dim": 512,
-    "head_dim": 256,
-    "hidden_activation": "gelu_pytorch_tanh",
-    "hidden_size": 2816,
-    "hidden_size_per_layer_input": 0,
-    "initializer_range": 0.02,
-    "intermediate_size": 2112,
-    "layer_types": [
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "full_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "full_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "full_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "full_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "sliding_attention",
-      "full_attention"
-    ],
-    "max_position_embeddings": 262144,
-    "model_type": "gemma4_text",
-    "moe_intermediate_size": 704,
-    "num_attention_heads": 16,
-    "num_experts": 128,
-    "num_global_key_value_heads": 2,
-    "num_hidden_layers": 30,
-    "num_key_value_heads": 8,
-    "num_kv_shared_layers": 0,
-    "pad_token_id": 0,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "full_attention": {
-        "partial_rotary_factor": 0.25,
-        "rope_theta": 1000000.0,
-        "rope_type": "proportional"
-      },
-      "sliding_attention": {
-        "rope_theta": 10000.0,
-        "rope_type": "default"
-      }
-    },
-    "sliding_window": 1024,
-    "tie_word_embeddings": true,
-    "top_k_experts": 8,
-    "use_bidirectional_attention": "vision",
-    "use_cache": true,
-    "use_double_wide_mlp": false,
-    "vocab_size": 262144,
-    "vocab_size_per_layer_input": 262144
-  },
-  "tie_word_embeddings": true,
-  "transformers_version": "5.5.0",
-  "video_token_id": 258884,
-  "vision_config": {
-    "_name_or_path": "",
-    "architectures": null,
-    "attention_bias": false,
-    "attention_dropout": 0.0,
-    "chunk_size_feed_forward": 0,
-    "default_output_length": 280,
-    "dtype": "bfloat16",
-    "global_head_dim": 72,
-    "head_dim": 72,
-    "hidden_activation": "gelu_pytorch_tanh",
-    "hidden_size": 1152,
-    "id2label": {
-      "0": "LABEL_0",
-      "1": "LABEL_1"
-    },
-    "initializer_range": 0.02,
-    "intermediate_size": 4304,
-    "is_encoder_decoder": false,
-    "label2id": {
-      "LABEL_0": 0,
-      "LABEL_1": 1
-    },
-    "max_position_embeddings": 131072,
-    "model_type": "gemma4_vision",
-    "num_attention_heads": 16,
-    "num_hidden_layers": 27,
-    "num_key_value_heads": 16,
-    "output_attentions": false,
-    "output_hidden_states": false,
-    "patch_size": 16,
-    "pooling_kernel_size": 3,
-    "position_embedding_size": 10240,
-    "problem_type": null,
-    "return_dict": true,
-    "rms_norm_eps": 1e-06,
-    "rope_parameters": {
-      "rope_theta": 100.0,
-      "rope_type": "default"
-    },
-    "standardize": true,
-    "use_clipped_linears": false
-  },
-  "vision_soft_tokens_per_image": 280
-}
+not a directory

environment/pip-freeze.txt

@@ -206,7 +206,7 @@ urllib3==2.7.0
 uv==0.12.7
 uvicorn==0.52.4
 uvloop==0.22.1
--e git+https://github.com/phaelon74/vllm.git@c798225a7f197994ed502068fc6ac1ba8a9706f3#egg=vllm
+-e git+https://github.com/phaelon74/vllm.git@a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61#egg=vllm
 watchfiles==1.2.0
 websockets==17.1
 xgrammar==0.2.3

environment/repo-git-describe.txt

@@ -1,5 +1,5 @@
 ### git -C /home/phaedawg/kld-bestofbreed/vllm describe --always --dirty --tags
 
-c798225a7
+a2ae11e29
 
 ### exit=0

environment/repo-git-log.txt

@@ -1,9 +1,9 @@
 ### git -C /home/phaedawg/kld-bestofbreed/vllm log -n 5 --format=%H %ad %s --date=iso
 
-c798225a7f197994ed502068fc6ac1ba8a9706f3 2026-09-07 20:40:31 -0500 Re-Order Campaign to be more complete.
-fccd81de7ce0e82e7afc23f413cea9fc7908dc5f 2026-09-07 20:16:00 -0500 Treat a report as stale when its bound capture manifest changed
-cbee5c2eb9921e66833f0a3034f314b9915c9250 2026-09-06 13:33:24 -0500 Document W4A4 NVFP4 emulation scoring and probe-based certification
-ae45172d91537370e6d84d6accb0da6a76339bac 2026-09-06 13:23:58 -0500 Pin the W4A4 emulation MoE backend on the student engine only
-575bb21679f76b899018528c9b17159ebf276ab0 2026-09-06 09:40:29 -0500 closed on unprobed MoE backends and score W4A4 NVFP4 faithfully
+a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61 2026-09-08 21:37:15 -0500 Updating based on finds related to Model Size.
+d20662a0ee1fed5cfe3ed6a5834380c9751b9a3a 2026-09-08 21:16:24 -0500 fix alignments on quantization.
+6596f9995824a1bf97fc44efd125de073587bf67 2026-09-08 20:42:56 -0500 Fix Addendum PR54444
+4514a12d0001bc092161196adb8efcc3768821b9 2026-09-08 20:32:03 -0500 Fix the Converged Alignment.
+fda1877acf59281fd62d3745b253cae889c93e06 2026-09-08 13:40:48 -0500 Update NVFP4
 
 ### exit=0

environment/runtime.json

@@ -29,7 +29,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -84,11 +84,6 @@
     "CUDA_HOME": "/usr/local/cuda-13.0",
     "HF_TOKEN": "<redacted>",
     "NCCL_DETERMINISTIC": "1",
-    "PYTORCH_NVML_BASED_CUDA_CHECK": "1",
-    "TORCHINDUCTOR_CACHE_DIR": "/tmp/torchinductor_phaedawg",
-    "TORCHINDUCTOR_COMPILE_THREADS": "1",
-    "TRITON_CACHE_AUTOTUNING": "1",
-    "TRITON_PTXAS_BLACKWELL_PATH": "/usr/local/cuda-13.0/bin/ptxas",
     "VLLM_BATCH_INVARIANT": "1",
     "VLLM_MARLIN_USE_ATOMIC_ADD": "0",
     "VLLM_MOE_USE_DEEP_GEMM": "0"
@@ -124,7 +119,7 @@
     "path": "/home/phaedawg/kld-bestofbreed/vllm/vllm/__init__.py",
     "version": "0.1.dev20446+gb2bc9171d"
   },
-  "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+  "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
   "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
   "vllm_tree_dirty": false
 }

environment/summary.md

@@ -1,11 +1,11 @@
 # KLD environment report
 
-- captured: 2026-09-08T01:43:27+00:00
+- captured: 2026-09-09T02:41:52+00:00
 - host: d011sd01
 - kernel: 6.8.0-137-generic
 - python: /home/phaedawg/kld-bestofbreed/kldbestofbreed/bin/python
 - repo: /home/phaedawg/kld-bestofbreed/vllm
-- commit: c798225a7f197994ed502068fc6ac1ba8a9706f3
+- commit: a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61
 - branch: feature/glm53-kld-determinism
 - driver: 580.173.02
 - gpus:       4 NVIDIA RTX PRO 6000 Blackwell Workstation Edition;

gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-attention-int4_g32_sym.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a",
+  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -65,6 +65,7 @@
     }
   ],
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "max_kld": 41.926361083984375,
   "mean_kld": 1.5059427601219246,
   "mean_kld_reverse": 1.6012680526979337,
@@ -72,6 +73,11 @@
   "model_runner_v2": false,
   "num_positions": 1572096,
   "num_rows": 768,
+  "nvfp4_dense_scale_inspection": {
+    "layers_filled": 0,
+    "layers_scanned": 0,
+    "workers": 1
+  },
   "p90_kld": 4.438774824142456,
   "p99_kld": 12.152195024490366,
   "per_context": [
@@ -8524,6 +8530,7 @@
       "top1_agreement": 0.7381533952125061
     }
   ],
+  "quantization_substitutions": [],
   "reference_weights_sha256": "201ebceb92c5b166cc028c3008607ba28c478c7699b1d382e7be7ede81f2ab75",
   "routing": {
     "buckets": [
@@ -8634,7 +8641,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8737,9 +8744,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.475490545388311,
-    "score_forward": 159.24197272164747,
-    "student_load": 22.502437429036945
+    "routing_compare": 8.53352264361456,
+    "score_forward": 159.3522308692336,
+    "student_load": 22.87009431095794
   },
   "top1_agreement": 0.6349357800032568,
   "topk_agreement": {

gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-int4_g32_sym.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a",
+  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -65,6 +65,7 @@
     }
   ],
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "max_kld": 42.107303619384766,
   "mean_kld": 0.6908063985435383,
   "mean_kld_reverse": 0.6849806829035872,
@@ -72,6 +73,11 @@
   "model_runner_v2": false,
   "num_positions": 1572096,
   "num_rows": 768,
+  "nvfp4_dense_scale_inspection": {
+    "layers_filled": 0,
+    "layers_scanned": 0,
+    "workers": 1
+  },
   "p90_kld": 1.8279701471328735,
   "p99_kld": 8.708542585372927,
   "per_context": [
@@ -8524,6 +8530,7 @@
       "top1_agreement": 0.8402540302882266
     }
   ],
+  "quantization_substitutions": [],
   "reference_weights_sha256": "201ebceb92c5b166cc028c3008607ba28c478c7699b1d382e7be7ede81f2ab75",
   "routing": {
     "buckets": [
@@ -8634,7 +8641,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8737,9 +8744,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.391019876580685,
-    "score_forward": 163.9489350831136,
-    "student_load": 22.62229167902842
+    "routing_compare": 8.304176480975002,
+    "score_forward": 162.99136068299413,
+    "student_load": 22.779514422174543
   },
   "top1_agreement": 0.7561681983797427,
   "topk_agreement": {

gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-mxfp8.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a",
+  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -65,6 +65,7 @@
     }
   ],
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "max_kld": 37.762569427490234,
   "mean_kld": 0.467187183179466,
   "mean_kld_reverse": 0.4616639767880551,
@@ -72,6 +73,11 @@
   "model_runner_v2": false,
   "num_positions": 1572096,
   "num_rows": 768,
+  "nvfp4_dense_scale_inspection": {
+    "layers_filled": 0,
+    "layers_scanned": 0,
+    "workers": 1
+  },
   "p90_kld": 1.1222012042999268,
   "p99_kld": 6.9373567581176765,
   "per_context": [
@@ -8524,6 +8530,7 @@
       "top1_agreement": 0.8622374206155349
     }
   ],
+  "quantization_substitutions": [],
   "reference_weights_sha256": "201ebceb92c5b166cc028c3008607ba28c478c7699b1d382e7be7ede81f2ab75",
   "routing": {
     "buckets": [
@@ -8634,7 +8641,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8737,9 +8744,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.068514080252498,
-    "score_forward": 166.34754822216928,
-    "student_load": 22.372262749820948
+    "routing_compare": 8.242558104451746,
+    "score_forward": 166.26376615883783,
+    "student_load": 22.620598616078496
   },
   "top1_agreement": 0.8053471289285132,
   "topk_agreement": {

gemma-4-26B-A4B-it-AWQ-4bit/attribution/gemma-4-26B-A4B-it-qdq-experts-nvfp4.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a",
+  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
   "confidence_buckets": [
     {
       "frac": 0.07217434558703795,
@@ -65,6 +65,7 @@
     }
   ],
   "kld_vocab_size": 262144,
+  "kv_cache_dtype": "bfloat16",
   "max_kld": 37.5721549987793,
   "mean_kld": 0.6665224749770983,
   "mean_kld_reverse": 0.6656885178338737,
@@ -72,6 +73,11 @@
   "model_runner_v2": false,
   "num_positions": 1572096,
   "num_rows": 768,
+  "nvfp4_dense_scale_inspection": {
+    "layers_filled": 0,
+    "layers_scanned": 0,
+    "workers": 1
+  },
   "p90_kld": 1.7603277564048767,
   "p99_kld": 8.449960660934465,
   "per_context": [
@@ -8524,6 +8530,7 @@
       "top1_agreement": 0.8612603810454323
     }
   ],
+  "quantization_substitutions": [],
   "reference_weights_sha256": "201ebceb92c5b166cc028c3008607ba28c478c7699b1d382e7be7ede81f2ab75",
   "routing": {
     "buckets": [
@@ -8634,7 +8641,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },
@@ -8737,9 +8744,9 @@
     }
   },
   "timings": {
-    "routing_compare": 8.323112379293889,
-    "score_forward": 163.84061056841165,
-    "student_load": 22.643151295837015
+    "routing_compare": 8.273550726938993,
+    "score_forward": 163.96420371392742,
+    "student_load": 22.59668383700773
   },
   "top1_agreement": 0.7592182665689627,
   "topk_agreement": {

gemma-4-26B-A4B-it-AWQ-4bit/attribution/routing/routing-manifest.json

@@ -1,5 +1,5 @@
 {
-  "capture_manifest_sha256": "d1f03dea74e9a9b9eaee1cba6a6131584c17e866198aa9660a55f0a79b3cf47a",
+  "capture_manifest_sha256": "b5ab5ede01f46551b1932e913620d9554df5f97dd7c5e2bcca05f61db68725ef",
   "file_hashes": {
     "routing_00000.safetensors": "b15d371f91aff64ce1f741b0fa20fec537834f00753b7e789418d5fbd24ab207",
     "routing_00001.safetensors": "91d03a3baff046afb21017604984267d9abd885408acc1cbcdce2e66784d94c5",
@@ -841,7 +841,7 @@
     "platform": "Linux-6.8.0-137-generic-x86_64-with-glibc2.39",
     "python": "3.12.3",
     "torch": "2.13.0+cu132",
-    "vllm_commit": "c798225a7f197994ed502068fc6ac1ba8a9706f3",
+    "vllm_commit": "a2ae11e29e34294cfaa4d768f1a1d8d8199d1d61",
     "vllm_dirty_digest": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
     "vllm_tree_dirty": false
   },