Hugging Face's logo

AutomatosX
/
AX-Unlimited-OCR-3B-MoE-MLX-MXFP8

Image-Text-to-Text
MLX
Safetensors
unlimited-ocr
ax-engine
mlx-vlm
ocr
mxfp8
int8
apple-silicon
automatosx
conversational
8-bit precision
Model card Files
xet
 Community
AutomatosX commited on
Commit
af1f1f3
·
verified ·
1 Parent(s): 16d38e0

Document AX Engine as primary inference; ship model-manifest.json

Browse files
Files changed (2) hide show
  1. README.md +82 -30
  2. model-manifest.json +2798 -0
README.md CHANGED
@@ -1,40 +1,23 @@
1
  ---
2
  license: mit
3
  base_model: baidu/Unlimited-OCR
4
- library_name: mlx-vlm
5
  pipeline_tag: image-text-to-text
6
  tags:
 
7
  - mlx
8
  - mlx-vlm
9
  - ocr
10
  - mxfp8
11
  - int8
12
  - apple-silicon
 
13
  widget:
14
  - text: "<image>document parsing."
15
  - text: "<image>Convert the document to markdown."
16
  - text: "<image>Free OCR. Extract all text."
17
  ---
18
 
19
- ## Release tags
20
-
21
- | Tag | Commit | Notes |
22
- |-----|--------|-------|
23
- | **v0.2.0** (current) | `6708e4d86b0a342c5891e47516aa412bbfd2eda2` | Sensitivity-guided MXFP8 + calibrated affine INT8 LM head |
24
- | **v0.1.0** | `fd0c04fc20db886f8a80fe1177081b07206586c7` | First OCR-aware MXFP8 publish |
25
-
26
- Pin production loads with `revision="v0.2.0"` (or the commit SHA). Floating `main` may move on future publishes.
27
-
28
- ```bash
29
- python -m mlx_vlm.generate \
30
- --model AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 \
31
- --revision v0.2.0 \
32
- --image document.png \
33
- --prompt '<image>document parsing.' \
34
- --max-tokens 4096
35
- ```
36
-
37
-
38
  # AX Unlimited-OCR 3B MoE — MLX (MXFP8 + affine INT8 head)
39
 
40
  This is an AutomatosX sensitivity-guided MLX conversion of
@@ -44,11 +27,24 @@ created from the upstream BF16 weights. The
44
  checkpoint was used only as a comparison reference; these are not repackaged
45
  Sahil weights.
46
 
 
 
 
 
47
  **Hybrid precision (name is shorthand):** MXFP8 (group size 32) on eligible
48
  language linears, **affine INT8** on the LM head, **BF16** on the vision
49
  encoder and other protected groups. The Hub “1B params” style size reflects
50
  active MoE capacity; the marketing name refers to the upstream 3B MoE family.
51
 
 
 
 
 
 
 
 
 
 
52
  Immutable upstream source revision: `ee63731b6461c8afcdcc7b15352e7d2ffecc2ead`.
53
  Reference checkpoint revision: `55b8031a6c867de675279d9604e38cc94b9882a4`.
54
 
@@ -61,11 +57,15 @@ CER, digit CER, and table structure. Eligible language linear layers use MXFP8
61
  MXFP8 missed the CER ceiling. OCR-critical or sensitive groups remain BF16.
62
  Retained groups: vision_encoder, vision_projector.
63
 
64
- The model uses mlx-vlm's native `model_type=unlimited-ocr` implementation,
65
  `processor_class=UnlimitedOCRHFProcessor`, and a sliding-window size of
66
  128. Prompts should be raw Unlimited-OCR strings such as
67
  `<image>document parsing.` (not multi-turn chat formatting).
68
 
 
 
 
 
69
  ## AutomatosX validation
70
 
71
  These are **local release measurements**, not upstream paper benchmark claims
@@ -88,9 +88,11 @@ CJK, and 3 with tables), digest
88
  - Small local set; some samples are synthetic. Do not treat mean CER as field accuracy.
89
  - The community reference may use a different `model_type` / load path; it is a
90
  throughput and packaging baseline, not an identical-recipe peer.
91
- - Generation profiles used in AutomatosX tooling live in the `unlimited-ocr`
92
- package (`profiles.py`); weight-only loads use mlx-vlm defaults unless you
93
- pass the same decoding settings.
 
 
94
 
95
  ### Weight integrity
96
 
@@ -132,25 +134,75 @@ Full per-sample evidence, sensitivity results, precision rules, PRD, ADR, and
132
  technical specification are included under `release/`, `quantization/`, and
133
  `docs/`.
134
 
135
- ## Usage
 
 
 
 
 
136
 
137
- Requires **Apple Silicon** and **mlx-vlm ≥ 0.6.4** (native Unlimited-OCR / R-SWA).
138
- Tested stack for this release: mlx-vlm `0.6.6`, mlx `0.32.0`.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
 
140
  ```bash
141
  pip install 'mlx-vlm>=0.6.4'
142
  python -m mlx_vlm.generate \
143
  --model AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 \
 
144
  --image document.png \
145
  --prompt '<image>document parsing.' \
146
  --max-tokens 4096
147
  ```
148
 
149
- Or with the AutomatosX toolkit (deskew, PDF, profiles):
150
 
151
  ```bash
152
- pip install 'unlimited-ocr[mlx]'
153
- # then use OCRPipeline / the unlimited-ocr CLI with model_path=AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8
154
  ```
155
 
156
  This is quantization/conversion work, not foundation-model retraining or domain
 
1
  ---
2
  license: mit
3
  base_model: baidu/Unlimited-OCR
4
+ library_name: mlx
5
  pipeline_tag: image-text-to-text
6
  tags:
7
+ - ax-engine
8
  - mlx
9
  - mlx-vlm
10
  - ocr
11
  - mxfp8
12
  - int8
13
  - apple-silicon
14
+ - unlimited-ocr
15
  widget:
16
  - text: "<image>document parsing."
17
  - text: "<image>Convert the document to markdown."
18
  - text: "<image>Free OCR. Extract all text."
19
  ---
20
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
21
  # AX Unlimited-OCR 3B MoE — MLX (MXFP8 + affine INT8 head)
22
 
23
  This is an AutomatosX sensitivity-guided MLX conversion of
 
27
  checkpoint was used only as a comparison reference; these are not repackaged
28
  Sahil weights.
29
 
30
+ **Primary inference engine:** [AX Engine](https://github.com/defai-digital/ax-engine)
31
+ (`model_family=unlimited_ocr`). Use it for Apple Silicon serving and native
32
+ MXFP8 / dual-vision loads. `mlx-vlm` remains a compatibility path.
33
+
34
  **Hybrid precision (name is shorthand):** MXFP8 (group size 32) on eligible
35
  language linears, **affine INT8** on the LM head, **BF16** on the vision
36
  encoder and other protected groups. The Hub “1B params” style size reflects
37
  active MoE capacity; the marketing name refers to the upstream 3B MoE family.
38
 
39
+ ## Release
40
+
41
+ - **Version tag:** `v0.2.0` (immutable Hub git tag; pin this in production)
42
+ - **Hub pin:** `revision="v0.2.0"` or the commit SHA after publish
43
+ - **Repo:** [`AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8`](https://huggingface.co/AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8)
44
+
45
+ **Release notes:** Sensitivity-guided MXFP8 with calibrated affine INT8 LM head; primary inference via AX Engine
46
+ - **Runtime:** [AX Engine](https://github.com/defai-digital/ax-engine)
47
+
48
  Immutable upstream source revision: `ee63731b6461c8afcdcc7b15352e7d2ffecc2ead`.
49
  Reference checkpoint revision: `55b8031a6c867de675279d9604e38cc94b9882a4`.
50
 
 
57
  MXFP8 missed the CER ceiling. OCR-critical or sensitive groups remain BF16.
58
  Retained groups: vision_encoder, vision_projector.
59
 
60
+ The checkpoint declares `model_type=unlimited-ocr`,
61
  `processor_class=UnlimitedOCRHFProcessor`, and a sliding-window size of
62
  128. Prompts should be raw Unlimited-OCR strings such as
63
  `<image>document parsing.` (not multi-turn chat formatting).
64
 
65
+ AX Engine maps this pack to native family `unlimited_ocr` (dual vision
66
+ SAM-ViT-B + CLIP-L, SWA MoE language tower, MXFP8 dense/expert packs, R-SWA
67
+ prefill). Prefer the shipped `model-manifest.json` for AX-ready loads.
68
+
69
  ## AutomatosX validation
70
 
71
  These are **local release measurements**, not upstream paper benchmark claims
 
88
  - Small local set; some samples are synthetic. Do not treat mean CER as field accuracy.
89
  - The community reference may use a different `model_type` / load path; it is a
90
  throughput and packaging baseline, not an identical-recipe peer.
91
+ - AX Engine is the primary runtime for AutomatosX packs. Document toolkits such
92
+ as AX-OCR may still call mlx-vlm for PDF/profile helpers; pin the same Hub
93
+ revision when comparing stacks.
94
+ - Generation profiles in AX-OCR live in `profiles.py`; raw mlx-vlm loads use
95
+ library defaults unless you pass matching decode settings.
96
 
97
  ### Weight integrity
98
 
 
134
  technical specification are included under `release/`, `quantization/`, and
135
  `docs/`.
136
 
137
+ ## Usage (primary: AX Engine)
138
+
139
+ Requires **Apple Silicon**. Primary runtime:
140
+ [AX Engine](https://github.com/defai-digital/ax-engine) (native `unlimited_ocr`).
141
+
142
+ Tested conversion/validation stack for this weight release: mlx-vlm `0.6.6`, mlx `0.32.0`, python `3.14.6`.
143
 
144
+ ### Install AX Engine
145
+
146
+ ```bash
147
+ # Homebrew (recommended on macOS)
148
+ brew tap defai-digital/ax-engine
149
+ brew install defai-digital/ax-engine/ax-engine
150
+ ax-engine doctor
151
+
152
+ # or Python wheel
153
+ python3 -m pip install --upgrade "ax-engine[download]>=6.11.0,<7"
154
+ ```
155
+
156
+ ### Download and prepare
157
+
158
+ ```bash
159
+ # Managed download (writes model-manifest.json when needed)
160
+ ax-engine download AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8
161
+
162
+ # Pin a release tag (recommended for production)
163
+ hf download AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 \
164
+ --revision v0.2.0 \
165
+ --local-dir ./AX-Unlimited-OCR-3B-MoE-MLX-MXFP8
166
+
167
+ # Ensure AX native manifest is present/valid
168
+ ax-engine-bench generate-manifest ./AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 --validate
169
+ ```
170
+
171
+ This repository ships `model-manifest.json` so downloads are AX-ready without a
172
+ manual convert step when the file is present.
173
+
174
+ ### Serve with AX Engine
175
+
176
+ ```bash
177
+ ax-engine serve ./AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 --port 31418
178
+ # or, after managed download of the Hub id:
179
+ # ax-engine serve AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 --download --port 31418
180
+ ```
181
+
182
+ OpenAI-compatible routes are exposed on the local server. See the
183
+ [AX Engine docs](https://github.com/defai-digital/ax-engine/blob/main/docs/GETTING-STARTED.md) for
184
+ install, doctor checks, and API details. Unlimited-OCR image paths use the
185
+ native dual-vision prefill in AX Engine (`model_family=unlimited_ocr`).
186
+
187
+ ## Compatibility paths
188
+
189
+ ### mlx-vlm (weight-compatible)
190
 
191
  ```bash
192
  pip install 'mlx-vlm>=0.6.4'
193
  python -m mlx_vlm.generate \
194
  --model AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8 \
195
+ --revision v0.2.0 \
196
  --image document.png \
197
  --prompt '<image>document parsing.' \
198
  --max-tokens 4096
199
  ```
200
 
201
+ ### AX-OCR toolkit (deskew, PDF, profiles)
202
 
203
  ```bash
204
+ pip install 'ax-ocr[mlx]'
205
+ # OCRPipeline / ax-ocr CLI: model_path=AutomatosX/AX-Unlimited-OCR-3B-MoE-MLX-MXFP8, revision=v0.2.0
206
  ```
207
 
208
  This is quantization/conversion work, not foundation-model retraining or domain
model-manifest.json ADDED
@@ -0,0 +1,2798 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "ax.native_model.v1",
3
+ "model_family": "unlimited_ocr",
4
+ "tensor_format": "safetensors",
5
+ "layer_count": 12,
6
+ "hidden_size": 1280,
7
+ "intermediate_size": 6848,
8
+ "attention_head_count": 10,
9
+ "attention_head_dim": 128,
10
+ "kv_head_count": 10,
11
+ "vocab_size": 129280,
12
+ "tie_word_embeddings": false,
13
+ "no_rope_layer_interval": 0,
14
+ "intermediate_size_mlp": 0,
15
+ "attn_output_gate": false,
16
+ "rms_norm_eps": 0.000001,
17
+ "moe_norm_topk_prob": false,
18
+ "hidden_size_per_layer_input": 0,
19
+ "moe": {
20
+ "expert_count": 64,
21
+ "experts_per_token": 6,
22
+ "expert_intermediate_size": 896,
23
+ "layer_freq": 1,
24
+ "first_dense_layers": 1,
25
+ "shared_expert_count": 2,
26
+ "sigmoid_routing": false
27
+ },
28
+ "tensors": [
29
+ {
30
+ "name": "language_model.model.norm.weight",
31
+ "role": "final_norm",
32
+ "dtype": "bf16",
33
+ "shape": [
34
+ 1280
35
+ ],
36
+ "file": "model.safetensors",
37
+ "offset_bytes": 175916953,
38
+ "length_bytes": 2560
39
+ },
40
+ {
41
+ "name": "language_model.lm_head.weight",
42
+ "role": "lm_head",
43
+ "dtype": "u32",
44
+ "source_quantized": true,
45
+ "quantization": {
46
+ "mode": "affine",
47
+ "group_size": 32,
48
+ "bits": 8
49
+ },
50
+ "shape": [
51
+ 129280,
52
+ 320
53
+ ],
54
+ "file": "model.safetensors",
55
+ "offset_bytes": 10438553,
56
+ "length_bytes": 165478400
57
+ },
58
+ {
59
+ "name": "language_model.model.embed_tokens.weight",
60
+ "role": "token_embedding",
61
+ "dtype": "u32",
62
+ "source_quantized": true,
63
+ "quantization": {
64
+ "mode": "mxfp8",
65
+ "group_size": 32,
66
+ "bits": 8
67
+ },
68
+ "shape": [
69
+ 129280,
70
+ 320
71
+ ],
72
+ "file": "model.safetensors",
73
+ "offset_bytes": 1756151193,
74
+ "length_bytes": 165478400
75
+ },
76
+ {
77
+ "name": "image_newline",
78
+ "role": "unlimited_ocr_image_newline",
79
+ "dtype": "bf16",
80
+ "shape": [
81
+ 1280
82
+ ],
83
+ "file": "model.safetensors",
84
+ "offset_bytes": 93593,
85
+ "length_bytes": 2560
86
+ },
87
+ {
88
+ "name": "projector.layers.weight",
89
+ "role": "unlimited_ocr_projector",
90
+ "dtype": "bf16",
91
+ "shape": [
92
+ 1280,
93
+ 2048
94
+ ],
95
+ "file": "model.safetensors",
96
+ "offset_bytes": 2562722713,
97
+ "length_bytes": 5242880
98
+ },
99
+ {
100
+ "name": "view_separator",
101
+ "role": "unlimited_ocr_view_separator",
102
+ "dtype": "bf16",
103
+ "shape": [
104
+ 1280
105
+ ],
106
+ "file": "model.safetensors",
107
+ "offset_bytes": 3039130137,
108
+ "length_bytes": 2560
109
+ },
110
+ {
111
+ "name": "language_model.model.layers.0.self_attn.k_proj.weight",
112
+ "role": "attention_k",
113
+ "layer_index": 0,
114
+ "dtype": "u32",
115
+ "source_quantized": true,
116
+ "quantization": {
117
+ "mode": "mxfp8",
118
+ "group_size": 32,
119
+ "bits": 8
120
+ },
121
+ "shape": [
122
+ 1280,
123
+ 320
124
+ ],
125
+ "file": "model.safetensors",
126
+ "offset_bytes": 2350949657,
127
+ "length_bytes": 1638400
128
+ },
129
+ {
130
+ "name": "language_model.model.layers.0.input_layernorm.weight",
131
+ "role": "attention_norm",
132
+ "layer_index": 0,
133
+ "dtype": "bf16",
134
+ "shape": [
135
+ 1280
136
+ ],
137
+ "file": "model.safetensors",
138
+ "offset_bytes": 1718457753,
139
+ "length_bytes": 2560
140
+ },
141
+ {
142
+ "name": "language_model.model.layers.0.self_attn.o_proj.weight",
143
+ "role": "attention_o",
144
+ "layer_index": 0,
145
+ "dtype": "u32",
146
+ "source_quantized": true,
147
+ "quantization": {
148
+ "mode": "mxfp8",
149
+ "group_size": 32,
150
+ "bits": 8
151
+ },
152
+ "shape": [
153
+ 1280,
154
+ 320
155
+ ],
156
+ "file": "model.safetensors",
157
+ "offset_bytes": 3606229785,
158
+ "length_bytes": 1638400
159
+ },
160
+ {
161
+ "name": "language_model.model.layers.0.post_attention_layernorm.weight",
162
+ "role": "attention_post_norm",
163
+ "layer_index": 0,
164
+ "dtype": "bf16",
165
+ "shape": [
166
+ 1280
167
+ ],
168
+ "file": "model.safetensors",
169
+ "offset_bytes": 1718455193,
170
+ "length_bytes": 2560
171
+ },
172
+ {
173
+ "name": "language_model.model.layers.0.self_attn.q_proj.weight",
174
+ "role": "attention_q",
175
+ "layer_index": 0,
176
+ "dtype": "u32",
177
+ "source_quantized": true,
178
+ "quantization": {
179
+ "mode": "mxfp8",
180
+ "group_size": 32,
181
+ "bits": 8
182
+ },
183
+ "shape": [
184
+ 1280,
185
+ 320
186
+ ],
187
+ "file": "model.safetensors",
188
+ "offset_bytes": 3342692889,
189
+ "length_bytes": 1638400
190
+ },
191
+ {
192
+ "name": "language_model.model.layers.0.self_attn.v_proj.weight",
193
+ "role": "attention_v",
194
+ "layer_index": 0,
195
+ "dtype": "u32",
196
+ "source_quantized": true,
197
+ "quantization": {
198
+ "mode": "mxfp8",
199
+ "group_size": 32,
200
+ "bits": 8
201
+ },
202
+ "shape": [
203
+ 1280,
204
+ 320
205
+ ],
206
+ "file": "model.safetensors",
207
+ "offset_bytes": 3027135001,
208
+ "length_bytes": 1638400
209
+ },
210
+ {
211
+ "name": "language_model.model.layers.0.mlp.down_proj.weight",
212
+ "role": "ffn_down",
213
+ "layer_index": 0,
214
+ "dtype": "u32",
215
+ "source_quantized": true,
216
+ "quantization": {
217
+ "mode": "mxfp8",
218
+ "group_size": 32,
219
+ "bits": 8
220
+ },
221
+ "shape": [
222
+ 1280,
223
+ 1712
224
+ ],
225
+ "file": "model.safetensors",
226
+ "offset_bytes": 1720372633,
227
+ "length_bytes": 8765440
228
+ },
229
+ {
230
+ "name": "language_model.model.layers.0.mlp.gate_proj.weight",
231
+ "role": "ffn_gate",
232
+ "layer_index": 0,
233
+ "dtype": "u32",
234
+ "source_quantized": true,
235
+ "quantization": {
236
+ "mode": "mxfp8",
237
+ "group_size": 32,
238
+ "bits": 8
239
+ },
240
+ "shape": [
241
+ 6848,
242
+ 320
243
+ ],
244
+ "file": "model.safetensors",
245
+ "offset_bytes": 1742109593,
246
+ "length_bytes": 8765440
247
+ },
248
+ {
249
+ "name": "language_model.model.layers.0.mlp.up_proj.weight",
250
+ "role": "ffn_up",
251
+ "layer_index": 0,
252
+ "dtype": "u32",
253
+ "source_quantized": true,
254
+ "quantization": {
255
+ "mode": "mxfp8",
256
+ "group_size": 32,
257
+ "bits": 8
258
+ },
259
+ "shape": [
260
+ 6848,
261
+ 320
262
+ ],
263
+ "file": "model.safetensors",
264
+ "offset_bytes": 1733344153,
265
+ "length_bytes": 8765440
266
+ },
267
+ {
268
+ "name": "language_model.model.layers.1.self_attn.k_proj.weight",
269
+ "role": "attention_k",
270
+ "layer_index": 1,
271
+ "dtype": "u32",
272
+ "source_quantized": true,
273
+ "quantization": {
274
+ "mode": "mxfp8",
275
+ "group_size": 32,
276
+ "bits": 8
277
+ },
278
+ "shape": [
279
+ 1280,
280
+ 320
281
+ ],
282
+ "file": "model.safetensors",
283
+ "offset_bytes": 2181890585,
284
+ "length_bytes": 1638400
285
+ },
286
+ {
287
+ "name": "language_model.model.layers.1.input_layernorm.weight",
288
+ "role": "attention_norm",
289
+ "layer_index": 1,
290
+ "dtype": "bf16",
291
+ "shape": [
292
+ 1280
293
+ ],
294
+ "file": "model.safetensors",
295
+ "offset_bytes": 1485049753,
296
+ "length_bytes": 2560
297
+ },
298
+ {
299
+ "name": "language_model.model.layers.1.self_attn.o_proj.weight",
300
+ "role": "attention_o",
301
+ "layer_index": 1,
302
+ "dtype": "u32",
303
+ "source_quantized": true,
304
+ "quantization": {
305
+ "mode": "mxfp8",
306
+ "group_size": 32,
307
+ "bits": 8
308
+ },
309
+ "shape": [
310
+ 1280,
311
+ 320
312
+ ],
313
+ "file": "model.safetensors",
314
+ "offset_bytes": 1718460313,
315
+ "length_bytes": 1638400
316
+ },
317
+ {
318
+ "name": "language_model.model.layers.1.post_attention_layernorm.weight",
319
+ "role": "attention_post_norm",
320
+ "layer_index": 1,
321
+ "dtype": "bf16",
322
+ "shape": [
323
+ 1280
324
+ ],
325
+ "file": "model.safetensors",
326
+ "offset_bytes": 3319586969,
327
+ "length_bytes": 2560
328
+ },
329
+ {
330
+ "name": "language_model.model.layers.1.self_attn.q_proj.weight",
331
+ "role": "attention_q",
332
+ "layer_index": 1,
333
+ "dtype": "u32",
334
+ "source_quantized": true,
335
+ "quantization": {
336
+ "mode": "mxfp8",
337
+ "group_size": 32,
338
+ "bits": 8
339
+ },
340
+ "shape": [
341
+ 1280,
342
+ 320
343
+ ],
344
+ "file": "model.safetensors",
345
+ "offset_bytes": 1716816793,
346
+ "length_bytes": 1638400
347
+ },
348
+ {
349
+ "name": "language_model.model.layers.1.self_attn.v_proj.weight",
350
+ "role": "attention_v",
351
+ "layer_index": 1,
352
+ "dtype": "u32",
353
+ "source_quantized": true,
354
+ "quantization": {
355
+ "mode": "mxfp8",
356
+ "group_size": 32,
357
+ "bits": 8
358
+ },
359
+ "shape": [
360
+ 1280,
361
+ 320
362
+ ],
363
+ "file": "model.safetensors",
364
+ "offset_bytes": 2808163225,
365
+ "length_bytes": 1638400
366
+ },
367
+ {
368
+ "name": "language_model.model.layers.1.mlp.switch_mlp.down_proj.weight",
369
+ "role": "ffn_down_exps",
370
+ "layer_index": 1,
371
+ "dtype": "u32",
372
+ "source_quantized": true,
373
+ "quantization": {
374
+ "mode": "mxfp8",
375
+ "group_size": 32,
376
+ "bits": 8
377
+ },
378
+ "shape": [
379
+ 64,
380
+ 1280,
381
+ 224
382
+ ],
383
+ "file": "model.safetensors",
384
+ "offset_bytes": 1565059993,
385
+ "length_bytes": 73400320
386
+ },
387
+ {
388
+ "name": "language_model.model.layers.1.mlp.switch_mlp.gate_proj.weight",
389
+ "role": "ffn_gate_exps",
390
+ "layer_index": 1,
391
+ "dtype": "u32",
392
+ "source_quantized": true,
393
+ "quantization": {
394
+ "mode": "mxfp8",
395
+ "group_size": 32,
396
+ "bits": 8
397
+ },
398
+ "shape": [
399
+ 64,
400
+ 896,
401
+ 320
402
+ ],
403
+ "file": "model.safetensors",
404
+ "offset_bytes": 3105961241,
405
+ "length_bytes": 73400320
406
+ },
407
+ {
408
+ "name": "language_model.model.layers.1.mlp.gate.weight",
409
+ "role": "ffn_gate_inp",
410
+ "layer_index": 1,
411
+ "dtype": "bf16",
412
+ "shape": [
413
+ 64,
414
+ 1280
415
+ ],
416
+ "file": "model.safetensors",
417
+ "offset_bytes": 1489202073,
418
+ "length_bytes": 163840
419
+ },
420
+ {
421
+ "name": "language_model.model.layers.1.mlp.shared_experts.down_proj.weight",
422
+ "role": "ffn_shared_expert_down",
423
+ "layer_index": 1,
424
+ "dtype": "u32",
425
+ "source_quantized": true,
426
+ "quantization": {
427
+ "mode": "mxfp8",
428
+ "group_size": 32,
429
+ "bits": 8
430
+ },
431
+ "shape": [
432
+ 1280,
433
+ 448
434
+ ],
435
+ "file": "model.safetensors",
436
+ "offset_bytes": 3561485593,
437
+ "length_bytes": 2293760
438
+ },
439
+ {
440
+ "name": "language_model.model.layers.1.mlp.shared_experts.gate_proj.weight",
441
+ "role": "ffn_shared_expert_gate",
442
+ "layer_index": 1,
443
+ "dtype": "u32",
444
+ "source_quantized": true,
445
+ "quantization": {
446
+ "mode": "mxfp8",
447
+ "group_size": 32,
448
+ "bits": 8
449
+ },
450
+ "shape": [
451
+ 1792,
452
+ 320
453
+ ],
454
+ "file": "model.safetensors",
455
+ "offset_bytes": 1485269913,
456
+ "length_bytes": 2293760
457
+ },
458
+ {
459
+ "name": "language_model.model.layers.1.mlp.shared_experts.up_proj.weight",
460
+ "role": "ffn_shared_expert_up",
461
+ "layer_index": 1,
462
+ "dtype": "u32",
463
+ "source_quantized": true,
464
+ "quantization": {
465
+ "mode": "mxfp8",
466
+ "group_size": 32,
467
+ "bits": 8
468
+ },
469
+ "shape": [
470
+ 1792,
471
+ 320
472
+ ],
473
+ "file": "model.safetensors",
474
+ "offset_bytes": 3815614873,
475
+ "length_bytes": 2293760
476
+ },
477
+ {
478
+ "name": "language_model.model.layers.1.mlp.switch_mlp.up_proj.weight",
479
+ "role": "ffn_up_exps",
480
+ "layer_index": 1,
481
+ "dtype": "u32",
482
+ "source_quantized": true,
483
+ "quantization": {
484
+ "mode": "mxfp8",
485
+ "group_size": 32,
486
+ "bits": 8
487
+ },
488
+ "shape": [
489
+ 64,
490
+ 896,
491
+ 320
492
+ ],
493
+ "file": "model.safetensors",
494
+ "offset_bytes": 1638460313,
495
+ "length_bytes": 73400320
496
+ },
497
+ {
498
+ "name": "language_model.model.layers.2.self_attn.k_proj.weight",
499
+ "role": "attention_k",
500
+ "layer_index": 2,
501
+ "dtype": "u32",
502
+ "source_quantized": true,
503
+ "quantization": {
504
+ "mode": "mxfp8",
505
+ "group_size": 32,
506
+ "bits": 8
507
+ },
508
+ "shape": [
509
+ 1280,
510
+ 320
511
+ ],
512
+ "file": "model.safetensors",
513
+ "offset_bytes": 3352739865,
514
+ "length_bytes": 1638400
515
+ },
516
+ {
517
+ "name": "language_model.model.layers.2.input_layernorm.weight",
518
+ "role": "attention_norm",
519
+ "layer_index": 2,
520
+ "dtype": "bf16",
521
+ "shape": [
522
+ 1280
523
+ ],
524
+ "file": "model.safetensors",
525
+ "offset_bytes": 1317443993,
526
+ "length_bytes": 2560
527
+ },
528
+ {
529
+ "name": "language_model.model.layers.2.self_attn.o_proj.weight",
530
+ "role": "attention_o",
531
+ "layer_index": 2,
532
+ "dtype": "u32",
533
+ "source_quantized": true,
534
+ "quantization": {
535
+ "mode": "mxfp8",
536
+ "group_size": 32,
537
+ "bits": 8
538
+ },
539
+ "shape": [
540
+ 1280,
541
+ 320
542
+ ],
543
+ "file": "model.safetensors",
544
+ "offset_bytes": 1478391193,
545
+ "length_bytes": 1638400
546
+ },
547
+ {
548
+ "name": "language_model.model.layers.2.post_attention_layernorm.weight",
549
+ "role": "attention_post_norm",
550
+ "layer_index": 2,
551
+ "dtype": "bf16",
552
+ "shape": [
553
+ 1280
554
+ ],
555
+ "file": "model.safetensors",
556
+ "offset_bytes": 2723929625,
557
+ "length_bytes": 2560
558
+ },
559
+ {
560
+ "name": "language_model.model.layers.2.self_attn.q_proj.weight",
561
+ "role": "attention_q",
562
+ "layer_index": 2,
563
+ "dtype": "u32",
564
+ "source_quantized": true,
565
+ "quantization": {
566
+ "mode": "mxfp8",
567
+ "group_size": 32,
568
+ "bits": 8
569
+ },
570
+ "shape": [
571
+ 1280,
572
+ 320
573
+ ],
574
+ "file": "model.safetensors",
575
+ "offset_bytes": 1483408793,
576
+ "length_bytes": 1638400
577
+ },
578
+ {
579
+ "name": "language_model.model.layers.2.self_attn.v_proj.weight",
580
+ "role": "attention_v",
581
+ "layer_index": 2,
582
+ "dtype": "u32",
583
+ "source_quantized": true,
584
+ "quantization": {
585
+ "mode": "mxfp8",
586
+ "group_size": 32,
587
+ "bits": 8
588
+ },
589
+ "shape": [
590
+ 1280,
591
+ 320
592
+ ],
593
+ "file": "model.safetensors",
594
+ "offset_bytes": 1481719193,
595
+ "length_bytes": 1638400
596
+ },
597
+ {
598
+ "name": "language_model.model.layers.2.mlp.switch_mlp.down_proj.weight",
599
+ "role": "ffn_down_exps",
600
+ "layer_index": 2,
601
+ "dtype": "u32",
602
+ "source_quantized": true,
603
+ "quantization": {
604
+ "mode": "mxfp8",
605
+ "group_size": 32,
606
+ "bits": 8
607
+ },
608
+ "shape": [
609
+ 64,
610
+ 1280,
611
+ 224
612
+ ],
613
+ "file": "model.safetensors",
614
+ "offset_bytes": 2943153689,
615
+ "length_bytes": 73400320
616
+ },
617
+ {
618
+ "name": "language_model.model.layers.2.mlp.switch_mlp.gate_proj.weight",
619
+ "role": "ffn_gate_exps",
620
+ "layer_index": 2,
621
+ "dtype": "u32",
622
+ "source_quantized": true,
623
+ "quantization": {
624
+ "mode": "mxfp8",
625
+ "group_size": 32,
626
+ "bits": 8
627
+ },
628
+ "shape": [
629
+ 64,
630
+ 896,
631
+ 320
632
+ ],
633
+ "file": "model.safetensors",
634
+ "offset_bytes": 2369328921,
635
+ "length_bytes": 73400320
636
+ },
637
+ {
638
+ "name": "language_model.model.layers.2.mlp.gate.weight",
639
+ "role": "ffn_gate_inp",
640
+ "layer_index": 2,
641
+ "dtype": "bf16",
642
+ "shape": [
643
+ 64,
644
+ 1280
645
+ ],
646
+ "file": "model.safetensors",
647
+ "offset_bytes": 1478227353,
648
+ "length_bytes": 163840
649
+ },
650
+ {
651
+ "name": "language_model.model.layers.2.mlp.shared_experts.down_proj.weight",
652
+ "role": "ffn_shared_expert_down",
653
+ "layer_index": 2,
654
+ "dtype": "u32",
655
+ "source_quantized": true,
656
+ "quantization": {
657
+ "mode": "mxfp8",
658
+ "group_size": 32,
659
+ "bits": 8
660
+ },
661
+ "shape": [
662
+ 1280,
663
+ 448
664
+ ],
665
+ "file": "model.safetensors",
666
+ "offset_bytes": 1317520793,
667
+ "length_bytes": 2293760
668
+ },
669
+ {
670
+ "name": "language_model.model.layers.2.mlp.shared_experts.gate_proj.weight",
671
+ "role": "ffn_shared_expert_gate",
672
+ "layer_index": 2,
673
+ "dtype": "u32",
674
+ "source_quantized": true,
675
+ "quantization": {
676
+ "mode": "mxfp8",
677
+ "group_size": 32,
678
+ "bits": 8
679
+ },
680
+ "shape": [
681
+ 1792,
682
+ 320
683
+ ],
684
+ "file": "model.safetensors",
685
+ "offset_bytes": 1395580313,
686
+ "length_bytes": 2293760
687
+ },
688
+ {
689
+ "name": "language_model.model.layers.2.mlp.shared_experts.up_proj.weight",
690
+ "role": "ffn_shared_expert_up",
691
+ "layer_index": 2,
692
+ "dtype": "u32",
693
+ "source_quantized": true,
694
+ "quantization": {
695
+ "mode": "mxfp8",
696
+ "group_size": 32,
697
+ "bits": 8
698
+ },
699
+ "shape": [
700
+ 1792,
701
+ 320
702
+ ],
703
+ "file": "model.safetensors",
704
+ "offset_bytes": 1393286553,
705
+ "length_bytes": 2293760
706
+ },
707
+ {
708
+ "name": "language_model.model.layers.2.mlp.switch_mlp.up_proj.weight",
709
+ "role": "ffn_up_exps",
710
+ "layer_index": 2,
711
+ "dtype": "u32",
712
+ "source_quantized": true,
713
+ "quantization": {
714
+ "mode": "mxfp8",
715
+ "group_size": 32,
716
+ "bits": 8
717
+ },
718
+ "shape": [
719
+ 64,
720
+ 896,
721
+ 320
722
+ ],
723
+ "file": "model.safetensors",
724
+ "offset_bytes": 1402461593,
725
+ "length_bytes": 73400320
726
+ },
727
+ {
728
+ "name": "language_model.model.layers.3.self_attn.k_proj.weight",
729
+ "role": "attention_k",
730
+ "layer_index": 3,
731
+ "dtype": "u32",
732
+ "source_quantized": true,
733
+ "quantization": {
734
+ "mode": "mxfp8",
735
+ "group_size": 32,
736
+ "bits": 8
737
+ },
738
+ "shape": [
739
+ 1280,
740
+ 320
741
+ ],
742
+ "file": "model.safetensors",
743
+ "offset_bytes": 2607808665,
744
+ "length_bytes": 1638400
745
+ },
746
+ {
747
+ "name": "language_model.model.layers.3.input_layernorm.weight",
748
+ "role": "attention_norm",
749
+ "layer_index": 3,
750
+ "dtype": "bf16",
751
+ "shape": [
752
+ 1280
753
+ ],
754
+ "file": "model.safetensors",
755
+ "offset_bytes": 1159653273,
756
+ "length_bytes": 2560
757
+ },
758
+ {
759
+ "name": "language_model.model.layers.3.self_attn.o_proj.weight",
760
+ "role": "attention_o",
761
+ "layer_index": 3,
762
+ "dtype": "u32",
763
+ "source_quantized": true,
764
+ "quantization": {
765
+ "mode": "mxfp8",
766
+ "group_size": 32,
767
+ "bits": 8
768
+ },
769
+ "shape": [
770
+ 1280,
771
+ 320
772
+ ],
773
+ "file": "model.safetensors",
774
+ "offset_bytes": 1480029593,
775
+ "length_bytes": 1638400
776
+ },
777
+ {
778
+ "name": "language_model.model.layers.3.post_attention_layernorm.weight",
779
+ "role": "attention_post_norm",
780
+ "layer_index": 3,
781
+ "dtype": "bf16",
782
+ "shape": [
783
+ 1280
784
+ ],
785
+ "file": "model.safetensors",
786
+ "offset_bytes": 2633741849,
787
+ "length_bytes": 2560
788
+ },
789
+ {
790
+ "name": "language_model.model.layers.3.self_attn.q_proj.weight",
791
+ "role": "attention_q",
792
+ "layer_index": 3,
793
+ "dtype": "u32",
794
+ "source_quantized": true,
795
+ "quantization": {
796
+ "mode": "mxfp8",
797
+ "group_size": 32,
798
+ "bits": 8
799
+ },
800
+ "shape": [
801
+ 1280,
802
+ 320
803
+ ],
804
+ "file": "model.safetensors",
805
+ "offset_bytes": 1315805593,
806
+ "length_bytes": 1638400
807
+ },
808
+ {
809
+ "name": "language_model.model.layers.3.self_attn.v_proj.weight",
810
+ "role": "attention_v",
811
+ "layer_index": 3,
812
+ "dtype": "u32",
813
+ "source_quantized": true,
814
+ "quantization": {
815
+ "mode": "mxfp8",
816
+ "group_size": 32,
817
+ "bits": 8
818
+ },
819
+ "shape": [
820
+ 1280,
821
+ 320
822
+ ],
823
+ "file": "model.safetensors",
824
+ "offset_bytes": 2832770841,
825
+ "length_bytes": 1638400
826
+ },
827
+ {
828
+ "name": "language_model.model.layers.3.mlp.switch_mlp.down_proj.weight",
829
+ "role": "ffn_down_exps",
830
+ "layer_index": 3,
831
+ "dtype": "u32",
832
+ "source_quantized": true,
833
+ "quantization": {
834
+ "mode": "mxfp8",
835
+ "group_size": 32,
836
+ "bits": 8
837
+ },
838
+ "shape": [
839
+ 64,
840
+ 1280,
841
+ 224
842
+ ],
843
+ "file": "model.safetensors",
844
+ "offset_bytes": 2258578329,
845
+ "length_bytes": 73400320
846
+ },
847
+ {
848
+ "name": "language_model.model.layers.3.mlp.switch_mlp.gate_proj.weight",
849
+ "role": "ffn_gate_exps",
850
+ "layer_index": 3,
851
+ "dtype": "u32",
852
+ "source_quantized": true,
853
+ "quantization": {
854
+ "mode": "mxfp8",
855
+ "group_size": 32,
856
+ "bits": 8
857
+ },
858
+ "shape": [
859
+ 64,
860
+ 896,
861
+ 320
862
+ ],
863
+ "file": "model.safetensors",
864
+ "offset_bytes": 3225070361,
865
+ "length_bytes": 73400320
866
+ },
867
+ {
868
+ "name": "language_model.model.layers.3.mlp.gate.weight",
869
+ "role": "ffn_gate_inp",
870
+ "layer_index": 3,
871
+ "dtype": "bf16",
872
+ "shape": [
873
+ 64,
874
+ 1280
875
+ ],
876
+ "file": "model.safetensors",
877
+ "offset_bytes": 1716601753,
878
+ "length_bytes": 163840
879
+ },
880
+ {
881
+ "name": "language_model.model.layers.3.mlp.shared_experts.down_proj.weight",
882
+ "role": "ffn_shared_expert_down",
883
+ "layer_index": 3,
884
+ "dtype": "u32",
885
+ "source_quantized": true,
886
+ "quantization": {
887
+ "mode": "mxfp8",
888
+ "group_size": 32,
889
+ "bits": 8
890
+ },
891
+ "shape": [
892
+ 1280,
893
+ 448
894
+ ],
895
+ "file": "model.safetensors",
896
+ "offset_bytes": 1159727513,
897
+ "length_bytes": 2293760
898
+ },
899
+ {
900
+ "name": "language_model.model.layers.3.mlp.shared_experts.gate_proj.weight",
901
+ "role": "ffn_shared_expert_gate",
902
+ "layer_index": 3,
903
+ "dtype": "u32",
904
+ "source_quantized": true,
905
+ "quantization": {
906
+ "mode": "mxfp8",
907
+ "group_size": 32,
908
+ "bits": 8
909
+ },
910
+ "shape": [
911
+ 1792,
912
+ 320
913
+ ],
914
+ "file": "model.safetensors",
915
+ "offset_bytes": 1164315033,
916
+ "length_bytes": 2293760
917
+ },
918
+ {
919
+ "name": "language_model.model.layers.3.mlp.shared_experts.up_proj.weight",
920
+ "role": "ffn_shared_expert_up",
921
+ "layer_index": 3,
922
+ "dtype": "u32",
923
+ "source_quantized": true,
924
+ "quantization": {
925
+ "mode": "mxfp8",
926
+ "group_size": 32,
927
+ "bits": 8
928
+ },
929
+ "shape": [
930
+ 1792,
931
+ 320
932
+ ],
933
+ "file": "model.safetensors",
934
+ "offset_bytes": 1162021273,
935
+ "length_bytes": 2293760
936
+ },
937
+ {
938
+ "name": "language_model.model.layers.3.mlp.switch_mlp.up_proj.weight",
939
+ "role": "ffn_up_exps",
940
+ "layer_index": 3,
941
+ "dtype": "u32",
942
+ "source_quantized": true,
943
+ "quantization": {
944
+ "mode": "mxfp8",
945
+ "group_size": 32,
946
+ "bits": 8
947
+ },
948
+ "shape": [
949
+ 64,
950
+ 896,
951
+ 320
952
+ ],
953
+ "file": "model.safetensors",
954
+ "offset_bytes": 1166608793,
955
+ "length_bytes": 73400320
956
+ },
957
+ {
958
+ "name": "language_model.model.layers.4.self_attn.k_proj.weight",
959
+ "role": "attention_k",
960
+ "layer_index": 4,
961
+ "dtype": "u32",
962
+ "source_quantized": true,
963
+ "quantization": {
964
+ "mode": "mxfp8",
965
+ "group_size": 32,
966
+ "bits": 8
967
+ },
968
+ "shape": [
969
+ 1280,
970
+ 320
971
+ ],
972
+ "file": "model.safetensors",
973
+ "offset_bytes": 1156376473,
974
+ "length_bytes": 1638400
975
+ },
976
+ {
977
+ "name": "language_model.model.layers.4.input_layernorm.weight",
978
+ "role": "attention_norm",
979
+ "layer_index": 4,
980
+ "dtype": "bf16",
981
+ "shape": [
982
+ 1280
983
+ ],
984
+ "file": "model.safetensors",
985
+ "offset_bytes": 1317446553,
986
+ "length_bytes": 2560
987
+ },
988
+ {
989
+ "name": "language_model.model.layers.4.self_attn.o_proj.weight",
990
+ "role": "attention_o",
991
+ "layer_index": 4,
992
+ "dtype": "u32",
993
+ "source_quantized": true,
994
+ "quantization": {
995
+ "mode": "mxfp8",
996
+ "group_size": 32,
997
+ "bits": 8
998
+ },
999
+ "shape": [
1000
+ 1280,
1001
+ 320
1002
+ ],
1003
+ "file": "model.safetensors",
1004
+ "offset_bytes": 1154686873,
1005
+ "length_bytes": 1638400
1006
+ },
1007
+ {
1008
+ "name": "language_model.model.layers.4.post_attention_layernorm.weight",
1009
+ "role": "attention_post_norm",
1010
+ "layer_index": 4,
1011
+ "dtype": "bf16",
1012
+ "shape": [
1013
+ 1280
1014
+ ],
1015
+ "file": "model.safetensors",
1016
+ "offset_bytes": 2183535513,
1017
+ "length_bytes": 2560
1018
+ },
1019
+ {
1020
+ "name": "language_model.model.layers.4.self_attn.q_proj.weight",
1021
+ "role": "attention_q",
1022
+ "layer_index": 4,
1023
+ "dtype": "u32",
1024
+ "source_quantized": true,
1025
+ "quantization": {
1026
+ "mode": "mxfp8",
1027
+ "group_size": 32,
1028
+ "bits": 8
1029
+ },
1030
+ "shape": [
1031
+ 1280,
1032
+ 320
1033
+ ],
1034
+ "file": "model.safetensors",
1035
+ "offset_bytes": 1158014873,
1036
+ "length_bytes": 1638400
1037
+ },
1038
+ {
1039
+ "name": "language_model.model.layers.4.self_attn.v_proj.weight",
1040
+ "role": "attention_v",
1041
+ "layer_index": 4,
1042
+ "dtype": "u32",
1043
+ "source_quantized": true,
1044
+ "quantization": {
1045
+ "mode": "mxfp8",
1046
+ "group_size": 32,
1047
+ "bits": 8
1048
+ },
1049
+ "shape": [
1050
+ 1280,
1051
+ 320
1052
+ ],
1053
+ "file": "model.safetensors",
1054
+ "offset_bytes": 2935148569,
1055
+ "length_bytes": 1638400
1056
+ },
1057
+ {
1058
+ "name": "language_model.model.layers.4.mlp.switch_mlp.down_proj.weight",
1059
+ "role": "ffn_down_exps",
1060
+ "layer_index": 4,
1061
+ "dtype": "u32",
1062
+ "source_quantized": true,
1063
+ "quantization": {
1064
+ "mode": "mxfp8",
1065
+ "group_size": 32,
1066
+ "bits": 8
1067
+ },
1068
+ "shape": [
1069
+ 64,
1070
+ 1280,
1071
+ 224
1072
+ ],
1073
+ "file": "model.safetensors",
1074
+ "offset_bytes": 1930365721,
1075
+ "length_bytes": 73400320
1076
+ },
1077
+ {
1078
+ "name": "language_model.model.layers.4.mlp.switch_mlp.gate_proj.weight",
1079
+ "role": "ffn_gate_exps",
1080
+ "layer_index": 4,
1081
+ "dtype": "u32",
1082
+ "source_quantized": true,
1083
+ "quantization": {
1084
+ "mode": "mxfp8",
1085
+ "group_size": 32,
1086
+ "bits": 8
1087
+ },
1088
+ "shape": [
1089
+ 64,
1090
+ 896,
1091
+ 320
1092
+ ],
1093
+ "file": "model.safetensors",
1094
+ "offset_bytes": 2461558297,
1095
+ "length_bytes": 73400320
1096
+ },
1097
+ {
1098
+ "name": "language_model.model.layers.4.mlp.gate.weight",
1099
+ "role": "ffn_gate_inp",
1100
+ "layer_index": 4,
1101
+ "dtype": "bf16",
1102
+ "shape": [
1103
+ 64,
1104
+ 1280
1105
+ ],
1106
+ "file": "model.safetensors",
1107
+ "offset_bytes": 1076381593,
1108
+ "length_bytes": 163840
1109
+ },
1110
+ {
1111
+ "name": "language_model.model.layers.4.mlp.shared_experts.down_proj.weight",
1112
+ "role": "ffn_shared_expert_down",
1113
+ "layer_index": 4,
1114
+ "dtype": "u32",
1115
+ "source_quantized": true,
1116
+ "quantization": {
1117
+ "mode": "mxfp8",
1118
+ "group_size": 32,
1119
+ "bits": 8
1120
+ },
1121
+ "shape": [
1122
+ 1280,
1123
+ 448
1124
+ ],
1125
+ "file": "model.safetensors",
1126
+ "offset_bytes": 1069500313,
1127
+ "length_bytes": 2293760
1128
+ },
1129
+ {
1130
+ "name": "language_model.model.layers.4.mlp.shared_experts.gate_proj.weight",
1131
+ "role": "ffn_shared_expert_gate",
1132
+ "layer_index": 4,
1133
+ "dtype": "u32",
1134
+ "source_quantized": true,
1135
+ "quantization": {
1136
+ "mode": "mxfp8",
1137
+ "group_size": 32,
1138
+ "bits": 8
1139
+ },
1140
+ "shape": [
1141
+ 1792,
1142
+ 320
1143
+ ],
1144
+ "file": "model.safetensors",
1145
+ "offset_bytes": 1074087833,
1146
+ "length_bytes": 2293760
1147
+ },
1148
+ {
1149
+ "name": "language_model.model.layers.4.mlp.shared_experts.up_proj.weight",
1150
+ "role": "ffn_shared_expert_up",
1151
+ "layer_index": 4,
1152
+ "dtype": "u32",
1153
+ "source_quantized": true,
1154
+ "quantization": {
1155
+ "mode": "mxfp8",
1156
+ "group_size": 32,
1157
+ "bits": 8
1158
+ },
1159
+ "shape": [
1160
+ 1792,
1161
+ 320
1162
+ ],
1163
+ "file": "model.safetensors",
1164
+ "offset_bytes": 1071794073,
1165
+ "length_bytes": 2293760
1166
+ },
1167
+ {
1168
+ "name": "language_model.model.layers.4.mlp.switch_mlp.up_proj.weight",
1169
+ "role": "ffn_up_exps",
1170
+ "layer_index": 4,
1171
+ "dtype": "u32",
1172
+ "source_quantized": true,
1173
+ "quantization": {
1174
+ "mode": "mxfp8",
1175
+ "group_size": 32,
1176
+ "bits": 8
1177
+ },
1178
+ "shape": [
1179
+ 64,
1180
+ 896,
1181
+ 320
1182
+ ],
1183
+ "file": "model.safetensors",
1184
+ "offset_bytes": 1078890393,
1185
+ "length_bytes": 73400320
1186
+ },
1187
+ {
1188
+ "name": "language_model.model.layers.5.self_attn.k_proj.weight",
1189
+ "role": "attention_k",
1190
+ "layer_index": 5,
1191
+ "dtype": "u32",
1192
+ "source_quantized": true,
1193
+ "quantization": {
1194
+ "mode": "mxfp8",
1195
+ "group_size": 32,
1196
+ "bits": 8
1197
+ },
1198
+ "shape": [
1199
+ 1280,
1200
+ 320
1201
+ ],
1202
+ "file": "model.safetensors",
1203
+ "offset_bytes": 1067790233,
1204
+ "length_bytes": 1638400
1205
+ },
1206
+ {
1207
+ "name": "language_model.model.layers.5.input_layernorm.weight",
1208
+ "role": "attention_norm",
1209
+ "layer_index": 5,
1210
+ "dtype": "bf16",
1211
+ "shape": [
1212
+ 1280
1213
+ ],
1214
+ "file": "model.safetensors",
1215
+ "offset_bytes": 830506393,
1216
+ "length_bytes": 2560
1217
+ },
1218
+ {
1219
+ "name": "language_model.model.layers.5.self_attn.o_proj.weight",
1220
+ "role": "attention_o",
1221
+ "layer_index": 5,
1222
+ "dtype": "u32",
1223
+ "source_quantized": true,
1224
+ "quantization": {
1225
+ "mode": "mxfp8",
1226
+ "group_size": 32,
1227
+ "bits": 8
1228
+ },
1229
+ "shape": [
1230
+ 1280,
1231
+ 320
1232
+ ],
1233
+ "file": "model.safetensors",
1234
+ "offset_bytes": 991061913,
1235
+ "length_bytes": 1638400
1236
+ },
1237
+ {
1238
+ "name": "language_model.model.layers.5.post_attention_layernorm.weight",
1239
+ "role": "attention_post_norm",
1240
+ "layer_index": 5,
1241
+ "dtype": "bf16",
1242
+ "shape": [
1243
+ 1280
1244
+ ],
1245
+ "file": "model.safetensors",
1246
+ "offset_bytes": 830503833,
1247
+ "length_bytes": 2560
1248
+ },
1249
+ {
1250
+ "name": "language_model.model.layers.5.self_attn.q_proj.weight",
1251
+ "role": "attention_q",
1252
+ "layer_index": 5,
1253
+ "dtype": "u32",
1254
+ "source_quantized": true,
1255
+ "quantization": {
1256
+ "mode": "mxfp8",
1257
+ "group_size": 32,
1258
+ "bits": 8
1259
+ },
1260
+ "shape": [
1261
+ 1280,
1262
+ 320
1263
+ ],
1264
+ "file": "model.safetensors",
1265
+ "offset_bytes": 2014376217,
1266
+ "length_bytes": 1638400
1267
+ },
1268
+ {
1269
+ "name": "language_model.model.layers.5.self_attn.v_proj.weight",
1270
+ "role": "attention_v",
1271
+ "layer_index": 5,
1272
+ "dtype": "u32",
1273
+ "source_quantized": true,
1274
+ "quantization": {
1275
+ "mode": "mxfp8",
1276
+ "group_size": 32,
1277
+ "bits": 8
1278
+ },
1279
+ "shape": [
1280
+ 1280,
1281
+ 320
1282
+ ],
1283
+ "file": "model.safetensors",
1284
+ "offset_bytes": 1066151833,
1285
+ "length_bytes": 1638400
1286
+ },
1287
+ {
1288
+ "name": "language_model.model.layers.5.mlp.switch_mlp.down_proj.weight",
1289
+ "role": "ffn_down_exps",
1290
+ "layer_index": 5,
1291
+ "dtype": "u32",
1292
+ "source_quantized": true,
1293
+ "quantization": {
1294
+ "mode": "mxfp8",
1295
+ "group_size": 32,
1296
+ "bits": 8
1297
+ },
1298
+ "shape": [
1299
+ 64,
1300
+ 1280,
1301
+ 224
1302
+ ],
1303
+ "file": "model.safetensors",
1304
+ "offset_bytes": 837984153,
1305
+ "length_bytes": 73400320
1306
+ },
1307
+ {
1308
+ "name": "language_model.model.layers.5.mlp.switch_mlp.gate_proj.weight",
1309
+ "role": "ffn_gate_exps",
1310
+ "layer_index": 5,
1311
+ "dtype": "u32",
1312
+ "source_quantized": true,
1313
+ "quantization": {
1314
+ "mode": "mxfp8",
1315
+ "group_size": 32,
1316
+ "bits": 8
1317
+ },
1318
+ "shape": [
1319
+ 64,
1320
+ 896,
1321
+ 320
1322
+ ],
1323
+ "file": "model.safetensors",
1324
+ "offset_bytes": 915971993,
1325
+ "length_bytes": 73400320
1326
+ },
1327
+ {
1328
+ "name": "language_model.model.layers.5.mlp.gate.weight",
1329
+ "role": "ffn_gate_inp",
1330
+ "layer_index": 5,
1331
+ "dtype": "bf16",
1332
+ "shape": [
1333
+ 64,
1334
+ 1280
1335
+ ],
1336
+ "file": "model.safetensors",
1337
+ "offset_bytes": 835362713,
1338
+ "length_bytes": 163840
1339
+ },
1340
+ {
1341
+ "name": "language_model.model.layers.5.mlp.shared_experts.down_proj.weight",
1342
+ "role": "ffn_shared_expert_down",
1343
+ "layer_index": 5,
1344
+ "dtype": "u32",
1345
+ "source_quantized": true,
1346
+ "quantization": {
1347
+ "mode": "mxfp8",
1348
+ "group_size": 32,
1349
+ "bits": 8
1350
+ },
1351
+ "shape": [
1352
+ 1280,
1353
+ 448
1354
+ ],
1355
+ "file": "model.safetensors",
1356
+ "offset_bytes": 2830401945,
1357
+ "length_bytes": 2293760
1358
+ },
1359
+ {
1360
+ "name": "language_model.model.layers.5.mlp.shared_experts.gate_proj.weight",
1361
+ "role": "ffn_shared_expert_gate",
1362
+ "layer_index": 5,
1363
+ "dtype": "u32",
1364
+ "source_quantized": true,
1365
+ "quantization": {
1366
+ "mode": "mxfp8",
1367
+ "group_size": 32,
1368
+ "bits": 8
1369
+ },
1370
+ "shape": [
1371
+ 1792,
1372
+ 320
1373
+ ],
1374
+ "file": "model.safetensors",
1375
+ "offset_bytes": 833017753,
1376
+ "length_bytes": 2293760
1377
+ },
1378
+ {
1379
+ "name": "language_model.model.layers.5.mlp.shared_experts.up_proj.weight",
1380
+ "role": "ffn_shared_expert_up",
1381
+ "layer_index": 5,
1382
+ "dtype": "u32",
1383
+ "source_quantized": true,
1384
+ "quantization": {
1385
+ "mode": "mxfp8",
1386
+ "group_size": 32,
1387
+ "bits": 8
1388
+ },
1389
+ "shape": [
1390
+ 1792,
1391
+ 320
1392
+ ],
1393
+ "file": "model.safetensors",
1394
+ "offset_bytes": 830652313,
1395
+ "length_bytes": 2293760
1396
+ },
1397
+ {
1398
+ "name": "language_model.model.layers.5.mlp.switch_mlp.up_proj.weight",
1399
+ "role": "ffn_up_exps",
1400
+ "layer_index": 5,
1401
+ "dtype": "u32",
1402
+ "source_quantized": true,
1403
+ "quantization": {
1404
+ "mode": "mxfp8",
1405
+ "group_size": 32,
1406
+ "bits": 8
1407
+ },
1408
+ "shape": [
1409
+ 64,
1410
+ 896,
1411
+ 320
1412
+ ],
1413
+ "file": "model.safetensors",
1414
+ "offset_bytes": 2185178009,
1415
+ "length_bytes": 73400320
1416
+ },
1417
+ {
1418
+ "name": "language_model.model.layers.6.self_attn.k_proj.weight",
1419
+ "role": "attention_k",
1420
+ "layer_index": 6,
1421
+ "dtype": "u32",
1422
+ "source_quantized": true,
1423
+ "quantization": {
1424
+ "mode": "mxfp8",
1425
+ "group_size": 32,
1426
+ "bits": 8
1427
+ },
1428
+ "shape": [
1429
+ 1280,
1430
+ 320
1431
+ ],
1432
+ "file": "model.safetensors",
1433
+ "offset_bytes": 3507597081,
1434
+ "length_bytes": 1638400
1435
+ },
1436
+ {
1437
+ "name": "language_model.model.layers.6.input_layernorm.weight",
1438
+ "role": "attention_norm",
1439
+ "layer_index": 6,
1440
+ "dtype": "bf16",
1441
+ "shape": [
1442
+ 1280
1443
+ ],
1444
+ "file": "model.safetensors",
1445
+ "offset_bytes": 749738393,
1446
+ "length_bytes": 2560
1447
+ },
1448
+ {
1449
+ "name": "language_model.model.layers.6.self_attn.o_proj.weight",
1450
+ "role": "attention_o",
1451
+ "layer_index": 6,
1452
+ "dtype": "u32",
1453
+ "source_quantized": true,
1454
+ "quantization": {
1455
+ "mode": "mxfp8",
1456
+ "group_size": 32,
1457
+ "bits": 8
1458
+ },
1459
+ "shape": [
1460
+ 1280,
1461
+ 320
1462
+ ],
1463
+ "file": "model.safetensors",
1464
+ "offset_bytes": 825486233,
1465
+ "length_bytes": 1638400
1466
+ },
1467
+ {
1468
+ "name": "language_model.model.layers.6.post_attention_layernorm.weight",
1469
+ "role": "attention_post_norm",
1470
+ "layer_index": 6,
1471
+ "dtype": "bf16",
1472
+ "shape": [
1473
+ 1280
1474
+ ],
1475
+ "file": "model.safetensors",
1476
+ "offset_bytes": 1485195673,
1477
+ "length_bytes": 2560
1478
+ },
1479
+ {
1480
+ "name": "language_model.model.layers.6.self_attn.q_proj.weight",
1481
+ "role": "attention_q",
1482
+ "layer_index": 6,
1483
+ "dtype": "u32",
1484
+ "source_quantized": true,
1485
+ "quantization": {
1486
+ "mode": "mxfp8",
1487
+ "group_size": 32,
1488
+ "bits": 8
1489
+ },
1490
+ "shape": [
1491
+ 1280,
1492
+ 320
1493
+ ],
1494
+ "file": "model.safetensors",
1495
+ "offset_bytes": 828865433,
1496
+ "length_bytes": 1638400
1497
+ },
1498
+ {
1499
+ "name": "language_model.model.layers.6.self_attn.v_proj.weight",
1500
+ "role": "attention_v",
1501
+ "layer_index": 6,
1502
+ "dtype": "u32",
1503
+ "source_quantized": true,
1504
+ "quantization": {
1505
+ "mode": "mxfp8",
1506
+ "group_size": 32,
1507
+ "bits": 8
1508
+ },
1509
+ "shape": [
1510
+ 1280,
1511
+ 320
1512
+ ],
1513
+ "file": "model.safetensors",
1514
+ "offset_bytes": 827227033,
1515
+ "length_bytes": 1638400
1516
+ },
1517
+ {
1518
+ "name": "language_model.model.layers.6.mlp.switch_mlp.down_proj.weight",
1519
+ "role": "ffn_down_exps",
1520
+ "layer_index": 6,
1521
+ "dtype": "u32",
1522
+ "source_quantized": true,
1523
+ "quantization": {
1524
+ "mode": "mxfp8",
1525
+ "group_size": 32,
1526
+ "bits": 8
1527
+ },
1528
+ "shape": [
1529
+ 64,
1530
+ 1280,
1531
+ 224
1532
+ ],
1533
+ "file": "model.safetensors",
1534
+ "offset_bytes": 752034713,
1535
+ "length_bytes": 73400320
1536
+ },
1537
+ {
1538
+ "name": "language_model.model.layers.6.mlp.switch_mlp.gate_proj.weight",
1539
+ "role": "ffn_gate_exps",
1540
+ "layer_index": 6,
1541
+ "dtype": "u32",
1542
+ "source_quantized": true,
1543
+ "quantization": {
1544
+ "mode": "mxfp8",
1545
+ "group_size": 32,
1546
+ "bits": 8
1547
+ },
1548
+ "shape": [
1549
+ 64,
1550
+ 896,
1551
+ 320
1552
+ ],
1553
+ "file": "model.safetensors",
1554
+ "offset_bytes": 2642134553,
1555
+ "length_bytes": 73400320
1556
+ },
1557
+ {
1558
+ "name": "language_model.model.layers.6.mlp.gate.weight",
1559
+ "role": "ffn_gate_inp",
1560
+ "layer_index": 6,
1561
+ "dtype": "bf16",
1562
+ "shape": [
1563
+ 64,
1564
+ 1280
1565
+ ],
1566
+ "file": "model.safetensors",
1567
+ "offset_bytes": 835526553,
1568
+ "length_bytes": 163840
1569
+ },
1570
+ {
1571
+ "name": "language_model.model.layers.6.mlp.shared_experts.down_proj.weight",
1572
+ "role": "ffn_shared_expert_down",
1573
+ "layer_index": 6,
1574
+ "dtype": "u32",
1575
+ "source_quantized": true,
1576
+ "quantization": {
1577
+ "mode": "mxfp8",
1578
+ "group_size": 32,
1579
+ "bits": 8
1580
+ },
1581
+ "shape": [
1582
+ 1280,
1583
+ 448
1584
+ ],
1585
+ "file": "model.safetensors",
1586
+ "offset_bytes": 749740953,
1587
+ "length_bytes": 2293760
1588
+ },
1589
+ {
1590
+ "name": "language_model.model.layers.6.mlp.shared_experts.gate_proj.weight",
1591
+ "role": "ffn_shared_expert_gate",
1592
+ "layer_index": 6,
1593
+ "dtype": "u32",
1594
+ "source_quantized": true,
1595
+ "quantization": {
1596
+ "mode": "mxfp8",
1597
+ "group_size": 32,
1598
+ "bits": 8
1599
+ },
1600
+ "shape": [
1601
+ 1792,
1602
+ 320
1603
+ ],
1604
+ "file": "model.safetensors",
1605
+ "offset_bytes": 1714154393,
1606
+ "length_bytes": 2293760
1607
+ },
1608
+ {
1609
+ "name": "language_model.model.layers.6.mlp.shared_experts.up_proj.weight",
1610
+ "role": "ffn_shared_expert_up",
1611
+ "layer_index": 6,
1612
+ "dtype": "u32",
1613
+ "source_quantized": true,
1614
+ "quantization": {
1615
+ "mode": "mxfp8",
1616
+ "group_size": 32,
1617
+ "bits": 8
1618
+ },
1619
+ "shape": [
1620
+ 1792,
1621
+ 320
1622
+ ],
1623
+ "file": "model.safetensors",
1624
+ "offset_bytes": 3198452505,
1625
+ "length_bytes": 2293760
1626
+ },
1627
+ {
1628
+ "name": "language_model.model.layers.6.mlp.switch_mlp.up_proj.weight",
1629
+ "role": "ffn_up_exps",
1630
+ "layer_index": 6,
1631
+ "dtype": "u32",
1632
+ "source_quantized": true,
1633
+ "quantization": {
1634
+ "mode": "mxfp8",
1635
+ "group_size": 32,
1636
+ "bits": 8
1637
+ },
1638
+ "shape": [
1639
+ 64,
1640
+ 896,
1641
+ 320
1642
+ ],
1643
+ "file": "model.safetensors",
1644
+ "offset_bytes": 3434024729,
1645
+ "length_bytes": 73400320
1646
+ },
1647
+ {
1648
+ "name": "language_model.model.layers.7.self_attn.k_proj.weight",
1649
+ "role": "attention_k",
1650
+ "layer_index": 7,
1651
+ "dtype": "u32",
1652
+ "source_quantized": true,
1653
+ "quantization": {
1654
+ "mode": "mxfp8",
1655
+ "group_size": 32,
1656
+ "bits": 8
1657
+ },
1658
+ "shape": [
1659
+ 1280,
1660
+ 320
1661
+ ],
1662
+ "file": "model.safetensors",
1663
+ "offset_bytes": 748048793,
1664
+ "length_bytes": 1638400
1665
+ },
1666
+ {
1667
+ "name": "language_model.model.layers.7.input_layernorm.weight",
1668
+ "role": "attention_norm",
1669
+ "layer_index": 7,
1670
+ "dtype": "bf16",
1671
+ "shape": [
1672
+ 1280
1673
+ ],
1674
+ "file": "model.safetensors",
1675
+ "offset_bytes": 3392695705,
1676
+ "length_bytes": 2560
1677
+ },
1678
+ {
1679
+ "name": "language_model.model.layers.7.self_attn.o_proj.weight",
1680
+ "role": "attention_o",
1681
+ "layer_index": 7,
1682
+ "dtype": "u32",
1683
+ "source_quantized": true,
1684
+ "quantization": {
1685
+ "mode": "mxfp8",
1686
+ "group_size": 32,
1687
+ "bits": 8
1688
+ },
1689
+ "shape": [
1690
+ 1280,
1691
+ 320
1692
+ ],
1693
+ "file": "model.safetensors",
1694
+ "offset_bytes": 2818202521,
1695
+ "length_bytes": 1638400
1696
+ },
1697
+ {
1698
+ "name": "language_model.model.layers.7.post_attention_layernorm.weight",
1699
+ "role": "attention_post_norm",
1700
+ "layer_index": 7,
1701
+ "dtype": "bf16",
1702
+ "shape": [
1703
+ 1280
1704
+ ],
1705
+ "file": "model.safetensors",
1706
+ "offset_bytes": 743284633,
1707
+ "length_bytes": 2560
1708
+ },
1709
+ {
1710
+ "name": "language_model.model.layers.7.self_attn.q_proj.weight",
1711
+ "role": "attention_q",
1712
+ "layer_index": 7,
1713
+ "dtype": "u32",
1714
+ "source_quantized": true,
1715
+ "quantization": {
1716
+ "mode": "mxfp8",
1717
+ "group_size": 32,
1718
+ "bits": 8
1719
+ },
1720
+ "shape": [
1721
+ 1280,
1722
+ 320
1723
+ ],
1724
+ "file": "model.safetensors",
1725
+ "offset_bytes": 1487563673,
1726
+ "length_bytes": 1638400
1727
+ },
1728
+ {
1729
+ "name": "language_model.model.layers.7.self_attn.v_proj.weight",
1730
+ "role": "attention_v",
1731
+ "layer_index": 7,
1732
+ "dtype": "u32",
1733
+ "source_quantized": true,
1734
+ "quantization": {
1735
+ "mode": "mxfp8",
1736
+ "group_size": 32,
1737
+ "bits": 8
1738
+ },
1739
+ "shape": [
1740
+ 1280,
1741
+ 320
1742
+ ],
1743
+ "file": "model.safetensors",
1744
+ "offset_bytes": 2183539609,
1745
+ "length_bytes": 1638400
1746
+ },
1747
+ {
1748
+ "name": "language_model.model.layers.7.mlp.switch_mlp.down_proj.weight",
1749
+ "role": "ffn_down_exps",
1750
+ "layer_index": 7,
1751
+ "dtype": "u32",
1752
+ "source_quantized": true,
1753
+ "quantization": {
1754
+ "mode": "mxfp8",
1755
+ "group_size": 32,
1756
+ "bits": 8
1757
+ },
1758
+ "shape": [
1759
+ 64,
1760
+ 1280,
1761
+ 224
1762
+ ],
1763
+ "file": "model.safetensors",
1764
+ "offset_bytes": 1319886233,
1765
+ "length_bytes": 73400320
1766
+ },
1767
+ {
1768
+ "name": "language_model.model.layers.7.mlp.switch_mlp.gate_proj.weight",
1769
+ "role": "ffn_gate_exps",
1770
+ "layer_index": 7,
1771
+ "dtype": "u32",
1772
+ "source_quantized": true,
1773
+ "quantization": {
1774
+ "mode": "mxfp8",
1775
+ "group_size": 32,
1776
+ "bits": 8
1777
+ },
1778
+ "shape": [
1779
+ 64,
1780
+ 896,
1781
+ 320
1782
+ ],
1783
+ "file": "model.safetensors",
1784
+ "offset_bytes": 2861748249,
1785
+ "length_bytes": 73400320
1786
+ },
1787
+ {
1788
+ "name": "language_model.model.layers.7.mlp.gate.weight",
1789
+ "role": "ffn_gate_inp",
1790
+ "layer_index": 7,
1791
+ "dtype": "bf16",
1792
+ "shape": [
1793
+ 64,
1794
+ 1280
1795
+ ],
1796
+ "file": "model.safetensors",
1797
+ "offset_bytes": 3507431193,
1798
+ "length_bytes": 163840
1799
+ },
1800
+ {
1801
+ "name": "language_model.model.layers.7.mlp.shared_experts.down_proj.weight",
1802
+ "role": "ffn_shared_expert_down",
1803
+ "layer_index": 7,
1804
+ "dtype": "u32",
1805
+ "source_quantized": true,
1806
+ "quantization": {
1807
+ "mode": "mxfp8",
1808
+ "group_size": 32,
1809
+ "bits": 8
1810
+ },
1811
+ "shape": [
1812
+ 1280,
1813
+ 448
1814
+ ],
1815
+ "file": "model.safetensors",
1816
+ "offset_bytes": 3179375897,
1817
+ "length_bytes": 2293760
1818
+ },
1819
+ {
1820
+ "name": "language_model.model.layers.7.mlp.shared_experts.gate_proj.weight",
1821
+ "role": "ffn_shared_expert_gate",
1822
+ "layer_index": 7,
1823
+ "dtype": "u32",
1824
+ "source_quantized": true,
1825
+ "quantization": {
1826
+ "mode": "mxfp8",
1827
+ "group_size": 32,
1828
+ "bits": 8
1829
+ },
1830
+ "shape": [
1831
+ 1792,
1832
+ 320
1833
+ ],
1834
+ "file": "model.safetensors",
1835
+ "offset_bytes": 3417343897,
1836
+ "length_bytes": 2293760
1837
+ },
1838
+ {
1839
+ "name": "language_model.model.layers.7.mlp.shared_experts.up_proj.weight",
1840
+ "role": "ffn_shared_expert_up",
1841
+ "layer_index": 7,
1842
+ "dtype": "u32",
1843
+ "source_quantized": true,
1844
+ "quantization": {
1845
+ "mode": "mxfp8",
1846
+ "group_size": 32,
1847
+ "bits": 8
1848
+ },
1849
+ "shape": [
1850
+ 1792,
1851
+ 320
1852
+ ],
1853
+ "file": "model.safetensors",
1854
+ "offset_bytes": 743358873,
1855
+ "length_bytes": 2293760
1856
+ },
1857
+ {
1858
+ "name": "language_model.model.layers.7.mlp.switch_mlp.up_proj.weight",
1859
+ "role": "ffn_up_exps",
1860
+ "layer_index": 7,
1861
+ "dtype": "u32",
1862
+ "source_quantized": true,
1863
+ "quantization": {
1864
+ "mode": "mxfp8",
1865
+ "group_size": 32,
1866
+ "bits": 8
1867
+ },
1868
+ "shape": [
1869
+ 64,
1870
+ 896,
1871
+ 320
1872
+ ],
1873
+ "file": "model.safetensors",
1874
+ "offset_bytes": 3642288409,
1875
+ "length_bytes": 73400320
1876
+ },
1877
+ {
1878
+ "name": "language_model.model.layers.8.self_attn.k_proj.weight",
1879
+ "role": "attention_k",
1880
+ "layer_index": 8,
1881
+ "dtype": "u32",
1882
+ "source_quantized": true,
1883
+ "quantization": {
1884
+ "mode": "mxfp8",
1885
+ "group_size": 32,
1886
+ "bits": 8
1887
+ },
1888
+ "shape": [
1889
+ 1280,
1890
+ 320
1891
+ ],
1892
+ "file": "model.safetensors",
1893
+ "offset_bytes": 989423513,
1894
+ "length_bytes": 1638400
1895
+ },
1896
+ {
1897
+ "name": "language_model.model.layers.8.input_layernorm.weight",
1898
+ "role": "attention_norm",
1899
+ "layer_index": 8,
1900
+ "dtype": "bf16",
1901
+ "shape": [
1902
+ 1280
1903
+ ],
1904
+ "file": "model.safetensors",
1905
+ "offset_bytes": 3342675993,
1906
+ "length_bytes": 2560
1907
+ },
1908
+ {
1909
+ "name": "language_model.model.layers.8.self_attn.o_proj.weight",
1910
+ "role": "attention_o",
1911
+ "layer_index": 8,
1912
+ "dtype": "u32",
1913
+ "source_quantized": true,
1914
+ "quantization": {
1915
+ "mode": "mxfp8",
1916
+ "group_size": 32,
1917
+ "bits": 8
1918
+ },
1919
+ "shape": [
1920
+ 1280,
1921
+ 320
1922
+ ],
1923
+ "file": "model.safetensors",
1924
+ "offset_bytes": 2364151577,
1925
+ "length_bytes": 1638400
1926
+ },
1927
+ {
1928
+ "name": "language_model.model.layers.8.post_attention_layernorm.weight",
1929
+ "role": "attention_post_norm",
1930
+ "layer_index": 8,
1931
+ "dtype": "bf16",
1932
+ "shape": [
1933
+ 1280
1934
+ ],
1935
+ "file": "model.safetensors",
1936
+ "offset_bytes": 578612633,
1937
+ "length_bytes": 2560
1938
+ },
1939
+ {
1940
+ "name": "language_model.model.layers.8.self_attn.q_proj.weight",
1941
+ "role": "attention_q",
1942
+ "layer_index": 8,
1943
+ "dtype": "u32",
1944
+ "source_quantized": true,
1945
+ "quantization": {
1946
+ "mode": "mxfp8",
1947
+ "group_size": 32,
1948
+ "bits": 8
1949
+ },
1950
+ "shape": [
1951
+ 1280,
1952
+ 320
1953
+ ],
1954
+ "file": "model.safetensors",
1955
+ "offset_bytes": 3632181529,
1956
+ "length_bytes": 1638400
1957
+ },
1958
+ {
1959
+ "name": "language_model.model.layers.8.self_attn.v_proj.weight",
1960
+ "role": "attention_v",
1961
+ "layer_index": 8,
1962
+ "dtype": "u32",
1963
+ "source_quantized": true,
1964
+ "quantization": {
1965
+ "mode": "mxfp8",
1966
+ "group_size": 32,
1967
+ "bits": 8
1968
+ },
1969
+ "shape": [
1970
+ 1280,
1971
+ 320
1972
+ ],
1973
+ "file": "model.safetensors",
1974
+ "offset_bytes": 1729138073,
1975
+ "length_bytes": 1638400
1976
+ },
1977
+ {
1978
+ "name": "language_model.model.layers.8.mlp.switch_mlp.down_proj.weight",
1979
+ "role": "ffn_down_exps",
1980
+ "layer_index": 8,
1981
+ "dtype": "u32",
1982
+ "source_quantized": true,
1983
+ "quantization": {
1984
+ "mode": "mxfp8",
1985
+ "group_size": 32,
1986
+ "bits": 8
1987
+ },
1988
+ "shape": [
1989
+ 64,
1990
+ 1280,
1991
+ 224
1992
+ ],
1993
+ "file": "model.safetensors",
1994
+ "offset_bytes": 992751513,
1995
+ "length_bytes": 73400320
1996
+ },
1997
+ {
1998
+ "name": "language_model.model.layers.8.mlp.switch_mlp.gate_proj.weight",
1999
+ "role": "ffn_gate_exps",
2000
+ "layer_index": 8,
2001
+ "dtype": "u32",
2002
+ "source_quantized": true,
2003
+ "quantization": {
2004
+ "mode": "mxfp8",
2005
+ "group_size": 32,
2006
+ "bits": 8
2007
+ },
2008
+ "shape": [
2009
+ 64,
2010
+ 896,
2011
+ 320
2012
+ ],
2013
+ "file": "model.safetensors",
2014
+ "offset_bytes": 668194713,
2015
+ "length_bytes": 73400320
2016
+ },
2017
+ {
2018
+ "name": "language_model.model.layers.8.mlp.gate.weight",
2019
+ "role": "ffn_gate_inp",
2020
+ "layer_index": 8,
2021
+ "dtype": "bf16",
2022
+ "shape": [
2023
+ 64,
2024
+ 1280
2025
+ ],
2026
+ "file": "model.safetensors",
2027
+ "offset_bytes": 2571522841,
2028
+ "length_bytes": 163840
2029
+ },
2030
+ {
2031
+ "name": "language_model.model.layers.8.mlp.shared_experts.down_proj.weight",
2032
+ "role": "ffn_shared_expert_down",
2033
+ "layer_index": 8,
2034
+ "dtype": "u32",
2035
+ "source_quantized": true,
2036
+ "quantization": {
2037
+ "mode": "mxfp8",
2038
+ "group_size": 32,
2039
+ "bits": 8
2040
+ },
2041
+ "shape": [
2042
+ 1280,
2043
+ 448
2044
+ ],
2045
+ "file": "model.safetensors",
2046
+ "offset_bytes": 578738073,
2047
+ "length_bytes": 2293760
2048
+ },
2049
+ {
2050
+ "name": "language_model.model.layers.8.mlp.shared_experts.gate_proj.weight",
2051
+ "role": "ffn_shared_expert_gate",
2052
+ "layer_index": 8,
2053
+ "dtype": "u32",
2054
+ "source_quantized": true,
2055
+ "quantization": {
2056
+ "mode": "mxfp8",
2057
+ "group_size": 32,
2058
+ "bits": 8
2059
+ },
2060
+ "shape": [
2061
+ 1792,
2062
+ 320
2063
+ ],
2064
+ "file": "model.safetensors",
2065
+ "offset_bytes": 583325593,
2066
+ "length_bytes": 2293760
2067
+ },
2068
+ {
2069
+ "name": "language_model.model.layers.8.mlp.shared_experts.up_proj.weight",
2070
+ "role": "ffn_shared_expert_up",
2071
+ "layer_index": 8,
2072
+ "dtype": "u32",
2073
+ "source_quantized": true,
2074
+ "quantization": {
2075
+ "mode": "mxfp8",
2076
+ "group_size": 32,
2077
+ "bits": 8
2078
+ },
2079
+ "shape": [
2080
+ 1792,
2081
+ 320
2082
+ ],
2083
+ "file": "model.safetensors",
2084
+ "offset_bytes": 581031833,
2085
+ "length_bytes": 2293760
2086
+ },
2087
+ {
2088
+ "name": "language_model.model.layers.8.mlp.switch_mlp.up_proj.weight",
2089
+ "role": "ffn_up_exps",
2090
+ "layer_index": 8,
2091
+ "dtype": "u32",
2092
+ "source_quantized": true,
2093
+ "quantization": {
2094
+ "mode": "mxfp8",
2095
+ "group_size": 32,
2096
+ "bits": 8
2097
+ },
2098
+ "shape": [
2099
+ 64,
2100
+ 896,
2101
+ 320
2102
+ ],
2103
+ "file": "model.safetensors",
2104
+ "offset_bytes": 592500633,
2105
+ "length_bytes": 73400320
2106
+ },
2107
+ {
2108
+ "name": "language_model.model.layers.9.self_attn.k_proj.weight",
2109
+ "role": "attention_k",
2110
+ "layer_index": 9,
2111
+ "dtype": "u32",
2112
+ "source_quantized": true,
2113
+ "quantization": {
2114
+ "mode": "mxfp8",
2115
+ "group_size": 32,
2116
+ "bits": 8
2117
+ },
2118
+ "shape": [
2119
+ 1280,
2120
+ 320
2121
+ ],
2122
+ "file": "model.safetensors",
2123
+ "offset_bytes": 576923033,
2124
+ "length_bytes": 1638400
2125
+ },
2126
+ {
2127
+ "name": "language_model.model.layers.9.input_layernorm.weight",
2128
+ "role": "attention_norm",
2129
+ "layer_index": 9,
2130
+ "dtype": "bf16",
2131
+ "shape": [
2132
+ 1280
2133
+ ],
2134
+ "file": "model.safetensors",
2135
+ "offset_bytes": 1485047193,
2136
+ "length_bytes": 2560
2137
+ },
2138
+ {
2139
+ "name": "language_model.model.layers.9.self_attn.o_proj.weight",
2140
+ "role": "attention_o",
2141
+ "layer_index": 9,
2142
+ "dtype": "u32",
2143
+ "source_quantized": true,
2144
+ "quantization": {
2145
+ "mode": "mxfp8",
2146
+ "group_size": 32,
2147
+ "bits": 8
2148
+ },
2149
+ "shape": [
2150
+ 1280,
2151
+ 320
2152
+ ],
2153
+ "file": "model.safetensors",
2154
+ "offset_bytes": 741646233,
2155
+ "length_bytes": 1638400
2156
+ },
2157
+ {
2158
+ "name": "language_model.model.layers.9.post_attention_layernorm.weight",
2159
+ "role": "attention_post_norm",
2160
+ "layer_index": 9,
2161
+ "dtype": "bf16",
2162
+ "shape": [
2163
+ 1280
2164
+ ],
2165
+ "file": "model.safetensors",
2166
+ "offset_bytes": 3405807001,
2167
+ "length_bytes": 2560
2168
+ },
2169
+ {
2170
+ "name": "language_model.model.layers.9.self_attn.q_proj.weight",
2171
+ "role": "attention_q",
2172
+ "layer_index": 9,
2173
+ "dtype": "u32",
2174
+ "source_quantized": true,
2175
+ "quantization": {
2176
+ "mode": "mxfp8",
2177
+ "group_size": 32,
2178
+ "bits": 8
2179
+ },
2180
+ "shape": [
2181
+ 1280,
2182
+ 320
2183
+ ],
2184
+ "file": "model.safetensors",
2185
+ "offset_bytes": 3559845145,
2186
+ "length_bytes": 1638400
2187
+ },
2188
+ {
2189
+ "name": "language_model.model.layers.9.self_attn.v_proj.weight",
2190
+ "role": "attention_v",
2191
+ "layer_index": 9,
2192
+ "dtype": "u32",
2193
+ "source_quantized": true,
2194
+ "quantization": {
2195
+ "mode": "mxfp8",
2196
+ "group_size": 32,
2197
+ "bits": 8
2198
+ },
2199
+ "shape": [
2200
+ 1280,
2201
+ 320
2202
+ ],
2203
+ "file": "model.safetensors",
2204
+ "offset_bytes": 575233433,
2205
+ "length_bytes": 1638400
2206
+ },
2207
+ {
2208
+ "name": "language_model.model.layers.9.mlp.switch_mlp.down_proj.weight",
2209
+ "role": "ffn_down_exps",
2210
+ "layer_index": 9,
2211
+ "dtype": "u32",
2212
+ "source_quantized": true,
2213
+ "quantization": {
2214
+ "mode": "mxfp8",
2215
+ "group_size": 32,
2216
+ "bits": 8
2217
+ },
2218
+ "shape": [
2219
+ 64,
2220
+ 1280,
2221
+ 224
2222
+ ],
2223
+ "file": "model.safetensors",
2224
+ "offset_bytes": 2102578713,
2225
+ "length_bytes": 73400320
2226
+ },
2227
+ {
2228
+ "name": "language_model.model.layers.9.mlp.switch_mlp.gate_proj.weight",
2229
+ "role": "ffn_gate_exps",
2230
+ "layer_index": 9,
2231
+ "dtype": "u32",
2232
+ "source_quantized": true,
2233
+ "quantization": {
2234
+ "mode": "mxfp8",
2235
+ "group_size": 32,
2236
+ "bits": 8
2237
+ },
2238
+ "shape": [
2239
+ 64,
2240
+ 896,
2241
+ 320
2242
+ ],
2243
+ "file": "model.safetensors",
2244
+ "offset_bytes": 501730713,
2245
+ "length_bytes": 73400320
2246
+ },
2247
+ {
2248
+ "name": "language_model.model.layers.9.mlp.gate.weight",
2249
+ "role": "ffn_gate_inp",
2250
+ "layer_index": 9,
2251
+ "dtype": "bf16",
2252
+ "shape": [
2253
+ 64,
2254
+ 1280
2255
+ ],
2256
+ "file": "model.safetensors",
2257
+ "offset_bytes": 419646873,
2258
+ "length_bytes": 163840
2259
+ },
2260
+ {
2261
+ "name": "language_model.model.layers.9.mlp.shared_experts.down_proj.weight",
2262
+ "role": "ffn_shared_expert_down",
2263
+ "layer_index": 9,
2264
+ "dtype": "u32",
2265
+ "source_quantized": true,
2266
+ "quantization": {
2267
+ "mode": "mxfp8",
2268
+ "group_size": 32,
2269
+ "bits": 8
2270
+ },
2271
+ "shape": [
2272
+ 1280,
2273
+ 448
2274
+ ],
2275
+ "file": "model.safetensors",
2276
+ "offset_bytes": 2012082457,
2277
+ "length_bytes": 2293760
2278
+ },
2279
+ {
2280
+ "name": "language_model.model.layers.9.mlp.shared_experts.gate_proj.weight",
2281
+ "role": "ffn_shared_expert_gate",
2282
+ "layer_index": 9,
2283
+ "dtype": "u32",
2284
+ "source_quantized": true,
2285
+ "quantization": {
2286
+ "mode": "mxfp8",
2287
+ "group_size": 32,
2288
+ "bits": 8
2289
+ },
2290
+ "shape": [
2291
+ 1792,
2292
+ 320
2293
+ ],
2294
+ "file": "model.safetensors",
2295
+ "offset_bytes": 3760249753,
2296
+ "length_bytes": 2293760
2297
+ },
2298
+ {
2299
+ "name": "language_model.model.layers.9.mlp.shared_experts.up_proj.weight",
2300
+ "role": "ffn_shared_expert_up",
2301
+ "layer_index": 9,
2302
+ "dtype": "u32",
2303
+ "source_quantized": true,
2304
+ "quantization": {
2305
+ "mode": "mxfp8",
2306
+ "group_size": 32,
2307
+ "bits": 8
2308
+ },
2309
+ "shape": [
2310
+ 1792,
2311
+ 320
2312
+ ],
2313
+ "file": "model.safetensors",
2314
+ "offset_bytes": 590206873,
2315
+ "length_bytes": 2293760
2316
+ },
2317
+ {
2318
+ "name": "language_model.model.layers.9.mlp.switch_mlp.up_proj.weight",
2319
+ "role": "ffn_up_exps",
2320
+ "layer_index": 9,
2321
+ "dtype": "u32",
2322
+ "source_quantized": true,
2323
+ "quantization": {
2324
+ "mode": "mxfp8",
2325
+ "group_size": 32,
2326
+ "bits": 8
2327
+ },
2328
+ "shape": [
2329
+ 64,
2330
+ 896,
2331
+ 320
2332
+ ],
2333
+ "file": "model.safetensors",
2334
+ "offset_bytes": 426036633,
2335
+ "length_bytes": 73400320
2336
+ },
2337
+ {
2338
+ "name": "language_model.model.layers.10.self_attn.k_proj.weight",
2339
+ "role": "attention_k",
2340
+ "layer_index": 10,
2341
+ "dtype": "u32",
2342
+ "source_quantized": true,
2343
+ "quantization": {
2344
+ "mode": "mxfp8",
2345
+ "group_size": 32,
2346
+ "bits": 8
2347
+ },
2348
+ "shape": [
2349
+ 1280,
2350
+ 320
2351
+ ],
2352
+ "file": "model.safetensors",
2353
+ "offset_bytes": 3813974425,
2354
+ "length_bytes": 1638400
2355
+ },
2356
+ {
2357
+ "name": "language_model.model.layers.10.input_layernorm.weight",
2358
+ "role": "attention_norm",
2359
+ "layer_index": 10,
2360
+ "dtype": "bf16",
2361
+ "shape": [
2362
+ 1280
2363
+ ],
2364
+ "file": "model.safetensors",
2365
+ "offset_bytes": 1750875033,
2366
+ "length_bytes": 2560
2367
+ },
2368
+ {
2369
+ "name": "language_model.model.layers.10.self_attn.o_proj.weight",
2370
+ "role": "attention_o",
2371
+ "layer_index": 10,
2372
+ "dtype": "u32",
2373
+ "source_quantized": true,
2374
+ "quantization": {
2375
+ "mode": "mxfp8",
2376
+ "group_size": 32,
2377
+ "bits": 8
2378
+ },
2379
+ "shape": [
2380
+ 1280,
2381
+ 320
2382
+ ],
2383
+ "file": "model.safetensors",
2384
+ "offset_bytes": 416103833,
2385
+ "length_bytes": 1638400
2386
+ },
2387
+ {
2388
+ "name": "language_model.model.layers.10.post_attention_layernorm.weight",
2389
+ "role": "attention_post_norm",
2390
+ "layer_index": 10,
2391
+ "dtype": "bf16",
2392
+ "shape": [
2393
+ 1280
2394
+ ],
2395
+ "file": "model.safetensors",
2396
+ "offset_bytes": 330873753,
2397
+ "length_bytes": 2560
2398
+ },
2399
+ {
2400
+ "name": "language_model.model.layers.10.self_attn.q_proj.weight",
2401
+ "role": "attention_q",
2402
+ "layer_index": 10,
2403
+ "dtype": "u32",
2404
+ "source_quantized": true,
2405
+ "quantization": {
2406
+ "mode": "mxfp8",
2407
+ "group_size": 32,
2408
+ "bits": 8
2409
+ },
2410
+ "shape": [
2411
+ 1280,
2412
+ 320
2413
+ ],
2414
+ "file": "model.safetensors",
2415
+ "offset_bytes": 417793433,
2416
+ "length_bytes": 1638400
2417
+ },
2418
+ {
2419
+ "name": "language_model.model.layers.10.self_attn.v_proj.weight",
2420
+ "role": "attention_v",
2421
+ "layer_index": 10,
2422
+ "dtype": "u32",
2423
+ "source_quantized": true,
2424
+ "quantization": {
2425
+ "mode": "mxfp8",
2426
+ "group_size": 32,
2427
+ "bits": 8
2428
+ },
2429
+ "shape": [
2430
+ 1280,
2431
+ 320
2432
+ ],
2433
+ "file": "model.safetensors",
2434
+ "offset_bytes": 2543350297,
2435
+ "length_bytes": 1638400
2436
+ },
2437
+ {
2438
+ "name": "language_model.model.layers.10.mlp.switch_mlp.down_proj.weight",
2439
+ "role": "ffn_down_exps",
2440
+ "layer_index": 10,
2441
+ "dtype": "u32",
2442
+ "source_quantized": true,
2443
+ "quantization": {
2444
+ "mode": "mxfp8",
2445
+ "group_size": 32,
2446
+ "bits": 8
2447
+ },
2448
+ "shape": [
2449
+ 64,
2450
+ 1280,
2451
+ 224
2452
+ ],
2453
+ "file": "model.safetensors",
2454
+ "offset_bytes": 1242405273,
2455
+ "length_bytes": 73400320
2456
+ },
2457
+ {
2458
+ "name": "language_model.model.layers.10.mlp.switch_mlp.gate_proj.weight",
2459
+ "role": "ffn_gate_exps",
2460
+ "layer_index": 10,
2461
+ "dtype": "u32",
2462
+ "source_quantized": true,
2463
+ "quantization": {
2464
+ "mode": "mxfp8",
2465
+ "group_size": 32,
2466
+ "bits": 8
2467
+ },
2468
+ "shape": [
2469
+ 64,
2470
+ 896,
2471
+ 320
2472
+ ],
2473
+ "file": "model.safetensors",
2474
+ "offset_bytes": 1491659673,
2475
+ "length_bytes": 73400320
2476
+ },
2477
+ {
2478
+ "name": "language_model.model.layers.10.mlp.gate.weight",
2479
+ "role": "ffn_gate_inp",
2480
+ "layer_index": 10,
2481
+ "dtype": "bf16",
2482
+ "shape": [
2483
+ 64,
2484
+ 1280
2485
+ ],
2486
+ "file": "model.safetensors",
2487
+ "offset_bytes": 335607193,
2488
+ "length_bytes": 163840
2489
+ },
2490
+ {
2491
+ "name": "language_model.model.layers.10.mlp.shared_experts.down_proj.weight",
2492
+ "role": "ffn_shared_expert_down",
2493
+ "layer_index": 10,
2494
+ "dtype": "u32",
2495
+ "source_quantized": true,
2496
+ "quantization": {
2497
+ "mode": "mxfp8",
2498
+ "group_size": 32,
2499
+ "bits": 8
2500
+ },
2501
+ "shape": [
2502
+ 1280,
2503
+ 448
2504
+ ],
2505
+ "file": "model.safetensors",
2506
+ "offset_bytes": 3796838297,
2507
+ "length_bytes": 2293760
2508
+ },
2509
+ {
2510
+ "name": "language_model.model.layers.10.mlp.shared_experts.gate_proj.weight",
2511
+ "role": "ffn_shared_expert_gate",
2512
+ "layer_index": 10,
2513
+ "dtype": "u32",
2514
+ "source_quantized": true,
2515
+ "quantization": {
2516
+ "mode": "mxfp8",
2517
+ "group_size": 32,
2518
+ "bits": 8
2519
+ },
2520
+ "shape": [
2521
+ 1792,
2522
+ 320
2523
+ ],
2524
+ "file": "model.safetensors",
2525
+ "offset_bytes": 333313433,
2526
+ "length_bytes": 2293760
2527
+ },
2528
+ {
2529
+ "name": "language_model.model.layers.10.mlp.shared_experts.up_proj.weight",
2530
+ "role": "ffn_shared_expert_up",
2531
+ "layer_index": 10,
2532
+ "dtype": "u32",
2533
+ "source_quantized": true,
2534
+ "quantization": {
2535
+ "mode": "mxfp8",
2536
+ "group_size": 32,
2537
+ "bits": 8
2538
+ },
2539
+ "shape": [
2540
+ 1792,
2541
+ 320
2542
+ ],
2543
+ "file": "model.safetensors",
2544
+ "offset_bytes": 330947993,
2545
+ "length_bytes": 2293760
2546
+ },
2547
+ {
2548
+ "name": "language_model.model.layers.10.mlp.switch_mlp.up_proj.weight",
2549
+ "role": "ffn_up_exps",
2550
+ "layer_index": 10,
2551
+ "dtype": "u32",
2552
+ "source_quantized": true,
2553
+ "quantization": {
2554
+ "mode": "mxfp8",
2555
+ "group_size": 32,
2556
+ "bits": 8
2557
+ },
2558
+ "shape": [
2559
+ 64,
2560
+ 896,
2561
+ 320
2562
+ ],
2563
+ "file": "model.safetensors",
2564
+ "offset_bytes": 340358553,
2565
+ "length_bytes": 73400320
2566
+ },
2567
+ {
2568
+ "name": "language_model.model.layers.11.self_attn.k_proj.weight",
2569
+ "role": "attention_k",
2570
+ "layer_index": 11,
2571
+ "dtype": "u32",
2572
+ "source_quantized": true,
2573
+ "quantization": {
2574
+ "mode": "mxfp8",
2575
+ "group_size": 32,
2576
+ "bits": 8
2577
+ },
2578
+ "shape": [
2579
+ 1280,
2580
+ 320
2581
+ ],
2582
+ "file": "model.safetensors",
2583
+ "offset_bytes": 329235353,
2584
+ "length_bytes": 1638400
2585
+ },
2586
+ {
2587
+ "name": "language_model.model.layers.11.input_layernorm.weight",
2588
+ "role": "attention_norm",
2589
+ "layer_index": 11,
2590
+ "dtype": "bf16",
2591
+ "shape": [
2592
+ 1280
2593
+ ],
2594
+ "file": "model.safetensors",
2595
+ "offset_bytes": 2177161753,
2596
+ "length_bytes": 2560
2597
+ },
2598
+ {
2599
+ "name": "language_model.model.layers.11.self_attn.o_proj.weight",
2600
+ "role": "attention_o",
2601
+ "layer_index": 11,
2602
+ "dtype": "u32",
2603
+ "source_quantized": true,
2604
+ "quantization": {
2605
+ "mode": "mxfp8",
2606
+ "group_size": 32,
2607
+ "bits": 8
2608
+ },
2609
+ "shape": [
2610
+ 1280,
2611
+ 320
2612
+ ],
2613
+ "file": "model.safetensors",
2614
+ "offset_bytes": 3362843673,
2615
+ "length_bytes": 1638400
2616
+ },
2617
+ {
2618
+ "name": "language_model.model.layers.11.post_attention_layernorm.weight",
2619
+ "role": "attention_post_norm",
2620
+ "layer_index": 11,
2621
+ "dtype": "bf16",
2622
+ "shape": [
2623
+ 1280
2624
+ ],
2625
+ "file": "model.safetensors",
2626
+ "offset_bytes": 175919513,
2627
+ "length_bytes": 2560
2628
+ },
2629
+ {
2630
+ "name": "language_model.model.layers.11.self_attn.q_proj.weight",
2631
+ "role": "attention_q",
2632
+ "layer_index": 11,
2633
+ "dtype": "u32",
2634
+ "source_quantized": true,
2635
+ "quantization": {
2636
+ "mode": "mxfp8",
2637
+ "group_size": 32,
2638
+ "bits": 8
2639
+ },
2640
+ "shape": [
2641
+ 1280,
2642
+ 320
2643
+ ],
2644
+ "file": "model.safetensors",
2645
+ "offset_bytes": 424398233,
2646
+ "length_bytes": 1638400
2647
+ },
2648
+ {
2649
+ "name": "language_model.model.layers.11.self_attn.v_proj.weight",
2650
+ "role": "attention_v",
2651
+ "layer_index": 11,
2652
+ "dtype": "u32",
2653
+ "source_quantized": true,
2654
+ "quantization": {
2655
+ "mode": "mxfp8",
2656
+ "group_size": 32,
2657
+ "bits": 8
2658
+ },
2659
+ "shape": [
2660
+ 1280,
2661
+ 320
2662
+ ],
2663
+ "file": "model.safetensors",
2664
+ "offset_bytes": 254196633,
2665
+ "length_bytes": 1638400
2666
+ },
2667
+ {
2668
+ "name": "language_model.model.layers.11.mlp.switch_mlp.down_proj.weight",
2669
+ "role": "ffn_down_exps",
2670
+ "layer_index": 11,
2671
+ "dtype": "u32",
2672
+ "source_quantized": true,
2673
+ "quantization": {
2674
+ "mode": "mxfp8",
2675
+ "group_size": 32,
2676
+ "bits": 8
2677
+ },
2678
+ "shape": [
2679
+ 64,
2680
+ 1280,
2681
+ 224
2682
+ ],
2683
+ "file": "model.safetensors",
2684
+ "offset_bytes": 2020812953,
2685
+ "length_bytes": 73400320
2686
+ },
2687
+ {
2688
+ "name": "language_model.model.layers.11.mlp.switch_mlp.gate_proj.weight",
2689
+ "role": "ffn_gate_exps",
2690
+ "layer_index": 11,
2691
+ "dtype": "u32",
2692
+ "source_quantized": true,
2693
+ "quantization": {
2694
+ "mode": "mxfp8",
2695
+ "group_size": 32,
2696
+ "bits": 8
2697
+ },
2698
+ "shape": [
2699
+ 64,
2700
+ 896,
2701
+ 320
2702
+ ],
2703
+ "file": "model.safetensors",
2704
+ "offset_bytes": 180745113,
2705
+ "length_bytes": 73400320
2706
+ },
2707
+ {
2708
+ "name": "language_model.model.layers.11.mlp.gate.weight",
2709
+ "role": "ffn_gate_inp",
2710
+ "layer_index": 11,
2711
+ "dtype": "bf16",
2712
+ "shape": [
2713
+ 64,
2714
+ 1280
2715
+ ],
2716
+ "file": "model.safetensors",
2717
+ "offset_bytes": 178287513,
2718
+ "length_bytes": 163840
2719
+ },
2720
+ {
2721
+ "name": "language_model.model.layers.11.mlp.shared_experts.down_proj.weight",
2722
+ "role": "ffn_shared_expert_down",
2723
+ "layer_index": 11,
2724
+ "dtype": "u32",
2725
+ "source_quantized": true,
2726
+ "quantization": {
2727
+ "mode": "mxfp8",
2728
+ "group_size": 32,
2729
+ "bits": 8
2730
+ },
2731
+ "shape": [
2732
+ 1280,
2733
+ 448
2734
+ ],
2735
+ "file": "model.safetensors",
2736
+ "offset_bytes": 3515535129,
2737
+ "length_bytes": 2293760
2738
+ },
2739
+ {
2740
+ "name": "language_model.model.layers.11.mlp.shared_experts.gate_proj.weight",
2741
+ "role": "ffn_shared_expert_gate",
2742
+ "layer_index": 11,
2743
+ "dtype": "u32",
2744
+ "source_quantized": true,
2745
+ "quantization": {
2746
+ "mode": "mxfp8",
2747
+ "group_size": 32,
2748
+ "bits": 8
2749
+ },
2750
+ "shape": [
2751
+ 1792,
2752
+ 320
2753
+ ],
2754
+ "file": "model.safetensors",
2755
+ "offset_bytes": 3720423705,
2756
+ "length_bytes": 2293760
2757
+ },
2758
+ {
2759
+ "name": "language_model.model.layers.11.mlp.shared_experts.up_proj.weight",
2760
+ "role": "ffn_shared_expert_up",
2761
+ "layer_index": 11,
2762
+ "dtype": "u32",
2763
+ "source_quantized": true,
2764
+ "quantization": {
2765
+ "mode": "mxfp8",
2766
+ "group_size": 32,
2767
+ "bits": 8
2768
+ },
2769
+ "shape": [
2770
+ 1792,
2771
+ 320
2772
+ ],
2773
+ "file": "model.safetensors",
2774
+ "offset_bytes": 175993753,
2775
+ "length_bytes": 2293760
2776
+ },
2777
+ {
2778
+ "name": "language_model.model.layers.11.mlp.switch_mlp.up_proj.weight",
2779
+ "role": "ffn_up_exps",
2780
+ "layer_index": 11,
2781
+ "dtype": "u32",
2782
+ "source_quantized": true,
2783
+ "quantization": {
2784
+ "mode": "mxfp8",
2785
+ "group_size": 32,
2786
+ "bits": 8
2787
+ },
2788
+ "shape": [
2789
+ 64,
2790
+ 896,
2791
+ 320
2792
+ ],
2793
+ "file": "model.safetensors",
2794
+ "offset_bytes": 255835033,
2795
+ "length_bytes": 73400320
2796
+ }
2797
+ ]
2798
+ }