Publish AX-Qwen3.8-27B-MLX-AXQ-MXFP4-MTP

README.md

@@ -24,13 +24,9 @@ tags:
 An **AXQuant (AXQ)** mixed-precision MLX checkpoint for Apple Silicon, converted directly from
 the BF16 source model. The language path is quantized while the multi-token-prediction (MTP) head and vision tower are preserved at BF16 in the checkpoint (or a bound sidecar when present).
 
-> **Checkpoint Tier 1 certified** on `df-macstudio-m2` (2026-08-15) at Hub commit
-> `594de6507dc9` — measured size against a matched uniform baseline, quality retention,
-> and conversion integrity. Current `main` preserves that revision's exact Safetensors
-> payloads while allowing metadata-only compatibility fixes. Tier 1 is a checkpoint claim,
-> **not** a speed claim: MTP
-> acceleration is **not certified**; no MTP speedup claim for this checkpoint.
-> See the [checkpoint Tier 1 certificate](https://github.com/defai-digital/axquant/blob/main/docs/certifications/qwen38-27b-axq-mxfp4-mtp-tier1.md) for the bound evidence and thresholds.
+> **Development evidence — not a certified AXQuant release.** This package has conversion and
+> artifact-integrity records, but it does not publish measured quality, long-context, kernel-speed,
+> or MTP-speed evidence. Do not interpret the AXQ product label as a benchmark claim.
 
 
 ## Model details
@@ -42,17 +38,17 @@ the BF16 source model. The language path is quantized while the multi-token-pred
 | Product family | `qwen3.8` |
 | Source architecture | `Qwen3_5ForConditionalGeneration` (dense); text path optimized |
 | Main-model parameters | 27.36B logical parameters |
-| Quantizer | AXQuant `1.8.1` |
+| Quantizer | AXQuant `1.9.0` |
 | Hub budget class | `MXFP4` |
 | AXQuant base precision class | `6bit` |
 | Planned storage-adjusted BPW | 5.6720 |
 | Measured main-model BPW | 4.8441 |
 | Measured total BPW, including MTP | **5.0147** |
 | Safetensors weight size | 17.41 GB |
-| Approximate complete download | 17.44 GB |
+| Approximate complete download | 17.45 GB |
 | Configured maximum context | 262,144 tokens; practical limits depend on unified memory |
 | Primary MLX runtime | MLX-LM |
-| AX Engine native execution | Native manifest included; execution still requires a runtime check |
+| AX Engine native execution | Not established; no validated native manifest is included |
 | MTP present | `True` |
 | Vision present | `True` |
 | Audio present | `False` |
@@ -85,7 +81,7 @@ python -m pip install -U huggingface_hub
 hf download AutomatosX/AX-Qwen3.8-27B-MLX-AXQ-MXFP4-MTP --local-dir ./AX-Qwen3.8-27B-MLX-AXQ-MXFP4-MTP
 ```
 
-Allow at least 17.44 GB of free disk space. Pin the resulting Hub commit in reproducible
+Allow at least 17.45 GB of free disk space. Pin the resulting Hub commit in reproducible
 deployments rather than relying indefinitely on `main`.
 
 ## Run with MLX-LM
@@ -102,29 +98,25 @@ mlx_lm.generate \
 MLX-LM compatibility covers standard **text/backbone inference**. It may ignore AXQuant runtime
 metadata and optional sidecars (`vision.safetensors`, `mtp.safetensors`); this command therefore
 does not establish MTP acceleration or vision-language quality. The artifact records MLX
-`0.32.0` and MLX-LM `0.31.3` from conversion.
+`0.32.1` and MLX-LM `0.31.3` from conversion.
 
-## Serve with AX Engine and MTP
+## AX Engine status
 
-After installing AX Engine, download the complete repository (see
-[AXQuant](https://github.com/defai-digital/axquant) for conversion, certificates, and
-model-card tooling) and serve the local directory:
-
-```bash
-ax-engine serve ./AX-Qwen3.8-27B-MLX-AXQ-MXFP4-MTP --port 31418
-```
-
-AX Engine is the authority for the AXQ runtime contract and native MTP sidecar.
-This development package does not claim runtime speedups until identical-checkpoint benchmarks are
-published. The artifact records AX Engine version `not recorded`. Native
-`model-manifest.json` status: included as `model-manifest.json`.
+This package does **not** include a validated native `model-manifest.json`, so AX Engine execution
+is not established by this release. The AX Engine fields in `axquant_runtime.json` describe the
+intended compatibility contract, not observed runtime evidence. Use the architecture-specific MLX
+runtime path above. The artifact records AX Engine version
+`7.5.7`, but version discovery alone is not a runtime check.
 
 ## Use the packaged Qwen MTP head with oMLX or MTPLX
 
-Download the complete repository to a writable local directory. In oMLX `0.6.3rc2` or newer, add
-that directory, open **Model Settings**, choose **Import MTP side-car**, and then enable
-**Lightning MTP**. The import changes only the local copy so the sidecar tensors become visible
-through the checkpoint index.
+This repository keeps the Qwen MTP head in `mtp.safetensors`; stock MLX-LM does not load that
+sidecar by itself. Download the complete repository to a writable local directory before using a
+sidecar-aware runtime.
+
+In oMLX `0.6.3rc2` or newer, add the local directory, open **Model Settings**, choose
+**Import MTP side-car**, and then enable **Lightning MTP**. The import changes only the local copy
+so the MTP tensors become visible through the checkpoint index.
 
 MTPLX can consume the packaged sidecar directly:
 
@@ -137,7 +129,8 @@ mtplx quickstart \
 ```
 
 `mtplx_runtime.json` declares the canonical `qwen3-next-mtp` execution contract. This enables
-strict runtime discovery; it does not establish oMLX or MTPLX exactness or speed certification.
+strict runtime discovery; it does not extend AXQuant quality, exactness, or speed certification to
+oMLX or MTPLX.
 
 ## Quantization layout
 
@@ -147,7 +140,7 @@ strict runtime discovery; it does not establish oMLX or MTPLX exactness or speed
 | `8bit` | 2.54B | 9.15% |
 | `bf16` | 888.07M | 3.20% |
 
-- Quantization methods: `affine, bf16`.
+- Quantization methods: `affine, bf16, mxfp4`.
 - Group sizes used by quantized assignments: `32, 64`.
 - MTP sidecar: 15 tensors, 424.70M parameters, 0.85 GB, BF16.
 - Vision sidecar: 333 tensors, 460.73M parameters, 0.92 GB, BF16.
@@ -165,23 +158,14 @@ establish MTP acceleration or vision-language quality.
 | Planning evidence | `architecture_prior` |
 | Calibration | none; the allocation is based on architecture priors |
 | Quantizer execution | 498/498 recorded module conversions succeeded; 0 fallbacks |
-| AX Engine native manifest | included as `model-manifest.json` |
+| AX Engine native manifest | not included |
 | Quality versus BF16 or uniform baselines | Not published; no quality-retention claim |
-| MTP acceptance and speed | **not certified** on `df-macstudio-m2` / AX Engine 6.16.1 ([Tier 2 record](https://github.com/defai-digital/axquant/blob/main/docs/certifications/qwen38-27b-axq-mxfp4-mtp-tier2.md)); greedy exactness failed and measured speedups were 0.0. Product default remains direct fallback. |
+| MTP acceptance and speed | not measured; no MTP speedup claim |
 | AX Engine kernel evidence | `unmeasured` |
-| Vision-language quality | Present, not certified; text Tier 1 does not imply VLM quality |
-| Speech-recognition quality | Not applicable (audio disabled for this pack) |
+| Vision-language quality | Not evaluated or claimed; vision tensors are preserved at BF16 |
+| Speech-recognition quality | Not applicable |
 | Long-context quality | 262,144-token capacity is config metadata, not a validated claim |
-| Release certification | **Checkpoint Tier 1 certified** on `df-macstudio-m2` (2026-08-15), Hub commit `594de6507dc9`; the formal AXQuant M0-M8 release campaign is a separate process and is not implied |
-
-## Modalities (capability-gated)
-
-Text checkpoint Tier 1 does **not** imply vision or audio quality. `Vision present=true` on a pack is not a quality pass.
-
-| Modality | Claim | Supported | Reason |
-| --- | --- | --- | --- |
-| Vision | `present-not-certified` | `true` | vision sidecar present; mlx-vlm smoke not a quality pass (prefixes=['model.visual']) |
-| Audio | `not-applicable` | `false` | audio not supported on this pack |
+| Release certification | **Not certified**; formal AXQuant M0-M8 gates are not closed |
 
 ## Intended use and limitations
 
@@ -190,10 +174,10 @@ Text checkpoint Tier 1 does **not** imply vision or audio quality. `Vision prese
   KV-cache policy, runtime buffers, and other processes using unified memory.
 - Architecture-prior allocation is not measured sensitivity. It must not be presented as measured
   model quality.
-- MTP requires a sidecar-aware runtime. oMLX/MTPLX discovery compatibility does not establish
-  exactness or speed certification for those runtimes.
+- MTP requires a sidecar-aware runtime. oMLX/MTPLX discovery compatibility does not establish exactness or speed certification for those runtimes.
 - Vision weights are preserved at BF16, but this release does not claim validated VLM quality.
 - The configured context window can require substantially more memory as the KV cache grows.
+- AX Engine execution is not established because this package has no validated native manifest.
 
 - Upstream capabilities, limitations, biases, and responsible-use guidance still apply.
 
@@ -207,7 +191,6 @@ Text checkpoint Tier 1 does **not** imply vision or audio quality. `Vision prese
 - [`axquant_runtime.json`](axquant_runtime.json): declared AX Engine and MLX compatibility metadata; runtime checks remain separate evidence.
 - [`axquant_mtp_sidecar_manifest.json`](axquant_mtp_sidecar_manifest.json): MTP tensor provenance.
 - [`axquant_vision_sidecar_manifest.json`](axquant_vision_sidecar_manifest.json): protected vision tensor provenance.
-- [`model-manifest.json`](model-manifest.json): AX Engine native tensor manifest.
 
 All published provenance uses repository-relative paths. Local source paths are stripped before
 publication. The checkpoint was converted from BF16 rather than re-quantized from an OptiQ

axquant_manifest.json

@@ -1,37 +1,47 @@
 {
-  "axquant_version": "1.8.1",
+  "axquant_version": "1.9.0",
   "calibration": null,
-  "created_at": "2026-08-15T16:16:21.884670Z",
+  "created_at": "2026-10-04T19:41:26.446575Z",
   "effective_bpw": 5.6719817910099914,
   "files": [
     {
       "path": "README.md",
-      "sha256": "1989a02ef81ae864b2efc2cf92536efdeb54f98bc5fc32217b21d1c05a41a1f0",
-      "size_bytes": 10490
+      "sha256": "a1c50ddf3102bc0496925805324d98952b15c85b8c88b87664e528b7578a4ea1",
+      "size_bytes": 9264
     },
     {
       "path": "axquant_mtp_sidecar_manifest.json",
-      "sha256": "11cc9ea99289a7021240024358df53674102c828a8b930930af0a83074e93b87",
+      "sha256": "65b5b43ba1951fbe2be29db9113643ed04a968aef4f36f93b3b3f7819e400504",
       "size_bytes": 880
     },
+    {
+      "path": "axquant_omlx_compat.json",
+      "sha256": "211a3b00bde87ce684d2817c6250738488fa308a9af21e199f9e00b5360d70a5",
+      "size_bytes": 1034
+    },
     {
       "path": "axquant_plan.json",
-      "sha256": "fce5a93299e4ccc15b7d5c4d861572f3ded153b2d76d8da2e8da5919b7435b68",
-      "size_bytes": 1091360
+      "sha256": "04cefba998b61c0938e571b4dc143354060aca9196245acced6cd212b60ba9d0",
+      "size_bytes": 1090819
     },
     {
       "path": "axquant_quantizer_execution.json",
-      "sha256": "c389d287323a820bbbd56cc49d29e0add0011db9c5cc75bc012aa5a43c6eb39b",
-      "size_bytes": 130100
+      "sha256": "f65d9c8333b146d1dee416450d959e9f403a0ca48262be7917b39030ac18a44f",
+      "size_bytes": 129604
     },
     {
       "path": "axquant_runtime.json",
-      "sha256": "35b5323bb62130be059d863dd684c17fab25157657e4788ba0a6a103192390c8",
-      "size_bytes": 1676
+      "sha256": "f0cd02cce04d192eced3dbd34169edb41b41b1b757fd2d7f6539a0d647a5d1aa",
+      "size_bytes": 1704
+    },
+    {
+      "path": "axquant_source_binding.json",
+      "sha256": "b25d5ef759649db215772b8535b5870c1130fe3cd39f619f548415d2337ac76c",
+      "size_bytes": 2293
     },
     {
       "path": "axquant_vision_sidecar_manifest.json",
-      "sha256": "aa54f3fc80257ea5433e7e161e6a66d4cb25335f4fb2a00f69c2b013e2b0ef28",
+      "sha256": "634c27a2e02b48dd6e3634c5972f17a78111e475c7e5a4a8a356a44f1b19092d",
       "size_bytes": 887
     },
     {
@@ -49,6 +59,11 @@
       "sha256": "e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e",
       "size_bytes": 202
     },
+    {
+      "path": "merges.txt",
+      "sha256": "a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d",
+      "size_bytes": 3353259
+    },
     {
       "path": "model-00001-of-00003.safetensors",
       "sha256": "cf4901a75d3c819b69df7002d8134d81982c94d20ba600522add8a2df858033b",
@@ -64,11 +79,6 @@
       "sha256": "5cd45370f41b0d6bf5b8f10def7b0be1cbaf76cdc392320c5778726763a639f5",
       "size_bytes": 4927126499
     },
-    {
-      "path": "model-manifest.json",
-      "sha256": "524163237ca2b879bff41f782a5eee6a1266b4f80f5d2c99e07117b7b899127a",
-      "size_bytes": 433842
-    },
     {
       "path": "model.safetensors.index.json",
       "sha256": "e20dbf6b96058c7b6ccfc1d79745c3f5f4cf8ea4d287ed408da95bdaa4833911",
@@ -84,6 +94,11 @@
       "sha256": "840c099003ecd5e25bcf8cbad13c83ff2335bfb9aebf5d743bfd82b4c0c058ae",
       "size_bytes": 919
     },
+    {
+      "path": "preprocessor_config.json",
+      "sha256": "27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516",
+      "size_bytes": 390
+    },
     {
       "path": "tokenizer.json",
       "sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523",
@@ -98,6 +113,11 @@
       "path": "vision.safetensors",
       "sha256": "d0d927c489250588557d5761dcd47ca267ef64c95f7e7d58dd6a0cccf54ac68f",
       "size_bytes": 921497320
+    },
+    {
+      "path": "vocab.json",
+      "sha256": "ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003",
+      "size_bytes": 6722759
     }
   ],
   "format": "mlx",
@@ -128,7 +148,7 @@
   },
   "mtp_present": true,
   "mtp_weight_file_size_bytes": 849400520,
-  "plan_sha256": "cbfbf9fbc718c89dfbd5241f97cd784bdb62bd8e570c137a0c3fc2c764d8bcb1",
+  "plan_sha256": "c80f0982facd6110ddd9355e025599081b1cfca641eda241fc09a5e1157f9b81",
   "profile": "agent-coding",
   "protected_weight_file_size_bytes": 921497320,
   "quantizer": "axquant",
@@ -155,9 +175,10 @@
         "support_level": "standard-inference"
       }
     ],
-    "created_at": "2026-08-15T16:16:13.776072Z",
+    "created_at": "2026-10-04T19:41:17.898482Z",
     "kv_cache": null,
     "memory_policy": {
+      "expert_stream": "off",
       "kv_cache_precision": "runtime-default",
       "mtp_buffers": "preallocate-when-enabled",
       "prefix_cache": "runtime-managed",
@@ -192,12 +213,12 @@
   },
   "schema_version": "axquant.artifact.v2",
   "software_versions": {
-    "ax_engine": null,
-    "axquant": "1.8.1",
-    "mlx": "0.32.0",
+    "ax_engine": "7.5.7",
+    "axquant": "1.9.0",
+    "mlx": "0.32.1",
     "mlx_lm": "0.31.3",
     "pydantic": "2.13.4",
-    "python": "3.13.14",
+    "python": "3.12.13",
     "safetensors": "0.8.0"
   },
   "source_model": {

axquant_mtp_sidecar_manifest.json

@@ -1,5 +1,5 @@
 {
-  "created_at": "2026-08-15T16:16:13.564577Z",
+  "created_at": "2026-10-04T19:41:17.848062Z",
   "dtypes": [
     "BF16"
   ],

axquant_runtime.json

@@ -21,9 +21,10 @@
       "support_level": "standard-inference"
     }
   ],
-  "created_at": "2026-08-15T16:16:13.776072Z",
+  "created_at": "2026-10-04T19:41:17.898482Z",
   "kv_cache": null,
   "memory_policy": {
+    "expert_stream": "off",
     "kv_cache_precision": "runtime-default",
     "mtp_buffers": "preallocate-when-enabled",
     "prefix_cache": "runtime-managed",

axquant_source_binding.json

@@ -0,0 +1,88 @@
+{
+  "config_sha256": "191e0af232104ed8b65258cf3fb2b842e288008baca7633c11b82a1ac7203aab",
+  "created_at": "2026-10-04T19:25:40.842603Z",
+  "index_sha256": "77042094076611b69791a610065f28b7013b8c621795fa86ddccc8bac7d1b9df",
+  "members": [
+    {
+      "path": "model-00001-of-00018.safetensors",
+      "size_bytes": 3966730552
+    },
+    {
+      "path": "model-00002-of-00018.safetensors",
+      "size_bytes": 3043080328
+    },
+    {
+      "path": "model-00003-of-00018.safetensors",
+      "size_bytes": 2542796952
+    },
+    {
+      "path": "model-00004-of-00018.safetensors",
+      "size_bytes": 3988973152
+    },
+    {
+      "path": "model-00005-of-00018.safetensors",
+      "size_bytes": 2099339864
+    },
+    {
+      "path": "model-00006-of-00018.safetensors",
+      "size_bytes": 3979553696
+    },
+    {
+      "path": "model-00007-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00008-of-00018.safetensors",
+      "size_bytes": 3979553696
+    },
+    {
+      "path": "model-00009-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00010-of-00018.safetensors",
+      "size_bytes": 3979553696
+    },
+    {
+      "path": "model-00011-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00012-of-00018.safetensors",
+      "size_bytes": 3979553696
+    },
+    {
+      "path": "model-00013-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00014-of-00018.safetensors",
+      "size_bytes": 3979553696
+    },
+    {
+      "path": "model-00015-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00016-of-00018.safetensors",
+      "size_bytes": 3979564040
+    },
+    {
+      "path": "model-00017-of-00018.safetensors",
+      "size_bytes": 2108759344
+    },
+    {
+      "path": "model-00018-of-00018.safetensors",
+      "size_bytes": 3392197344
+    }
+  ],
+  "plan_sha256": "c80f0982facd6110ddd9355e025599081b1cfca641eda241fc09a5e1157f9b81",
+  "schema_version": "axquant.source-plan-binding.v1",
+  "source_model": {
+    "architecture": "Qwen3_5ForConditionalGeneration",
+    "format": "mlx",
+    "local_path": null,
+    "model_id": "Qwen/Qwen3.8-27B",
+    "revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0"
+  }
+}

axquant_vision_sidecar_manifest.json

@@ -1,5 +1,5 @@
 {
-  "created_at": "2026-08-15T16:16:11.098120Z",
+  "created_at": "2026-10-04T19:41:15.156713Z",
   "dtypes": [
     "BF16"
   ],

preprocessor_config.json

@@ -0,0 +1,21 @@
+{
+    "size": {
+        "longest_edge": 16777216,
+        "shortest_edge": 65536
+    },
+    "patch_size": 16,
+    "temporal_patch_size": 2,
+    "merge_size": 2,
+    "image_mean": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "image_std": [
+        0.5,
+        0.5,
+        0.5
+    ],
+    "processor_class": "Qwen3VLProcessor",
+    "image_processor_type": "Qwen2VLImageProcessorFast"
+}