Hugging Face's logo

juspay
/
xor

Text Generation
Transformers
Safetensors
qwen3_5_moe
image-text-to-text
sglang
mixture-of-experts
typed-classification
conversational
Model card Files
xet
 Community
pramodchandra commited on
Commit
a286a46
·
verified ·
1 Parent(s): 8d2d708

Release Xor 1.2

Browse files
Files changed (23) hide show
  1. README.md +24 -25
  2. RELEASE_PROVENANCE.json +37 -15
  3. checksums.sha256 +18 -18
  4. generation_config.json +11 -10
  5. model-00001-of-00016.safetensors +1 -1
  6. model-00002-of-00016.safetensors +1 -1
  7. model-00003-of-00016.safetensors +1 -1
  8. model-00004-of-00016.safetensors +1 -1
  9. model-00005-of-00016.safetensors +1 -1
  10. model-00006-of-00016.safetensors +1 -1
  11. model-00007-of-00016.safetensors +1 -1
  12. model-00008-of-00016.safetensors +1 -1
  13. model-00009-of-00016.safetensors +1 -1
  14. model-00010-of-00016.safetensors +1 -1
  15. model-00011-of-00016.safetensors +1 -1
  16. model-00012-of-00016.safetensors +1 -1
  17. model-00013-of-00016.safetensors +1 -1
  18. model-00014-of-00016.safetensors +1 -1
  19. model-00015-of-00016.safetensors +1 -1
  20. model-00016-of-00016.safetensors +1 -1
  21. serving/{xor-1.1-serving.tar.gz → xor-1.2-serving.tar.gz} +2 -2
  22. serving/{xor-1.1-serving.tar.gz.sha256 → xor-1.2-serving.tar.gz.sha256} +1 -1
  23. tokenizer_config.json +2 -0
README.md CHANGED
@@ -10,30 +10,29 @@ tags:
10
  - typed-classification
11
  ---
12
 
13
- # Xor 1.1
14
 
15
- Xor 1.1 (`xor-1.1`) is a post-trained version of [Qwen/Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) for typed decision tasks. It is served through a TypeSafe-compatible `/v1/systemone` API.
16
 
17
- ## Changes from Xor 1.0
18
 
19
  - New post-trained weights (LoRA rank 16, merged into BF16 base weights)
20
- - `choice` and `score` questions accept 2 to 255 candidates (previously 26)
21
- - Video input in addition to images; remote media URLs are rejected, only data URLs are accepted
22
- - Calibration as per question types
23
- - Validated on one GPU (TP1) with a pinned, patched SGLang runtime
24
 
25
  ## Revisions
26
 
27
  | Revision | Release |
28
  |---|---|
29
- | `main` | Latest release (currently Xor 1.1) |
 
30
  | `v1.1` | Xor 1.1, immutable |
31
  | `xor-v1` | Xor 1.0, immutable |
32
 
33
- Pin a revision for reproducible results:
34
 
35
  ```bash
36
- hf download juspay/xor --revision v1.1 --local-dir xor-1.1
37
  ```
38
 
39
  ## Base model
@@ -68,16 +67,16 @@ The serving layer performs deterministic single-token candidate readout, forward
68
  Download the release, verify and extract the serving bundle, and start Xor on one GPU:
69
 
70
  ```bash
71
- hf download juspay/xor --revision v1.1 --local-dir xor-1.1
72
 
73
- (cd xor-1.1/serving && sha256sum -c xor-1.1-serving.tar.gz.sha256)
74
 
75
- mkdir -p xor-1.1-runtime
76
- tar -xzf xor-1.1/serving/xor-1.1-serving.tar.gz -C xor-1.1-runtime --strip-components=1
77
 
78
- cd xor-1.1-runtime
79
  cp .env.example .env
80
- sed -i "s|^MODEL_DIR=.*|MODEL_DIR=$(cd ../xor-1.1 && pwd)|" .env
81
  ./run.sh
82
  ```
83
 
@@ -93,22 +92,22 @@ curl -sS -X POST http://127.0.0.1:30002/v1/systemone \
93
  --data @examples/image-request.json
94
  ```
95
 
96
- The setup requires Linux x86-64, the Hugging Face CLI, Docker Engine with Docker Compose v2, the NVIDIA Container Toolkit, and approximately 120 GB of free disk space.
97
 
98
  ## Public JEVBench self-run
99
 
100
- Xor 1.1 was evaluated locally on the public tiers of [JEVBench](https://github.com/fstandhartinger/jevbench) using harness commit `1bcc55eb6c8cffde2306b3db03ede39b61c6152a`, the existing `typesafe` adapter, and one request at a time.
101
 
102
- The run used the released serving bundle unmodified on 1 x NVIDIA H200 (143 GB), tensor parallelism 1, data parallelism 1, with request caching disabled.
103
 
104
  | Tier | Attempted | Valid | Correct | Accuracy | p50 | p95 |
105
  |---|---:|---:|---:|---:|---:|---:|
106
- | Easy | 48 | 48 | 48 | 1.0000 | 0.0739 s | 0.0772 s |
107
- | Original | 72 | 72 | 70 | 0.9722 | 0.0742 s | 0.0780 s |
108
- | Hard public | 111 | 111 | 89 | 0.8018 | 0.0781 s | 0.1378 s |
109
- | **All public** | **231** | **231** | **207** | **0.8961** | 0.0753 s | 0.1207 s |
110
 
111
- Across all 231 public decisions: macro accuracy 0.9056, Brier mean 0.1704, ECE 0.0559. Operational success, coverage, schema validity, and strict schema validity were 1.0000.
112
 
113
  These are self-run public-tier results, not an official JEVBench rank. Latency is hardware-specific and was measured locally without network overhead.
114
 
@@ -120,7 +119,7 @@ These are self-run public-tier results, not an official JEVBench rank. Latency i
120
  | Upstream SGLang base | `lmsysorg/sglang@sha256:6bcaa47db52f78ce0d67863b8b2431221b79bc23204a80cad757fa819d00e921` |
121
  | Tensor parallelism | 1 |
122
  | Data parallelism | 1 |
123
- | Validated GPU | 1 x NVIDIA H200, 143 GB |
124
  | Maximum prefill tokens | 250,000 |
125
  | Static memory fraction | 0.85 |
126
 
 
10
  - typed-classification
11
  ---
12
 
13
+ # Xor 1.2
14
 
15
+ Xor 1.2 (`xor-1.2`) is a post-trained version of [Qwen/Qwen3.6-35B-A3B](https://huggingface.co/Qwen/Qwen3.6-35B-A3B) for typed decision tasks. It is served through a TypeSafe-compatible `/v1/systemone` API.
16
 
17
+ ## Changes from Xor 1.1
18
 
19
  - New post-trained weights (LoRA rank 16, merged into BF16 base weights)
20
+ - Serving bundle, API, calibration, and media handling are unchanged from Xor 1.1
21
+ - Validated on one GPU (TP1) with the same pinned, patched SGLang runtime
 
 
22
 
23
  ## Revisions
24
 
25
  | Revision | Release |
26
  |---|---|
27
+ | `main` | Latest stable release (currently Xor 1.1) |
28
+ | `release/xor-1.2` | Xor 1.2 |
29
  | `v1.1` | Xor 1.1, immutable |
30
  | `xor-v1` | Xor 1.0, immutable |
31
 
32
+ Pin a revision for reproducible results. `release/xor-1.2` is a branch; pin its commit hash for byte-exact reproducibility.
33
 
34
  ```bash
35
+ hf download juspay/xor --revision release/xor-1.2 --local-dir xor-1.2
36
  ```
37
 
38
  ## Base model
 
67
  Download the release, verify and extract the serving bundle, and start Xor on one GPU:
68
 
69
  ```bash
70
+ hf download juspay/xor --revision release/xor-1.2 --local-dir xor-1.2
71
 
72
+ (cd xor-1.2/serving && sha256sum -c xor-1.2-serving.tar.gz.sha256)
73
 
74
+ mkdir -p xor-1.2-runtime
75
+ tar -xzf xor-1.2/serving/xor-1.2-serving.tar.gz -C xor-1.2-runtime --strip-components=1
76
 
77
+ cd xor-1.2-runtime
78
  cp .env.example .env
79
+ sed -i "s|^MODEL_DIR=.*|MODEL_DIR=$(cd ../xor-1.2 && pwd)|" .env
80
  ./run.sh
81
  ```
82
 
 
92
  --data @examples/image-request.json
93
  ```
94
 
95
+ The setup requires Linux x86-64, the Hugging Face CLI, Docker Engine with a recent Docker Compose v2 (the bundle uses the service-level `gpus` key, which older Compose releases reject), the NVIDIA Container Toolkit, and approximately 120 GB of free disk space.
96
 
97
  ## Public JEVBench self-run
98
 
99
+ Xor 1.2 was evaluated locally on the public tiers of [JEVBench](https://github.com/fstandhartinger/jevbench) using harness commit `1bcc55eb6c8cffde2306b3db03ede39b61c6152a`, the existing `typesafe` adapter, and one request at a time.
100
 
101
+ The run used the released serving bundle unmodified on 1 x NVIDIA H200 NVL (143 GB), tensor parallelism 1, data parallelism 1, with request caching disabled.
102
 
103
  | Tier | Attempted | Valid | Correct | Accuracy | p50 | p95 |
104
  |---|---:|---:|---:|---:|---:|---:|
105
+ | Easy | 48 | 48 | 48 | 1.0000 | 0.0679 s | 0.0709 s |
106
+ | Original | 72 | 72 | 69 | 0.9583 | 0.0670 s | 0.0701 s |
107
+ | Hard public | 111 | 111 | 91 | 0.8198 | 0.0832 s | 0.1845 s |
108
+ | **All public** | **231** | **231** | **208** | **0.9004** | 0.0691 s | 0.1616 s |
109
 
110
+ Across all 231 public decisions: macro accuracy 0.9070, Brier mean 0.1790, ECE 0.0733. Operational success, coverage, schema validity, and strict schema validity were 1.0000.
111
 
112
  These are self-run public-tier results, not an official JEVBench rank. Latency is hardware-specific and was measured locally without network overhead.
113
 
 
119
  | Upstream SGLang base | `lmsysorg/sglang@sha256:6bcaa47db52f78ce0d67863b8b2431221b79bc23204a80cad757fa819d00e921` |
120
  | Tensor parallelism | 1 |
121
  | Data parallelism | 1 |
122
+ | Validated GPU | 1 x NVIDIA H200 NVL, 143 GB |
123
  | Maximum prefill tokens | 250,000 |
124
  | Static memory fraction | 0.85 |
125
 
RELEASE_PROVENANCE.json CHANGED
@@ -1,30 +1,49 @@
1
  {
2
  "release": {
3
- "model_id": "xor-1.1",
4
- "release_date": "2026-09-27",
5
  "precision": "bfloat16",
6
  "parameters": 35107181936,
7
  "weight_shards": 16,
8
- "candidate": "F10 (lora35-f10)"
 
 
9
  },
10
  "base_model": {
11
  "repository": "Qwen/Qwen3.6-35B-A3B",
12
  "revision": "995ad96eacd98c81ed38be0c5b274b04031597b0",
13
- "revision_evidence": "Hugging Face download metadata retained with the merge input"
14
  },
15
  "adapter": {
16
  "peft_type": "LORA",
17
  "rank": 16,
18
  "alpha": 32,
 
 
19
  "dropout": 0.05,
20
- "config_sha256": "13123d5699769f67c903e5cadc03c258007e0e88d3904c8ac4758aac651804be",
21
- "weights_sha256": "2f2e528d0e7c0d4b2d0f6f6fac60e6e50a0d35e8ca291b2cda555df54db277aa",
 
22
  "peft_version": "0.21.0"
23
  },
24
  "merge": {
25
- "method": "PEFT merge_and_unload(safe_merge=True) followed by BF16 save_pretrained (5 GB shards)",
26
- "script": "merge_adapter.py",
27
- "script_sha256": "5feca24cb3a8d42d8b1115b9c4a82e4ec0b05201de2f7682eb5d89fbb75c07f7",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  "merged_index_sha256": "f8448aa8b2fbf3519723b0c2965fc94ed64cafdd463381636a1df351952b3f1a",
29
  "base_tokenizer_files_copied_from_pinned_base": [
30
  "merges.txt",
@@ -36,9 +55,9 @@
36
  "sglang_image": "prakhar1611/xor-sglang@sha256:94c48d2a6cc98dc456cf93f723707ea7dd81dddfe1061e823b348d68bbe8158f",
37
  "sglang_upstream_image": "lmsysorg/sglang@sha256:6bcaa47db52f78ce0d67863b8b2431221b79bc23204a80cad757fa819d00e921",
38
  "sglang_patch_dockerfile_sha256": "e05f93d4537cad3e1837fff5e80f511ab1599cf64827cb87aa0853f3d4837d08",
39
- "wrapper_sha256": "9d13e9f9c0ac9386cfa09d2c07e30fd8f4d1f947429061a42902fbd29fa4d63a",
40
- "compose_sha256": "a461e57bce290e31f197763b8c97178a3ed6c118c258d2ac6c2ae2448ad35a82",
41
- "source_parent_commit": "3f102081f90e78ac802f80ffdedb8afb1bed1be9",
42
  "tensor_parallel_size": 1,
43
  "data_parallel_size": 1,
44
  "marker_count": 255,
@@ -48,11 +67,14 @@
48
  "noul": 1.4,
49
  "score": 1.0
50
  },
51
- "release_benchmark": "benchmarks/20260927-xor11-release-jevbench.json",
52
- "release_benchmark_sha256": "dc8fa6edfb191177a1512221742ac3f010d42524d2970c931bd5c909bb6380e6"
53
  },
54
  "known_provenance_gaps": [
55
  "The adapter artifact does not contain trainer_state.json.",
56
- "adapter_config.json does not record the base revision; the pinned revision was verified by hashing all base files against Hugging Face download metadata before merging."
 
 
 
57
  ]
58
  }
 
1
  {
2
  "release": {
3
+ "model_id": "xor-1.2",
4
+ "release_date": "2026-09-28",
5
  "precision": "bfloat16",
6
  "parameters": 35107181936,
7
  "weight_shards": 16,
8
+ "candidate": "F12 a24 (lora35-f12, merged at lora_alpha 24)",
9
+ "received_archive": "f12-a24-champion-52.23-v0.2.1.zip",
10
+ "received_archive_sha256": "0ba00678790e57555e842302a4540726b1c8503d7b199350c6a5eb4208c7133e"
11
  },
12
  "base_model": {
13
  "repository": "Qwen/Qwen3.6-35B-A3B",
14
  "revision": "995ad96eacd98c81ed38be0c5b274b04031597b0",
15
+ "revision_evidence": "All 26 weight shards and tokenizer.json of the merge input were hashed and matched the pinned revision; merges.txt, vocab.json and configuration.json matched the Xor 1.1 release copies of the same pinned files."
16
  },
17
  "adapter": {
18
  "peft_type": "LORA",
19
  "rank": 16,
20
  "alpha": 32,
21
+ "merge_alpha": 24,
22
+ "merge_scale": 1.5,
23
  "dropout": 0.05,
24
+ "target_tables": 310,
25
+ "config_sha256": "9fe94e209f1662e12cdad136c62b58b21e655a28c21e986618a76064fed71be3",
26
+ "weights_sha256": "eb52e577d5f9da0945f3743c3e50ccfc7fb20f3aa10ac6b56774ab76c092c1e0",
27
  "peft_version": "0.21.0"
28
  },
29
  "merge": {
30
+ "method": "PEFT merge_and_unload(safe_merge=True) at lora_alpha 24 followed by BF16 save_pretrained, then re-split without changing any tensor into the 16-shard layout of Xor 1.1",
31
+ "script": "merge_box.py (supplied with the adapter archive), variant a24",
32
+ "script_sha256": "18ba79692fb89cadb7d42073bd60e9d44feeefc70a3fbff55672d3559adad40f",
33
+ "environment": {
34
+ "image": "lmsysorg/sglang@sha256:6bcaa47db52f78ce0d67863b8b2431221b79bc23204a80cad757fa819d00e921",
35
+ "torch": "2.13.0+cu130",
36
+ "transformers": "5.12.1",
37
+ "peft": "0.21.0",
38
+ "safetensors": "0.8.0",
39
+ "gpu": "1 x NVIDIA H200 NVL"
40
+ },
41
+ "trainer_merge_match": "The unsplit merge output was byte-identical (both weight shards, index and config) to the trainer's own F12 a24 merge.",
42
+ "reshard": {
43
+ "method": "Every tensor copied byte for byte into the shard assigned by the Xor 1.1 index; all 1026 tensors compared equal to the unsplit merge afterwards",
44
+ "layout_template_index_sha256": "f8448aa8b2fbf3519723b0c2965fc94ed64cafdd463381636a1df351952b3f1a",
45
+ "script_sha256": "276e342d4377e372149f6fb6ff5cd4c0adfc8952393677d6cf66ad2662acb4f9"
46
+ },
47
  "merged_index_sha256": "f8448aa8b2fbf3519723b0c2965fc94ed64cafdd463381636a1df351952b3f1a",
48
  "base_tokenizer_files_copied_from_pinned_base": [
49
  "merges.txt",
 
55
  "sglang_image": "prakhar1611/xor-sglang@sha256:94c48d2a6cc98dc456cf93f723707ea7dd81dddfe1061e823b348d68bbe8158f",
56
  "sglang_upstream_image": "lmsysorg/sglang@sha256:6bcaa47db52f78ce0d67863b8b2431221b79bc23204a80cad757fa819d00e921",
57
  "sglang_patch_dockerfile_sha256": "e05f93d4537cad3e1837fff5e80f511ab1599cf64827cb87aa0853f3d4837d08",
58
+ "wrapper_sha256": "a44d9dcde4d2e48b43d6b367dae03a0d4e62edd05e639d1a483481435420ec21",
59
+ "compose_sha256": "24338aa871af898cec3b5dc1262d3e57926aefc5f9fe07e7f19ef231a118d32d",
60
+ "source_parent_commit": "fed366a6bc7c0f00b53253a4afc9eb29c811f931",
61
  "tensor_parallel_size": 1,
62
  "data_parallel_size": 1,
63
  "marker_count": 255,
 
67
  "noul": 1.4,
68
  "score": 1.0
69
  },
70
+ "release_benchmark": "benchmarks/20260928-xor12-release-jevbench.json",
71
+ "release_benchmark_sha256": "1143ac7225a80466579a3abaa57da2d919dc3edfa96551e9b9db2cae41bdcc14"
72
  },
73
  "known_provenance_gaps": [
74
  "The adapter artifact does not contain trainer_state.json.",
75
+ "adapter_config.json does not record the base revision; the pinned revision was verified by hashing the base files used by the merge before merging.",
76
+ "adapter_config.json records lora_alpha 32; the release merges at lora_alpha 24 (scale 1.5), the variant selected by the trainer. The alpha change is applied by merge_box.py at merge time.",
77
+ "The merge used merge_box.py supplied with the adapter rather than serving/merge_adapter.py; both call PEFT merge_and_unload(safe_merge=True).",
78
+ "The 19 mtp.* (multi-token prediction) tensors of the base checkpoint are not saved by the transformers model class used for the merge, as in Xor 1.1; MTP speculative decoding is not available."
79
  ]
80
  }
checksums.sha256 CHANGED
@@ -1,27 +1,27 @@
1
  e84f32a23fdda27689f868aa4a1a5621f41133e51a48d7f3efcbea2839574259 chat_template.jinja
2
  85eeca65b582011bb78e97bdbef4eee4127df73c96eed698dadac34c1ebbb7b0 config.json
3
  c1b09db419119513247e9b8b912c4b9897106c9b20c6cada7e107d993c5435eb configuration.json
4
- e70c136c1b78ddc1fb0905bac8e733a4dc448d4f852a5dd75143fffc70be550e generation_config.json
5
  a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d merges.txt
6
- 3306361f668701765dfdbdbb0391c3c38b2c58babf785d9c442fa0a6a46bcdf2 model-00001-of-00016.safetensors
7
- b28819fd258addd1f2787fc12dce2c557573edd44ea8081adf8d5856a3854510 model-00002-of-00016.safetensors
8
- 3bd4a6e3985b4468fd9ff4bc466bf3fcddb1aa2ab1cba8854980f993c17c60b0 model-00003-of-00016.safetensors
9
- 44d10bd135468117e6d81fda5d404f900190cc6040acdce98832023a61c4e0e1 model-00004-of-00016.safetensors
10
- 75bc1b5a3407e028bf2f90e26ffb23f3e173a9a645bded33645ed9d31a4de7b9 model-00005-of-00016.safetensors
11
- ef2e5aa5c98951f1ffdfef4029d46e1e984fa6f288f4a13efb0d9f844b69ff18 model-00006-of-00016.safetensors
12
- 74a8519f6c6797f9e411cf1b22a586c1c9cdbc4f908e8f025e3302d9e0ec3421 model-00007-of-00016.safetensors
13
- 1ad9c92ba443364f9d0253ef4c5514eba27c287c26abe2cec1526858fd93e140 model-00008-of-00016.safetensors
14
- 72ee7fc1639f2fc2b942077fe1f5b772a7c7733d75a5b3521fdfcbd460c680a0 model-00009-of-00016.safetensors
15
- 24b6f3bb385590bb67b32696035174e4dbf42f89e3a369c1718541e58a33954d model-00010-of-00016.safetensors
16
- bb7416d18517d6facd0111ca94e1c7d9bda9a6942043be7d8f7bf9514db628bb model-00011-of-00016.safetensors
17
- 2ca4cd3b98a341e409e2803ba9a7b2738cddb236803c3e771ca530bc1207de47 model-00012-of-00016.safetensors
18
- e2f4658acc20abfadf9ca99d21b21a966faed00d18fe0ffe3f49cc6c5477aee2 model-00013-of-00016.safetensors
19
- e9b5af4ced763d15cded104db78bf5275796394e1832afa57e05dce967bb98f3 model-00014-of-00016.safetensors
20
- 0797c28fe924c2437e457bbb3218f860042b767873027d8938f210bdd4ba1ee9 model-00015-of-00016.safetensors
21
- 578a9b700e0100df1e2f543be6161646e9f0fd8701f658720a70ec1bfbf28e73 model-00016-of-00016.safetensors
22
  f8448aa8b2fbf3519723b0c2965fc94ed64cafdd463381636a1df351952b3f1a model.safetensors.index.json
23
  27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516 preprocessor_config.json
 
24
  06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523 tokenizer.json
25
- 91a08f825d370d085d692e04cf117cdd7faad7bf18e996f1e6031b6dab03db72 tokenizer_config.json
26
  7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13 video_preprocessor_config.json
27
  ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003 vocab.json
 
1
  e84f32a23fdda27689f868aa4a1a5621f41133e51a48d7f3efcbea2839574259 chat_template.jinja
2
  85eeca65b582011bb78e97bdbef4eee4127df73c96eed698dadac34c1ebbb7b0 config.json
3
  c1b09db419119513247e9b8b912c4b9897106c9b20c6cada7e107d993c5435eb configuration.json
4
+ a4cef85934ea1fdcb207944dbc6eee70dbbf16806874428556ae33023336c0a4 generation_config.json
5
  a9d356d7bdf1ef4949e3e748e95b8e10ad9d4e2e838eddc38a0a7b6b94d1db8d merges.txt
6
+ 953b595b4bcfb712d84b14120a37cd3b0072fe7175ff1ec24c3499356e7f03af model-00001-of-00016.safetensors
7
+ 22baafc887c1f9aaeab5f97dde7b64b071572b9a8c23219ff5c58d53f7d6add7 model-00002-of-00016.safetensors
8
+ 8013cd5a13f958466eb4d7c0c52111183bddfb80452f8b817f2a8000611822a0 model-00003-of-00016.safetensors
9
+ de62ae1755ba59e1a4760abf17c735ad088343a5b9288e37d69641a218ba5b22 model-00004-of-00016.safetensors
10
+ c57248e9eeef143c5d307db3e2867616714950670c6e7b81eba73f3a97824c62 model-00005-of-00016.safetensors
11
+ 27a7d888234efccb2bc33bdc94522bed3adbc2c20e93b72bdfe9144dae8b78cf model-00006-of-00016.safetensors
12
+ caaf6974487cc1329a6f70c6694f84197d07831f4802975e5e99720c1fe237a0 model-00007-of-00016.safetensors
13
+ e72f8676d4daa8a418c69601fb5970c35ad7a194ddf4d3795846392ca1bcc1f8 model-00008-of-00016.safetensors
14
+ ce940ef333525fd39e68cb19407c5d831adf511f2eca77a93d567f582e41bdd4 model-00009-of-00016.safetensors
15
+ a21c0a631a4fddc954c73a0b9070ecfd4da952669b20a5abc799c18248eb5544 model-00010-of-00016.safetensors
16
+ 929c3a1e9e93e6c6b5e70d709bd920f96ec367492a1e1d096bb94f6138a332c9 model-00011-of-00016.safetensors
17
+ 89378fb259e321e2a053da95a363f6e2f0cad9178943388c6e65e392c0af1550 model-00012-of-00016.safetensors
18
+ 9460d07ce62430e5a52fbbea91ec4fb455b2914b3efe75c4c32e5d8ecbb36fe0 model-00013-of-00016.safetensors
19
+ 0b8bd8de460c8b72721fd35bb94e421697d923d4e059063a5e6d15af1a14e7f6 model-00014-of-00016.safetensors
20
+ 0972ff1c2678a16c32187b114298f23ab5a7ea3d5ab97668a84388a3d17d1808 model-00015-of-00016.safetensors
21
+ 9112172742877d7530f891b25cbedf1d10cfe4165a814d3aeb648fd257bd4a8e model-00016-of-00016.safetensors
22
  f8448aa8b2fbf3519723b0c2965fc94ed64cafdd463381636a1df351952b3f1a model.safetensors.index.json
23
  27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516 preprocessor_config.json
24
+ 66e427c470fe580fe8c7b5725d857af23d8417e37fae62667ec698306a19987b tokenizer_config.json
25
  06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523 tokenizer.json
 
26
  7768af27c1fafa9cc9011c1dc20067e03f8915e03b63504550e11d5066986d13 video_preprocessor_config.json
27
  ce99b4cb2983d118806ce0a8b777a35b093e2000a503ebde25853284c9dfa003 vocab.json
generation_config.json CHANGED
@@ -1,12 +1,13 @@
1
  {
2
- "bos_token_id": 248044,
3
- "do_sample": true,
4
- "eos_token_id": [
5
- 248046,
6
- 248044
7
- ],
8
- "pad_token_id": 248044,
9
- "temperature": 1.0,
10
- "top_k": 20,
11
- "top_p": 0.95
 
12
  }
 
1
  {
2
+ "bos_token_id": 248044,
3
+ "do_sample": true,
4
+ "eos_token_id": [
5
+ 248046,
6
+ 248044
7
+ ],
8
+ "pad_token_id": 248044,
9
+ "temperature": 1.0,
10
+ "top_k": 20,
11
+ "top_p": 0.95,
12
+ "transformers_version": "5.12.1"
13
  }
model-00001-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3306361f668701765dfdbdbb0391c3c38b2c58babf785d9c442fa0a6a46bcdf2
3
  size 4323955448
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:953b595b4bcfb712d84b14120a37cd3b0072fe7175ff1ec24c3499356e7f03af
3
  size 4323955448
model-00002-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b28819fd258addd1f2787fc12dce2c557573edd44ea8081adf8d5856a3854510
3
  size 4506431768
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22baafc887c1f9aaeab5f97dde7b64b071572b9a8c23219ff5c58d53f7d6add7
3
  size 4506431768
model-00003-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3bd4a6e3985b4468fd9ff4bc466bf3fcddb1aa2ab1cba8854980f993c17c60b0
3
  size 4988775056
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8013cd5a13f958466eb4d7c0c52111183bddfb80452f8b817f2a8000611822a0
3
  size 4988775056
model-00004-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:44d10bd135468117e6d81fda5d404f900190cc6040acdce98832023a61c4e0e1
3
  size 3962207584
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de62ae1755ba59e1a4760abf17c735ad088343a5b9288e37d69641a218ba5b22
3
  size 3962207584
model-00005-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:75bc1b5a3407e028bf2f90e26ffb23f3e173a9a645bded33645ed9d31a4de7b9
3
  size 4506431736
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c57248e9eeef143c5d307db3e2867616714950670c6e7b81eba73f3a97824c62
3
  size 4506431736
model-00006-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ef2e5aa5c98951f1ffdfef4029d46e1e984fa6f288f4a13efb0d9f844b69ff18
3
  size 4988775104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:27a7d888234efccb2bc33bdc94522bed3adbc2c20e93b72bdfe9144dae8b78cf
3
  size 4988775104
model-00007-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:74a8519f6c6797f9e411cf1b22a586c1c9cdbc4f908e8f025e3302d9e0ec3421
3
  size 3962207632
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:caaf6974487cc1329a6f70c6694f84197d07831f4802975e5e99720c1fe237a0
3
  size 3962207632
model-00008-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1ad9c92ba443364f9d0253ef4c5514eba27c287c26abe2cec1526858fd93e140
3
  size 4506431816
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e72f8676d4daa8a418c69601fb5970c35ad7a194ddf4d3795846392ca1bcc1f8
3
  size 4506431816
model-00009-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:72ee7fc1639f2fc2b942077fe1f5b772a7c7733d75a5b3521fdfcbd460c680a0
3
  size 4988775104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce940ef333525fd39e68cb19407c5d831adf511f2eca77a93d567f582e41bdd4
3
  size 4988775104
model-00010-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:24b6f3bb385590bb67b32696035174e4dbf42f89e3a369c1718541e58a33954d
3
  size 3962207632
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a21c0a631a4fddc954c73a0b9070ecfd4da952669b20a5abc799c18248eb5544
3
  size 3962207632
model-00011-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:bb7416d18517d6facd0111ca94e1c7d9bda9a6942043be7d8f7bf9514db628bb
3
  size 4506431816
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:929c3a1e9e93e6c6b5e70d709bd920f96ec367492a1e1d096bb94f6138a332c9
3
  size 4506431816
model-00012-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2ca4cd3b98a341e409e2803ba9a7b2738cddb236803c3e771ca530bc1207de47
3
  size 4988775104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89378fb259e321e2a053da95a363f6e2f0cad9178943388c6e65e392c0af1550
3
  size 4988775104
model-00013-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e2f4658acc20abfadf9ca99d21b21a966faed00d18fe0ffe3f49cc6c5477aee2
3
  size 3962207632
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9460d07ce62430e5a52fbbea91ec4fb455b2914b3efe75c4c32e5d8ecbb36fe0
3
  size 3962207632
model-00014-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e9b5af4ced763d15cded104db78bf5275796394e1832afa57e05dce967bb98f3
3
  size 4506431816
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b8bd8de460c8b72721fd35bb94e421697d923d4e059063a5e6d15af1a14e7f6
3
  size 4506431816
model-00015-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0797c28fe924c2437e457bbb3218f860042b767873027d8938f210bdd4ba1ee9
3
  size 4988775104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0972ff1c2678a16c32187b114298f23ab5a7ea3d5ab97668a84388a3d17d1808
3
  size 4988775104
model-00016-of-00016.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:578a9b700e0100df1e2f543be6161646e9f0fd8701f658720a70ec1bfbf28e73
3
  size 2565674304
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9112172742877d7530f891b25cbedf1d10cfe4165a814d3aeb648fd257bd4a8e
3
  size 2565674304
serving/{xor-1.1-serving.tar.gz → xor-1.2-serving.tar.gz} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9d25e74bdda5faebb0168fd538934715114e5e7f9c00cccf9ba2ad7f6f708088
3
- size 32756
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bda761ac9f21185d096b5e459eab77793208aa4e3277eb3d85722916bf4093da
3
+ size 33522
serving/{xor-1.1-serving.tar.gz.sha256 → xor-1.2-serving.tar.gz.sha256} RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:df01bdfd3c22bf9ba231b402bb09d460596f0d5cc4cdca335943fa3f994778a7
3
  size 89
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa78c759522671589c1c3f16012dbd33362c90bdc00903248a90f5a29b8635c3
3
  size 89
tokenizer_config.json CHANGED
@@ -9,6 +9,8 @@
9
  "eos_token": "<|im_end|>",
10
  "errors": "replace",
11
  "image_token": "<|image_pad|>",
 
 
12
  "model_max_length": 262144,
13
  "model_specific_special_tokens": {
14
  "audio_bos_token": "<|audio_start|>",
 
9
  "eos_token": "<|im_end|>",
10
  "errors": "replace",
11
  "image_token": "<|image_pad|>",
12
+ "is_local": true,
13
+ "local_files_only": false,
14
  "model_max_length": 262144,
15
  "model_specific_special_tokens": {
16
  "audio_bos_token": "<|audio_start|>",