Add files using upload-large-folder tool
Browse files- .gitattributes +1 -0
- LICENSE +10 -0
- NOTICE +5 -0
- README.md +242 -3
- added_tokens.json +28 -0
- chat_template.json +3 -0
- config.json +73 -0
- configuration_ovis2_5.py +96 -0
- generation_config.json +15 -0
- merges.txt +0 -0
- model-00001-of-00002.safetensors +3 -0
- model-00002-of-00002.safetensors +3 -0
- model.safetensors.index.json +758 -0
- modeling_ovis2_5.py +903 -0
- preprocessor_config.json +24 -0
- special_tokens_map.json +31 -0
- tokenizer.json +3 -0
- tokenizer_config.json +240 -0
- vocab.json +0 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
LICENSE
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Copyright (C) 2025 AIDC-AI
|
| 2 |
+
Licensed under the Apache License, Version 2.0 (the "License");
|
| 3 |
+
you may not use this file except in compliance with the License.
|
| 4 |
+
You may obtain a copy of the License at
|
| 5 |
+
http://www.apache.org/licenses/LICENSE-2.0
|
| 6 |
+
Unless required by applicable law or agreed to in writing, software
|
| 7 |
+
distributed under the License is distributed on an "AS IS" BASIS,
|
| 8 |
+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
| 9 |
+
See the License for the specific language governing permissions and
|
| 10 |
+
limitations under the License.
|
NOTICE
ADDED
|
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
Copyright (C) 2025 AIDC-AI
|
| 2 |
+
Licensed under the Apache 2.0 (the "License").
|
| 3 |
+
The model was trained based on the following models:
|
| 4 |
+
1. Qwen3-1.7B (https://huggingface.co/Qwen/Qwen3-1.7B), license: Apache License 2.0 (https://huggingface.co/Qwen/Qwen3-1.7B/blob/main/LICENSE).
|
| 5 |
+
2. Siglip2 (https://huggingface.co/google/siglip2-so400m-patch16-512), license: Apache License 2.0 (https://choosealicense.com/licenses/apache-2.0/).
|
README.md
CHANGED
|
@@ -1,3 +1,242 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- AIDC-AI/Ovis-dataset
|
| 5 |
+
library_name: transformers
|
| 6 |
+
tags:
|
| 7 |
+
- MLLM
|
| 8 |
+
pipeline_tag: image-text-to-text
|
| 9 |
+
language:
|
| 10 |
+
- en
|
| 11 |
+
- zh
|
| 12 |
+
---
|
| 13 |
+
# Ovis2.5-2B
|
| 14 |
+
<div align="center">
|
| 15 |
+
<img src=https://cdn-uploads.huggingface.co/production/uploads/637aebed7ce76c3b834cea37/3IK823BZ8w-mz_QfeYkDn.png width="30%"/>
|
| 16 |
+
</div>
|
| 17 |
+
|
| 18 |
+
<p align="center">
|
| 19 |
+
<a href="https://arxiv.org/pdf/2405.20797"><img src="https://img.shields.io/badge/📖_Technical_Report-Ovis2.5-b31b1b.svg" alt="technical report"></a>
|
| 20 |
+
<a href="https://github.com/AIDC-AI/Ovis"><img src="https://img.shields.io/badge/GitHub-AIDC--AI/Ovis-blue?style=flat&logo=github" alt="code"></a>
|
| 21 |
+
<a href="https://huggingface.co/spaces/AIDC-AI/Ovis2.5-2B"><img src="https://img.shields.io/badge/🎨_HF_Spaces-AIDC--AI/Ovis2.5--2B-lightblack" alt="demo"></a>
|
| 22 |
+
<a href="https://huggingface.co/collections/AIDC-AI/ovis25-689ec1474633b2aab8809335"><img src="https://img.shields.io/badge/🤗_Models-AIDC--AI/Ovis2.5-yellow" alt="models"></a>
|
| 23 |
+
</p>
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
## Introduction
|
| 27 |
+
|
| 28 |
+
We are pleased to announce the release of **Ovis2.5**, the successor to Ovis2, designed for native-resolution visual perception and enhanced multimodal reasoning.
|
| 29 |
+
It integrates a native-resolution vision transformer (NaViT) that processes images at their original, variable resolutions, eliminating the need for fixed-resolution tiling and preserving both fine details and global layout—crucial for visually dense content such as charts and diagrams.
|
| 30 |
+
To strengthen reasoning, Ovis2.5 is trained not only on linear chain-of-thought (CoT) but also on reflective reasoning, including self-checking and revision.
|
| 31 |
+
This advanced capability is available at inference as an optional *thinking mode*, enabling users to trade latency for higher accuracy on complex inputs.
|
| 32 |
+
|
| 33 |
+
Building on these advances, **Ovis2.5-9B** achieves an average score of 78.3 on the OpenCompass multimodal evaluation suite (SOTA among open-source MLLMs under 40B parameters), while the lightweight **Ovis2.5-2B** scores 73.9, continuing the “small model, big performance” philosophy for resource-constrained scenarios.
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
<div align="center">
|
| 37 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/637aebed7ce76c3b834cea37/kh-1dhZRAduP-P4SkIhXr.png" width="100%" />
|
| 38 |
+
</div>
|
| 39 |
+
|
| 40 |
+
**Key Features**
|
| 41 |
+
* **Native-Resolution Perception** — NaViT vision encoder preserves fine details and global structure without lossy tiling.
|
| 42 |
+
* **Deep-Reasoning Capability** — Optional *thinking mode* for self-checking and revision beyond linear CoT.
|
| 43 |
+
* **Chart & Document OCR** — State-of-the-art at its scale for complex chart analysis, document understanding (including tables and forms), and OCR.
|
| 44 |
+
* **Broad Task Coverage** — Demonstrates leading performance on image reasoning, video understanding, and grounding benchmarks, showcasing strong general multimodal capability.
|
| 45 |
+
|
| 46 |
+
<div align="center">
|
| 47 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/637aebed7ce76c3b834cea37/4kw2RRUhXDiMZdU7wGOfP.png" width="100%" />
|
| 48 |
+
</div>
|
| 49 |
+
|
| 50 |
+
## Quick Inference
|
| 51 |
+
Below is a simple example demonstrating how to run Ovis2.5 with a single image input.
|
| 52 |
+
|
| 53 |
+
First, install the required dependencies:
|
| 54 |
+
```bash
|
| 55 |
+
pip install torch==2.4.0 transformers==4.51.3 numpy==1.25.0 pillow==10.3.0 moviepy==1.0.3
|
| 56 |
+
pip install flash-attn==2.7.0.post2 --no-build-isolation
|
| 57 |
+
```
|
| 58 |
+
Then, run the following code:
|
| 59 |
+
```python
|
| 60 |
+
import torch
|
| 61 |
+
import requests
|
| 62 |
+
from PIL import Image
|
| 63 |
+
from transformers import AutoModelForCausalLM
|
| 64 |
+
|
| 65 |
+
MODEL_PATH = "AIDC-AI/Ovis2.5-2B"
|
| 66 |
+
THINKING = True # Controls whether to enable thinking mode
|
| 67 |
+
|
| 68 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 69 |
+
MODEL_PATH,
|
| 70 |
+
torch_dtype=torch.bfloat16,
|
| 71 |
+
trust_remote_code=True
|
| 72 |
+
).cuda()
|
| 73 |
+
|
| 74 |
+
messages = [{
|
| 75 |
+
"role": "user",
|
| 76 |
+
"content": [
|
| 77 |
+
{"type": "image", "image": Image.open(requests.get("https://cdn-uploads.huggingface.co/production/uploads/658a8a837959448ef5500ce5/TIlymOb86R6_Mez3bpmcB.png", stream=True).raw)},
|
| 78 |
+
{"type": "text", "text": "Calculate the sum of the numbers in the middle box in figure (c)."},
|
| 79 |
+
],
|
| 80 |
+
}]
|
| 81 |
+
|
| 82 |
+
input_ids, pixel_values, grid_thws = model.preprocess_inputs(
|
| 83 |
+
messages=messages,
|
| 84 |
+
add_generation_prompt=True,
|
| 85 |
+
enable_thinking=THINKING
|
| 86 |
+
)
|
| 87 |
+
input_ids = input_ids.cuda()
|
| 88 |
+
pixel_values = pixel_values.cuda() if pixel_values is not None else None
|
| 89 |
+
grid_thws = grid_thws.cuda() if grid_thws is not None else None
|
| 90 |
+
|
| 91 |
+
outputs = model.generate(
|
| 92 |
+
inputs=input_ids,
|
| 93 |
+
pixel_values=pixel_values,
|
| 94 |
+
grid_thws=grid_thws,
|
| 95 |
+
max_new_tokens=3072
|
| 96 |
+
)
|
| 97 |
+
|
| 98 |
+
response = model.text_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 99 |
+
print(response)
|
| 100 |
+
```
|
| 101 |
+
|
| 102 |
+
<details>
|
| 103 |
+
<summary>Example: Multi-image</summary>
|
| 104 |
+
Demonstrates how to run inference with multiple images and a related question.
|
| 105 |
+
|
| 106 |
+
```python
|
| 107 |
+
# Multi-image inference
|
| 108 |
+
multi_image_files = [
|
| 109 |
+
"/path/to/image_1.jpg",
|
| 110 |
+
"/path/to/image_2.jpg",
|
| 111 |
+
"/path/to/image_3.jpg",
|
| 112 |
+
]
|
| 113 |
+
|
| 114 |
+
content = [{"type": "image", "image": Image.open(p).convert("RGB")} for p in multi_image_files]
|
| 115 |
+
content.append({"type": "text", "text": "Describe the images."})
|
| 116 |
+
messages = [{"role": "user", "content": content}]
|
| 117 |
+
|
| 118 |
+
input_ids, pixel_values, grid_thws = model.preprocess_inputs(messages=messages, add_generation_prompt=True, max_pixels=896*896)
|
| 119 |
+
input_ids = input_ids.cuda()
|
| 120 |
+
pixel_values = pixel_values.cuda().to(model.dtype) if pixel_values is not None else None
|
| 121 |
+
grid_thws = grid_thws.cuda() if grid_thws is not None else None
|
| 122 |
+
|
| 123 |
+
with torch.no_grad():
|
| 124 |
+
outputs = model.generate(inputs=input_ids, pixel_values=pixel_values, grid_thws=grid_thws,
|
| 125 |
+
max_new_tokens=1024, do_sample=True,
|
| 126 |
+
eos_token_id=model.text_tokenizer.eos_token_id,
|
| 127 |
+
pad_token_id=model.text_tokenizer.pad_token_id)
|
| 128 |
+
print(model.text_tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 129 |
+
```
|
| 130 |
+
</details>
|
| 131 |
+
|
| 132 |
+
<details>
|
| 133 |
+
<summary>Example: Video</summary>
|
| 134 |
+
Demonstrates how to run inference on a video by sampling multiple frames and asking the model to describe the content.
|
| 135 |
+
|
| 136 |
+
```python
|
| 137 |
+
# Video inference
|
| 138 |
+
from moviepy.editor import VideoFileClip # pip install moviepy==1.0.3
|
| 139 |
+
|
| 140 |
+
video_file = "/path/to/video_1.mp4"
|
| 141 |
+
num_frames = 8
|
| 142 |
+
|
| 143 |
+
with VideoFileClip(video_file) as clip:
|
| 144 |
+
total_frames = int(clip.fps * clip.duration)
|
| 145 |
+
indices = [int(i * total_frames / num_frames) for i in range(num_frames)]
|
| 146 |
+
frames = [Image.fromarray(clip.get_frame(t)) for t in (idx / clip.fps for idx in indices)]
|
| 147 |
+
|
| 148 |
+
messages = [{"role": "user", "content": [
|
| 149 |
+
{"type": "video", "video": frames},
|
| 150 |
+
{"type": "text", "text": "Describe this video in detail."},
|
| 151 |
+
]}]
|
| 152 |
+
|
| 153 |
+
input_ids, pixel_values, grid_thws = model.preprocess_inputs(messages=messages, add_generation_prompt=True, max_pixels=896*896)
|
| 154 |
+
input_ids = input_ids.cuda()
|
| 155 |
+
pixel_values = pixel_values.cuda().to(model.dtype) if pixel_values is not None else None
|
| 156 |
+
grid_thws = grid_thws.cuda() if grid_thws is not None else None
|
| 157 |
+
|
| 158 |
+
with torch.no_grad():
|
| 159 |
+
outputs = model.generate(inputs=input_ids, pixel_values=pixel_values, grid_thws=grid_thws,
|
| 160 |
+
max_new_tokens=1024, do_sample=True,
|
| 161 |
+
eos_token_id=model.text_tokenizer.eos_token_id,
|
| 162 |
+
pad_token_id=model.text_tokenizer.pad_token_id)
|
| 163 |
+
print(model.text_tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 164 |
+
```
|
| 165 |
+
|
| 166 |
+
</details>
|
| 167 |
+
|
| 168 |
+
<details>
|
| 169 |
+
<summary>Example: Text-only</summary>
|
| 170 |
+
Demonstrates how to run inference using only text input without any images or videos.
|
| 171 |
+
|
| 172 |
+
```python
|
| 173 |
+
# Text-only inference
|
| 174 |
+
messages = [{"role": "user", "content": "Hi, please introduce Yellow Mountain."}]
|
| 175 |
+
|
| 176 |
+
input_ids, _, _ = model.preprocess_inputs(messages=messages, add_generation_prompt=True)
|
| 177 |
+
input_ids = input_ids.cuda()
|
| 178 |
+
|
| 179 |
+
with torch.no_grad():
|
| 180 |
+
outputs = model.generate(inputs=input_ids, max_new_tokens=1024, do_sample=True,
|
| 181 |
+
eos_token_id=model.text_tokenizer.eos_token_id,
|
| 182 |
+
pad_token_id=model.text_tokenizer.pad_token_id)
|
| 183 |
+
print(model.text_tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 184 |
+
```
|
| 185 |
+
|
| 186 |
+
</details>
|
| 187 |
+
|
| 188 |
+
To enable grounding, end your prompt with `Please provide the bounding box coordinates.` (for boxes) or `Please provide the point coordinates.` (for points). To target a specific object, wrap its description in `<ref>` tags, e.g.:
|
| 189 |
+
|
| 190 |
+
```text
|
| 191 |
+
Find the <ref>red apple</ref> in the image. Please provide the bounding box coordinates.
|
| 192 |
+
```
|
| 193 |
+
|
| 194 |
+
Coordinates are normalized to `[0,1)` with the origin `(0,0)` at the top-left corner of the image.
|
| 195 |
+
|
| 196 |
+
* Point: `<point>(x,y)</point>`
|
| 197 |
+
* Bounding box: `<box>(x1,y1),(x2,y2)</box>` where `(x1,y1)` is top-left, `(x2,y2)` is bottom-right.
|
| 198 |
+
* Multiple results can be listed in square brackets: `[<box>(...),<box>(...) ]`
|
| 199 |
+
|
| 200 |
+
Example:
|
| 201 |
+
|
| 202 |
+
```text
|
| 203 |
+
The image features a serene scene with <ref>three birds</ref>[
|
| 204 |
+
<box>(0.401,0.526),(0.430,0.557)</box>,
|
| 205 |
+
<box>(0.489,0.494),(0.516,0.526)</box>,
|
| 206 |
+
<box>(0.296,0.529),(0.324,0.576)</box>
|
| 207 |
+
] flying in formation against a clear blue sky.
|
| 208 |
+
```
|
| 209 |
+
|
| 210 |
+
|
| 211 |
+
|
| 212 |
+
## Model Zoo
|
| 213 |
+
|
| 214 |
+
| Ovis MLLMs | ViT | LLM | Model Weights | Demo |
|
| 215 |
+
|:-----------|:-----------------------:|:---------------------:|:-------------------------------------------------------:|:--------------------------------------------------------:|
|
| 216 |
+
| Ovis2.5-2B | siglip2-so400m-patch16-512 | Qwen3-1.7B | [Huggingface](https://huggingface.co/AIDC-AI/Ovis2.5-2B) | [Space](https://huggingface.co/spaces/AIDC-AI/Ovis2.5-2B) |
|
| 217 |
+
| Ovis2.5-9B | siglip2-so400m-patch16-512 | Qwen3-8B | [Huggingface](https://huggingface.co/AIDC-AI/Ovis2.5-9B) | [Space](https://huggingface.co/spaces/AIDC-AI/Ovis2.5-9B) |
|
| 218 |
+
|
| 219 |
+
## Performance
|
| 220 |
+
We evaluate Ovis2.5 using [VLMEvalKit](https://github.com/open-compass/VLMEvalKit), as employed in the OpenCompass multimodal and reasoning evaluation suite.
|
| 221 |
+
|
| 222 |
+
|
| 223 |
+
|
| 224 |
+
|
| 225 |
+
|
| 226 |
+
|
| 227 |
+
## Citation
|
| 228 |
+
If you find Ovis useful, please consider citing the paper
|
| 229 |
+
```
|
| 230 |
+
@article{lu2024ovis,
|
| 231 |
+
title={Ovis: Structural Embedding Alignment for Multimodal Large Language Model},
|
| 232 |
+
author={Shiyin Lu and Yang Li and Qing-Guo Chen and Zhao Xu and Weihua Luo and Kaifu Zhang and Han-Jia Ye},
|
| 233 |
+
year={2024},
|
| 234 |
+
journal={arXiv:2405.20797}
|
| 235 |
+
}
|
| 236 |
+
```
|
| 237 |
+
|
| 238 |
+
## License
|
| 239 |
+
This project is licensed under the [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0.txt) (SPDX-License-Identifier: Apache-2.0).
|
| 240 |
+
|
| 241 |
+
## Disclaimer
|
| 242 |
+
We used compliance-checking algorithms during the training process, to ensure the compliance of the trained model to the best of our ability. Due to the complexity of the data and the diversity of language model usage scenarios, we cannot guarantee that the model is completely free of copyright issues or improper content. If you believe anything infringes on your rights or generates improper content, please contact us, and we will promptly address the matter.
|
added_tokens.json
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"</think>": 151668,
|
| 3 |
+
"</tool_call>": 151658,
|
| 4 |
+
"</tool_response>": 151666,
|
| 5 |
+
"<think>": 151667,
|
| 6 |
+
"<tool_call>": 151657,
|
| 7 |
+
"<tool_response>": 151665,
|
| 8 |
+
"<|box_end|>": 151649,
|
| 9 |
+
"<|box_start|>": 151648,
|
| 10 |
+
"<|endoftext|>": 151643,
|
| 11 |
+
"<|file_sep|>": 151664,
|
| 12 |
+
"<|fim_middle|>": 151660,
|
| 13 |
+
"<|fim_pad|>": 151662,
|
| 14 |
+
"<|fim_prefix|>": 151659,
|
| 15 |
+
"<|fim_suffix|>": 151661,
|
| 16 |
+
"<|im_end|>": 151645,
|
| 17 |
+
"<|im_start|>": 151644,
|
| 18 |
+
"<|image_pad|>": 151655,
|
| 19 |
+
"<|object_ref_end|>": 151647,
|
| 20 |
+
"<|object_ref_start|>": 151646,
|
| 21 |
+
"<|quad_end|>": 151651,
|
| 22 |
+
"<|quad_start|>": 151650,
|
| 23 |
+
"<|repo_name|>": 151663,
|
| 24 |
+
"<|video_pad|>": 151656,
|
| 25 |
+
"<|vision_end|>": 151653,
|
| 26 |
+
"<|vision_pad|>": 151654,
|
| 27 |
+
"<|vision_start|>": 151652
|
| 28 |
+
}
|
chat_template.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"chat_template": "{%- for message in messages %}{{- '<|im_start|>' + message.role + '\n'}}{%- if message.role == 'system' or message.role == 'user' %}{%- if message.content is string %}{{- message.content | replace('<image>', '') | replace('<video>', '') }}{%- else %}{%- for item in message.content %}{%- if item.type == 'text' and 'text' in item %}{{- item.text | replace('<image>', '') | replace('<video>', '') }}{%- elif item.type == 'image' and 'image' in item %}{{- '<image>'}}{%- elif item.type == 'video' and 'video' in item %}{{- '<video>'}}{%- else %}{{- raise_exception('Invalid content type. Supported types for system and user are text, image, video.')}}{%- endif %}{%- if not loop.last %}{{- '\n'}}{%- endif %}{%- endfor %}{%- endif %}{%- elif message.role == 'assistant' %}{%- set content = '' %}{%- if message.content is string %}{%- set content = message.content | replace('<image>', '') | replace('<video>', '') %}{%- else %}{%- for item in message.content %}{%- if item.type == 'text' and 'text' in item %}{%- set content = content ~ (item.text | replace('<image>', '') | replace('<video>', '')) %}{%- else %}{{- raise_exception('Invalid content type. Supported type for assistant is text.')}}{%- endif %}{%- endfor %}{%- endif %}{%- set content = content.split('</think>')[-1].lstrip('\n') %}{{- content }}{%- else %}{{- raise_exception('Invalid role. Supported roles are system, user, assistant.')}}{%- endif %}{{- '<|im_end|>\n'}}{%- endfor %}{%- if add_generation_prompt %}{{- '<|im_start|>assistant\n' }}{%- if enable_thinking is defined and enable_thinking is false %}{{- '<think>\n\n</think>\n\n' }}{%- endif %}{%- endif %}"
|
| 3 |
+
}
|
config.json
ADDED
|
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Ovis2_5"
|
| 4 |
+
],
|
| 5 |
+
"auto_map": {
|
| 6 |
+
"AutoConfig": "configuration_ovis2_5.Ovis2_5_Config",
|
| 7 |
+
"AutoModelForCausalLM": "modeling_ovis2_5.Ovis2_5"
|
| 8 |
+
},
|
| 9 |
+
"conversation_formatter_class": "Qwen3ConversationFormatter",
|
| 10 |
+
"hidden_size": 2048,
|
| 11 |
+
"vocab_size": 151936,
|
| 12 |
+
"num_attention_heads": 32,
|
| 13 |
+
"max_position_embeddings": 40960,
|
| 14 |
+
"llm_config": {
|
| 15 |
+
"_attn_implementation_autoset": true,
|
| 16 |
+
"_name_or_path": "Qwen/Qwen3-1.7B",
|
| 17 |
+
"architectures": [
|
| 18 |
+
"Qwen3ForCausalLM"
|
| 19 |
+
],
|
| 20 |
+
"attention_bias": false,
|
| 21 |
+
"attention_dropout": 0.0,
|
| 22 |
+
"bos_token_id": 151643,
|
| 23 |
+
"eos_token_id": 151645,
|
| 24 |
+
"head_dim": 128,
|
| 25 |
+
"hidden_act": "silu",
|
| 26 |
+
"hidden_size": 2048,
|
| 27 |
+
"initializer_range": 0.02,
|
| 28 |
+
"intermediate_size": 6144,
|
| 29 |
+
"max_position_embeddings": 40960,
|
| 30 |
+
"max_window_layers": 28,
|
| 31 |
+
"model_type": "qwen3",
|
| 32 |
+
"num_attention_heads": 16,
|
| 33 |
+
"num_hidden_layers": 28,
|
| 34 |
+
"num_key_value_heads": 8,
|
| 35 |
+
"rms_norm_eps": 1e-06,
|
| 36 |
+
"rope_scaling": null,
|
| 37 |
+
"rope_theta": 1000000,
|
| 38 |
+
"sliding_window": null,
|
| 39 |
+
"tie_word_embeddings": true,
|
| 40 |
+
"torch_dtype": "float32",
|
| 41 |
+
"use_cache": true,
|
| 42 |
+
"use_sliding_window": false,
|
| 43 |
+
"vocab_size": 151936
|
| 44 |
+
},
|
| 45 |
+
"model_type": "ovis2_5",
|
| 46 |
+
"torch_dtype": "bfloat16",
|
| 47 |
+
"transformers_version": "4.51.3",
|
| 48 |
+
"use_cache": true,
|
| 49 |
+
"visual_vocab_size": 65536,
|
| 50 |
+
"vit_config": {
|
| 51 |
+
"_attn_implementation_autoset": true,
|
| 52 |
+
"_name_or_path": "google/siglip2-so400m-patch16-512",
|
| 53 |
+
"attention_dropout": 0.0,
|
| 54 |
+
"fullatt_block_indexes": null,
|
| 55 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 56 |
+
"hidden_size": 1152,
|
| 57 |
+
"hidden_stride": 2,
|
| 58 |
+
"image_size": 512,
|
| 59 |
+
"intermediate_size": 4304,
|
| 60 |
+
"layer_norm_eps": 1e-06,
|
| 61 |
+
"model_type": "siglip2_navit",
|
| 62 |
+
"num_attention_heads": 16,
|
| 63 |
+
"num_channels": 3,
|
| 64 |
+
"num_hidden_layers": 27,
|
| 65 |
+
"num_patches": -1,
|
| 66 |
+
"patch_size": 16,
|
| 67 |
+
"preserve_original_pe": true,
|
| 68 |
+
"temporal_patch_size": 1,
|
| 69 |
+
"torch_dtype": "float32",
|
| 70 |
+
"use_rope": true,
|
| 71 |
+
"window_size": 112
|
| 72 |
+
}
|
| 73 |
+
}
|
configuration_ovis2_5.py
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import Any, Optional, List, Union
|
| 2 |
+
|
| 3 |
+
from transformers import Qwen3Config
|
| 4 |
+
from transformers.configuration_utils import PretrainedConfig
|
| 5 |
+
|
| 6 |
+
__all__ = ["Siglip2NavitConfig", "Ovis2_5_Config"]
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
class Siglip2NavitConfig(PretrainedConfig):
|
| 10 |
+
"""This is the configuration class to store the configuration of an [`AIMv2Model`].
|
| 11 |
+
|
| 12 |
+
Instantiating a configuration with the defaults will yield a similar configuration
|
| 13 |
+
to that of the [apple/aimv2-large-patch14-224](https://huggingface.co/apple/aimv2-large-patch14-224).
|
| 14 |
+
|
| 15 |
+
Args:
|
| 16 |
+
hidden_size: Dimension of the hidden representations.
|
| 17 |
+
intermediate_size: Dimension of the SwiGLU representations.
|
| 18 |
+
num_hidden_layers: Number of hidden layers in the Transformer.
|
| 19 |
+
num_attention_heads: Number of attention heads for each attention layer
|
| 20 |
+
in the Transformer.
|
| 21 |
+
num_channels: Number of input channels.
|
| 22 |
+
image_size: Image size.
|
| 23 |
+
patch_size: Patch size.
|
| 24 |
+
rms_norm_eps: Epsilon value used for the RMS normalization layer.
|
| 25 |
+
attention_dropout: Dropout ratio for attention probabilities.
|
| 26 |
+
projection_dropout: Dropout ratio for the projection layer after the attention.
|
| 27 |
+
qkv_bias: Whether to add a bias to the queries, keys and values.
|
| 28 |
+
use_bias: Whether to add a bias in the feed-forward and projection layers.
|
| 29 |
+
kwargs: Keyword arguments for the [`PretrainedConfig`].
|
| 30 |
+
"""
|
| 31 |
+
|
| 32 |
+
model_type: str = "siglip2_navit"
|
| 33 |
+
|
| 34 |
+
def __init__(
|
| 35 |
+
self,
|
| 36 |
+
hidden_size: int = 1024,
|
| 37 |
+
intermediate_size: int = 4096,
|
| 38 |
+
num_hidden_layers: int = 24,
|
| 39 |
+
num_attention_heads: int = 16,
|
| 40 |
+
num_channels: int = 3,
|
| 41 |
+
num_patches: int = -1,
|
| 42 |
+
image_size: int = 512,
|
| 43 |
+
patch_size: int = 16,
|
| 44 |
+
hidden_act: str="gelu_pytorch_tanh",
|
| 45 |
+
layer_norm_eps: float = 1e-6,
|
| 46 |
+
attention_dropout: float = 0.0,
|
| 47 |
+
hidden_stride: int = 2,
|
| 48 |
+
window_size: int = 112,
|
| 49 |
+
fullatt_block_indexes: Optional[list] = None,
|
| 50 |
+
temporal_patch_size: int = 1,
|
| 51 |
+
preserve_original_pe: bool = True,
|
| 52 |
+
use_rope: bool = True,
|
| 53 |
+
**kwargs: Any,
|
| 54 |
+
):
|
| 55 |
+
super().__init__(**kwargs)
|
| 56 |
+
self.hidden_size = hidden_size
|
| 57 |
+
self.intermediate_size = intermediate_size
|
| 58 |
+
self.num_hidden_layers = num_hidden_layers
|
| 59 |
+
self.num_attention_heads = num_attention_heads
|
| 60 |
+
self.num_channels = num_channels
|
| 61 |
+
self.num_patches = num_patches
|
| 62 |
+
self.patch_size = patch_size
|
| 63 |
+
self.image_size = image_size
|
| 64 |
+
self.hidden_act = hidden_act
|
| 65 |
+
self.attention_dropout = attention_dropout
|
| 66 |
+
self.layer_norm_eps = layer_norm_eps
|
| 67 |
+
self.hidden_stride = hidden_stride
|
| 68 |
+
self.window_size = window_size
|
| 69 |
+
self.fullatt_block_indexes = fullatt_block_indexes
|
| 70 |
+
self.temporal_patch_size = temporal_patch_size
|
| 71 |
+
self.preserve_original_pe = preserve_original_pe
|
| 72 |
+
self.use_rope = use_rope
|
| 73 |
+
|
| 74 |
+
class Ovis2_5_Config(PretrainedConfig):
|
| 75 |
+
model_type = "ovis2_5"
|
| 76 |
+
sub_configs = dict(llm_config=Qwen3Config, vit_config=Siglip2NavitConfig)
|
| 77 |
+
|
| 78 |
+
def __init__(self,
|
| 79 |
+
llm_config: Optional[Union[Qwen3Config, dict]] = None,
|
| 80 |
+
vit_config: Optional[Union[Siglip2NavitConfig, dict]] = None,
|
| 81 |
+
visual_vocab_size=65536,
|
| 82 |
+
hidden_size=None,
|
| 83 |
+
**kwargs
|
| 84 |
+
):
|
| 85 |
+
super().__init__(**kwargs)
|
| 86 |
+
if isinstance(llm_config, dict):
|
| 87 |
+
llm_config = Qwen3Config(**llm_config)
|
| 88 |
+
self.llm_config = llm_config
|
| 89 |
+
if isinstance(vit_config, dict):
|
| 90 |
+
vit_config = Siglip2NavitConfig(**vit_config)
|
| 91 |
+
self.vit_config = vit_config
|
| 92 |
+
self.visual_vocab_size = visual_vocab_size
|
| 93 |
+
self.hidden_size = hidden_size
|
| 94 |
+
if kwargs.get('attn_implementation'):
|
| 95 |
+
self.llm_config._attn_implementation = kwargs['attn_implementation']
|
| 96 |
+
self.vit_config._attn_implementation = kwargs['attn_implementation']
|
generation_config.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 151643,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
151645,
|
| 6 |
+
151643
|
| 7 |
+
],
|
| 8 |
+
"multimodal_max_length": 8192,
|
| 9 |
+
"pad_token_id": 151643,
|
| 10 |
+
"repetition_penalty": 1.05,
|
| 11 |
+
"temperature": 0.6,
|
| 12 |
+
"top_k": 20,
|
| 13 |
+
"top_p": 0.95,
|
| 14 |
+
"transformers_version": "4.51.3"
|
| 15 |
+
}
|
merges.txt
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
model-00001-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4bd7cec79fedabcd8149c62934fdffff7d840de111e8a17335d346bcbeba0c12
|
| 3 |
+
size 4872524976
|
model-00002-of-00002.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0bb2d37fcbf03254938b72d4bf88da9f84c8c0307b4f7d6ee79aa0a3c2f7732d
|
| 3 |
+
size 268435576
|
model.safetensors.index.json
ADDED
|
@@ -0,0 +1,758 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_size": 5140859344
|
| 4 |
+
},
|
| 5 |
+
"weight_map": {
|
| 6 |
+
"llm.model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
| 7 |
+
"llm.model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 8 |
+
"llm.model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 9 |
+
"llm.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 10 |
+
"llm.model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 11 |
+
"llm.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 12 |
+
"llm.model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 13 |
+
"llm.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 14 |
+
"llm.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 15 |
+
"llm.model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 16 |
+
"llm.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 17 |
+
"llm.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 18 |
+
"llm.model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 19 |
+
"llm.model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 20 |
+
"llm.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 21 |
+
"llm.model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 22 |
+
"llm.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 23 |
+
"llm.model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 24 |
+
"llm.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 25 |
+
"llm.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 26 |
+
"llm.model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 27 |
+
"llm.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 28 |
+
"llm.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 29 |
+
"llm.model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 30 |
+
"llm.model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 31 |
+
"llm.model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 32 |
+
"llm.model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 33 |
+
"llm.model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 34 |
+
"llm.model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 35 |
+
"llm.model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 36 |
+
"llm.model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 37 |
+
"llm.model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 38 |
+
"llm.model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 39 |
+
"llm.model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 40 |
+
"llm.model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 41 |
+
"llm.model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 42 |
+
"llm.model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 43 |
+
"llm.model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 44 |
+
"llm.model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 45 |
+
"llm.model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 46 |
+
"llm.model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 47 |
+
"llm.model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 48 |
+
"llm.model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 49 |
+
"llm.model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 50 |
+
"llm.model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 51 |
+
"llm.model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 52 |
+
"llm.model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 53 |
+
"llm.model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 54 |
+
"llm.model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 55 |
+
"llm.model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 56 |
+
"llm.model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 57 |
+
"llm.model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 58 |
+
"llm.model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 59 |
+
"llm.model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 60 |
+
"llm.model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 61 |
+
"llm.model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 62 |
+
"llm.model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 63 |
+
"llm.model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 64 |
+
"llm.model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 65 |
+
"llm.model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 66 |
+
"llm.model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 67 |
+
"llm.model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 68 |
+
"llm.model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 69 |
+
"llm.model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 70 |
+
"llm.model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 71 |
+
"llm.model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 72 |
+
"llm.model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 73 |
+
"llm.model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 74 |
+
"llm.model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 75 |
+
"llm.model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 76 |
+
"llm.model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 77 |
+
"llm.model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 78 |
+
"llm.model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 79 |
+
"llm.model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 80 |
+
"llm.model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 81 |
+
"llm.model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 82 |
+
"llm.model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 83 |
+
"llm.model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 84 |
+
"llm.model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 85 |
+
"llm.model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 86 |
+
"llm.model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 87 |
+
"llm.model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 88 |
+
"llm.model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 89 |
+
"llm.model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 90 |
+
"llm.model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 91 |
+
"llm.model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 92 |
+
"llm.model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 93 |
+
"llm.model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 94 |
+
"llm.model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 95 |
+
"llm.model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 96 |
+
"llm.model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 97 |
+
"llm.model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 98 |
+
"llm.model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 99 |
+
"llm.model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 100 |
+
"llm.model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 101 |
+
"llm.model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 102 |
+
"llm.model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 103 |
+
"llm.model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 104 |
+
"llm.model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 105 |
+
"llm.model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 106 |
+
"llm.model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 107 |
+
"llm.model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 108 |
+
"llm.model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 109 |
+
"llm.model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 110 |
+
"llm.model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 111 |
+
"llm.model.layers.17.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 112 |
+
"llm.model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 113 |
+
"llm.model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 114 |
+
"llm.model.layers.17.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 115 |
+
"llm.model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 116 |
+
"llm.model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 117 |
+
"llm.model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 118 |
+
"llm.model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 119 |
+
"llm.model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 120 |
+
"llm.model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 121 |
+
"llm.model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 122 |
+
"llm.model.layers.18.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 123 |
+
"llm.model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 124 |
+
"llm.model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 125 |
+
"llm.model.layers.18.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 126 |
+
"llm.model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 127 |
+
"llm.model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 128 |
+
"llm.model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 129 |
+
"llm.model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 130 |
+
"llm.model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 131 |
+
"llm.model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 132 |
+
"llm.model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 133 |
+
"llm.model.layers.19.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 134 |
+
"llm.model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 135 |
+
"llm.model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 136 |
+
"llm.model.layers.19.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 137 |
+
"llm.model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 138 |
+
"llm.model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 139 |
+
"llm.model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 140 |
+
"llm.model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 141 |
+
"llm.model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 142 |
+
"llm.model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 143 |
+
"llm.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 144 |
+
"llm.model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 145 |
+
"llm.model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 146 |
+
"llm.model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 147 |
+
"llm.model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 148 |
+
"llm.model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 149 |
+
"llm.model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 150 |
+
"llm.model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 151 |
+
"llm.model.layers.20.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 152 |
+
"llm.model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 153 |
+
"llm.model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 154 |
+
"llm.model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 155 |
+
"llm.model.layers.20.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 156 |
+
"llm.model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 157 |
+
"llm.model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 158 |
+
"llm.model.layers.20.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 159 |
+
"llm.model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 160 |
+
"llm.model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 161 |
+
"llm.model.layers.21.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 162 |
+
"llm.model.layers.21.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 163 |
+
"llm.model.layers.21.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 164 |
+
"llm.model.layers.21.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 165 |
+
"llm.model.layers.21.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 166 |
+
"llm.model.layers.21.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 167 |
+
"llm.model.layers.21.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 168 |
+
"llm.model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 169 |
+
"llm.model.layers.21.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 170 |
+
"llm.model.layers.21.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 171 |
+
"llm.model.layers.21.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 172 |
+
"llm.model.layers.22.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 173 |
+
"llm.model.layers.22.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 174 |
+
"llm.model.layers.22.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 175 |
+
"llm.model.layers.22.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 176 |
+
"llm.model.layers.22.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 177 |
+
"llm.model.layers.22.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 178 |
+
"llm.model.layers.22.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 179 |
+
"llm.model.layers.22.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 180 |
+
"llm.model.layers.22.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 181 |
+
"llm.model.layers.22.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 182 |
+
"llm.model.layers.22.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 183 |
+
"llm.model.layers.23.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 184 |
+
"llm.model.layers.23.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 185 |
+
"llm.model.layers.23.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 186 |
+
"llm.model.layers.23.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 187 |
+
"llm.model.layers.23.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 188 |
+
"llm.model.layers.23.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 189 |
+
"llm.model.layers.23.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 190 |
+
"llm.model.layers.23.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 191 |
+
"llm.model.layers.23.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 192 |
+
"llm.model.layers.23.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 193 |
+
"llm.model.layers.23.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 194 |
+
"llm.model.layers.24.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 195 |
+
"llm.model.layers.24.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 196 |
+
"llm.model.layers.24.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 197 |
+
"llm.model.layers.24.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 198 |
+
"llm.model.layers.24.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 199 |
+
"llm.model.layers.24.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 200 |
+
"llm.model.layers.24.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 201 |
+
"llm.model.layers.24.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 202 |
+
"llm.model.layers.24.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 203 |
+
"llm.model.layers.24.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 204 |
+
"llm.model.layers.24.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 205 |
+
"llm.model.layers.25.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 206 |
+
"llm.model.layers.25.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 207 |
+
"llm.model.layers.25.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 208 |
+
"llm.model.layers.25.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 209 |
+
"llm.model.layers.25.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 210 |
+
"llm.model.layers.25.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 211 |
+
"llm.model.layers.25.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 212 |
+
"llm.model.layers.25.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 213 |
+
"llm.model.layers.25.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 214 |
+
"llm.model.layers.25.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 215 |
+
"llm.model.layers.25.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 216 |
+
"llm.model.layers.26.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 217 |
+
"llm.model.layers.26.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 218 |
+
"llm.model.layers.26.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 219 |
+
"llm.model.layers.26.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 220 |
+
"llm.model.layers.26.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 221 |
+
"llm.model.layers.26.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 222 |
+
"llm.model.layers.26.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 223 |
+
"llm.model.layers.26.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 224 |
+
"llm.model.layers.26.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 225 |
+
"llm.model.layers.26.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 226 |
+
"llm.model.layers.26.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 227 |
+
"llm.model.layers.27.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 228 |
+
"llm.model.layers.27.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 229 |
+
"llm.model.layers.27.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 230 |
+
"llm.model.layers.27.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 231 |
+
"llm.model.layers.27.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 232 |
+
"llm.model.layers.27.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 233 |
+
"llm.model.layers.27.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 234 |
+
"llm.model.layers.27.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 235 |
+
"llm.model.layers.27.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 236 |
+
"llm.model.layers.27.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 237 |
+
"llm.model.layers.27.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 238 |
+
"llm.model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 239 |
+
"llm.model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 240 |
+
"llm.model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 241 |
+
"llm.model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 242 |
+
"llm.model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 243 |
+
"llm.model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 244 |
+
"llm.model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 245 |
+
"llm.model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 246 |
+
"llm.model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 247 |
+
"llm.model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 248 |
+
"llm.model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 249 |
+
"llm.model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 250 |
+
"llm.model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 251 |
+
"llm.model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 252 |
+
"llm.model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 253 |
+
"llm.model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 254 |
+
"llm.model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 255 |
+
"llm.model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 256 |
+
"llm.model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 257 |
+
"llm.model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 258 |
+
"llm.model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 259 |
+
"llm.model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 260 |
+
"llm.model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 261 |
+
"llm.model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 262 |
+
"llm.model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 263 |
+
"llm.model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 264 |
+
"llm.model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 265 |
+
"llm.model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 266 |
+
"llm.model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 267 |
+
"llm.model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 268 |
+
"llm.model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 269 |
+
"llm.model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 270 |
+
"llm.model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 271 |
+
"llm.model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 272 |
+
"llm.model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 273 |
+
"llm.model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 274 |
+
"llm.model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 275 |
+
"llm.model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 276 |
+
"llm.model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 277 |
+
"llm.model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 278 |
+
"llm.model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 279 |
+
"llm.model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 280 |
+
"llm.model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 281 |
+
"llm.model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 282 |
+
"llm.model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 283 |
+
"llm.model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 284 |
+
"llm.model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 285 |
+
"llm.model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 286 |
+
"llm.model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 287 |
+
"llm.model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 288 |
+
"llm.model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 289 |
+
"llm.model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 290 |
+
"llm.model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 291 |
+
"llm.model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 292 |
+
"llm.model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 293 |
+
"llm.model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 294 |
+
"llm.model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 295 |
+
"llm.model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 296 |
+
"llm.model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 297 |
+
"llm.model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 298 |
+
"llm.model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 299 |
+
"llm.model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 300 |
+
"llm.model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 301 |
+
"llm.model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 302 |
+
"llm.model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 303 |
+
"llm.model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 304 |
+
"llm.model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 305 |
+
"llm.model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
| 306 |
+
"llm.model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
| 307 |
+
"llm.model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
| 308 |
+
"llm.model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 309 |
+
"llm.model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
| 310 |
+
"llm.model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 311 |
+
"llm.model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
| 312 |
+
"llm.model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
| 313 |
+
"llm.model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 314 |
+
"llm.model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 315 |
+
"llm.model.norm.weight": "model-00001-of-00002.safetensors",
|
| 316 |
+
"visual_tokenizer.head.0.weight": "model-00001-of-00002.safetensors",
|
| 317 |
+
"visual_tokenizer.head.1.bias": "model-00001-of-00002.safetensors",
|
| 318 |
+
"visual_tokenizer.head.1.weight": "model-00001-of-00002.safetensors",
|
| 319 |
+
"visual_tokenizer.vit.vision_model.embeddings.patch_embedding.bias": "model-00001-of-00002.safetensors",
|
| 320 |
+
"visual_tokenizer.vit.vision_model.embeddings.patch_embedding.weight": "model-00001-of-00002.safetensors",
|
| 321 |
+
"visual_tokenizer.vit.vision_model.embeddings.position_embedding.weight": "model-00001-of-00002.safetensors",
|
| 322 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 323 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 324 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 325 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 326 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 327 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 328 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 329 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 330 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 331 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 332 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 333 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 334 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 335 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 336 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 337 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 338 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 339 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 340 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 341 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 342 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 343 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 344 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 345 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 346 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 347 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 348 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 349 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 350 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 351 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 352 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 353 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 354 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 355 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 356 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 357 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 358 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 359 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 360 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 361 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 362 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 363 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 364 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 365 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 366 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 367 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 368 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 369 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 370 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 371 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 372 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 373 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 374 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 375 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 376 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 377 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 378 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 379 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 380 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 381 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 382 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 383 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 384 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 385 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 386 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 387 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 388 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 389 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 390 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 391 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 392 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 393 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 394 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 395 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 396 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 397 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 398 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 399 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 400 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 401 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 402 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 403 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 404 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 405 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 406 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 407 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 408 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 409 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 410 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 411 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 412 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 413 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 414 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 415 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 416 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 417 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 418 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 419 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 420 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 421 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 422 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 423 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 424 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 425 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 426 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 427 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 428 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 429 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 430 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 431 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 432 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 433 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 434 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 435 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 436 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 437 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 438 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 439 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 440 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 441 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 442 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 443 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 444 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 445 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 446 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 447 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 448 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 449 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 450 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 451 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 452 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 453 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 454 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 455 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 456 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 457 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 458 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 459 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 460 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 461 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 462 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 463 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 464 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 465 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 466 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 467 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 468 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 469 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 470 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 471 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 472 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 473 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 474 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 475 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 476 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 477 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 478 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 479 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 480 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 481 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 482 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 483 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 484 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 485 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 486 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 487 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 488 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 489 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 490 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 491 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 492 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 493 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 494 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 495 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 496 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 497 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 498 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 499 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 500 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 501 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 502 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 503 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 504 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 505 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 506 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 507 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 508 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 509 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 510 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 511 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 512 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 513 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 514 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 515 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 516 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 517 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 518 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 519 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 520 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 521 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 522 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 523 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 524 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 525 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 526 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 527 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 528 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 529 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 530 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 531 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 532 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 533 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 534 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 535 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 536 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 537 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 538 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 539 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 540 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 541 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 542 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 543 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 544 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 545 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 546 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 547 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 548 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 549 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 550 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 551 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 552 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 553 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 554 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 555 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 556 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 557 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 558 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 559 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 560 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 561 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.21.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 562 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 563 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 564 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 565 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 566 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 567 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 568 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 569 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 570 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 571 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 572 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 573 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 574 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 575 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 576 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 577 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.22.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 578 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 579 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 580 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 581 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 582 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 583 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 584 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 585 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 586 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 587 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 588 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 589 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 590 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 591 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 592 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 593 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.23.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 594 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 595 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 596 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 597 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 598 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 599 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 600 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 601 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 602 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 603 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 604 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 605 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 606 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 607 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 608 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 609 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.24.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 610 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 611 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 612 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 613 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 614 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 615 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 616 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 617 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 618 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 619 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 620 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 621 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 622 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 623 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 624 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 625 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.25.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 626 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 627 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 628 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 629 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 630 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 631 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 632 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 633 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 634 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 635 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 636 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 637 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 638 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 639 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 640 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 641 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.26.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 642 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 643 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 644 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 645 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 646 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 647 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 648 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 649 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 650 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 651 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 652 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 653 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 654 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 655 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 656 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 657 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 658 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 659 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 660 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 661 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 662 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 663 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 664 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 665 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 666 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 667 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 668 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 669 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 670 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 671 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 672 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 673 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 674 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 675 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 676 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 677 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 678 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 679 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 680 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 681 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 682 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 683 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 684 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 685 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 686 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 687 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 688 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 689 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 690 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 691 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 692 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 693 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 694 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 695 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 696 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 697 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 698 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 699 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 700 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 701 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 702 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 703 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 704 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 705 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 706 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 707 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 708 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 709 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 710 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 711 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 712 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 713 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 714 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 715 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 716 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 717 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 718 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 719 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 720 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 721 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 722 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 723 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 724 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 725 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 726 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 727 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 728 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 729 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 730 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 731 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 732 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 733 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 734 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 735 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 736 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 737 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 738 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.layer_norm1.bias": "model-00001-of-00002.safetensors",
|
| 739 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.layer_norm1.weight": "model-00001-of-00002.safetensors",
|
| 740 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.layer_norm2.bias": "model-00001-of-00002.safetensors",
|
| 741 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.layer_norm2.weight": "model-00001-of-00002.safetensors",
|
| 742 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.mlp.fc1.bias": "model-00001-of-00002.safetensors",
|
| 743 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.mlp.fc1.weight": "model-00001-of-00002.safetensors",
|
| 744 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.mlp.fc2.bias": "model-00001-of-00002.safetensors",
|
| 745 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.mlp.fc2.weight": "model-00001-of-00002.safetensors",
|
| 746 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.k_proj.bias": "model-00001-of-00002.safetensors",
|
| 747 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
| 748 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.out_proj.bias": "model-00001-of-00002.safetensors",
|
| 749 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.out_proj.weight": "model-00001-of-00002.safetensors",
|
| 750 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.q_proj.bias": "model-00001-of-00002.safetensors",
|
| 751 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
| 752 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.v_proj.bias": "model-00001-of-00002.safetensors",
|
| 753 |
+
"visual_tokenizer.vit.vision_model.encoder.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
| 754 |
+
"visual_tokenizer.vit.vision_model.post_layernorm.bias": "model-00001-of-00002.safetensors",
|
| 755 |
+
"visual_tokenizer.vit.vision_model.post_layernorm.weight": "model-00001-of-00002.safetensors",
|
| 756 |
+
"vte.weight": "model-00002-of-00002.safetensors"
|
| 757 |
+
}
|
| 758 |
+
}
|
modeling_ovis2_5.py
ADDED
|
@@ -0,0 +1,903 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import math
|
| 2 |
+
from typing import Dict, List, Optional, Tuple, Union
|
| 3 |
+
|
| 4 |
+
import PIL.Image
|
| 5 |
+
import numpy as np
|
| 6 |
+
import torch
|
| 7 |
+
from flash_attn import flash_attn_varlen_func
|
| 8 |
+
from flash_attn.layers.rotary import apply_rotary_emb
|
| 9 |
+
from torch import Tensor, nn
|
| 10 |
+
from torch.nn import functional as F
|
| 11 |
+
from transformers import (
|
| 12 |
+
AutoConfig,
|
| 13 |
+
AutoImageProcessor,
|
| 14 |
+
AutoModel,
|
| 15 |
+
AutoModelForCausalLM,
|
| 16 |
+
AutoTokenizer,
|
| 17 |
+
)
|
| 18 |
+
from transformers.activations import ACT2FN
|
| 19 |
+
from transformers.generation.utils import GenerateOutput
|
| 20 |
+
from transformers.modeling_outputs import BaseModelOutputWithNoAttention
|
| 21 |
+
from transformers.modeling_utils import PreTrainedModel
|
| 22 |
+
|
| 23 |
+
from .configuration_ovis2_5 import Siglip2NavitConfig, Ovis2_5_Config
|
| 24 |
+
|
| 25 |
+
IMAGE_PLACEHOLDER = "<image>"
|
| 26 |
+
IMAGE_PLACEHOLDER_ID = -200
|
| 27 |
+
VIDEO_PLACEHOLDER = "<video>"
|
| 28 |
+
VIDEO_PLACEHOLDER_ID = -201
|
| 29 |
+
|
| 30 |
+
VISUAL_ATOM_ID = -300
|
| 31 |
+
INDICATOR_IDS = [-301, -302, -303, -304]
|
| 32 |
+
|
| 33 |
+
# copied from qwen2.5-vl
|
| 34 |
+
class VisionRotaryEmbedding(nn.Module):
|
| 35 |
+
def __init__(self, dim: int, theta: float = 10000.0) -> None:
|
| 36 |
+
super().__init__()
|
| 37 |
+
inv_freq = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float) / dim))
|
| 38 |
+
self.register_buffer("inv_freq", inv_freq, persistent=False)
|
| 39 |
+
|
| 40 |
+
def forward(self, seqlen: int) -> torch.Tensor:
|
| 41 |
+
seq = torch.arange(seqlen, device=self.inv_freq.device, dtype=self.inv_freq.dtype)
|
| 42 |
+
freqs = torch.outer(seq, self.inv_freq)
|
| 43 |
+
return freqs
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
class Siglip2VisionEmbeddings(nn.Module):
|
| 47 |
+
def __init__(self, config: Siglip2NavitConfig):
|
| 48 |
+
super().__init__()
|
| 49 |
+
self.config = config
|
| 50 |
+
self.embed_dim = config.hidden_size
|
| 51 |
+
self.patch_size = config.patch_size
|
| 52 |
+
self.image_size = config.image_size
|
| 53 |
+
self.num_patches = config.num_patches
|
| 54 |
+
self.preserve_original_pe = config.preserve_original_pe
|
| 55 |
+
self.hidden_stride = config.hidden_stride
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
# siglip2 naflex
|
| 59 |
+
if self.num_patches > 0:
|
| 60 |
+
self.patch_embedding = nn.Linear(
|
| 61 |
+
in_features=config.num_channels * self.patch_size * self.patch_size,
|
| 62 |
+
out_features=self.embed_dim,
|
| 63 |
+
)
|
| 64 |
+
if self.preserve_original_pe:
|
| 65 |
+
self.position_embedding_size = int(self.num_patches**0.5)
|
| 66 |
+
self.position_embedding = nn.Embedding(self.num_patches, self.embed_dim)
|
| 67 |
+
|
| 68 |
+
else:
|
| 69 |
+
self.patch_embedding = nn.Conv2d(
|
| 70 |
+
in_channels=config.num_channels,
|
| 71 |
+
out_channels=self.embed_dim,
|
| 72 |
+
kernel_size=self.patch_size,
|
| 73 |
+
stride=self.patch_size,
|
| 74 |
+
padding="valid",
|
| 75 |
+
)
|
| 76 |
+
if self.preserve_original_pe:
|
| 77 |
+
self.num_patches = (self.image_size // self.patch_size) ** 2
|
| 78 |
+
self.position_embedding_size = self.image_size // self.patch_size
|
| 79 |
+
self.position_embedding = nn.Embedding(self.num_patches, self.embed_dim)
|
| 80 |
+
|
| 81 |
+
@staticmethod
|
| 82 |
+
def resize_positional_embeddings(
|
| 83 |
+
positional_embeddings: torch.Tensor,
|
| 84 |
+
spatial_shapes: torch.LongTensor,
|
| 85 |
+
max_length: int,
|
| 86 |
+
) -> torch.Tensor:
|
| 87 |
+
"""
|
| 88 |
+
Resize positional embeddings to image-specific size and pad to a fixed size.
|
| 89 |
+
|
| 90 |
+
Args:
|
| 91 |
+
positional_embeddings (`torch.Tensor`):
|
| 92 |
+
Position embeddings of shape (height, width, embed_dim)
|
| 93 |
+
spatial_shapes (`torch.LongTensor`):
|
| 94 |
+
Spatial shapes of shape (batch_size, 2) to resize the positional embeddings to
|
| 95 |
+
max_length (`int`):
|
| 96 |
+
Maximum length of the positional embeddings to pad resized positional embeddings to
|
| 97 |
+
|
| 98 |
+
Returns:
|
| 99 |
+
`torch.Tensor`: Embeddings of shape (batch_size, max_length, embed_dim)
|
| 100 |
+
"""
|
| 101 |
+
batch_size = spatial_shapes.shape[0]
|
| 102 |
+
embed_dim = positional_embeddings.shape[-1]
|
| 103 |
+
source_dtype = positional_embeddings.dtype
|
| 104 |
+
|
| 105 |
+
resulted_positional_embeddings = torch.empty(
|
| 106 |
+
(batch_size, max_length, embed_dim),
|
| 107 |
+
device=positional_embeddings.device,
|
| 108 |
+
dtype=source_dtype,
|
| 109 |
+
)
|
| 110 |
+
|
| 111 |
+
# (height, width, embed_dim) -> (1, embed_dim, height, width) for interpolation
|
| 112 |
+
positional_embeddings = positional_embeddings.permute(2, 0, 1).unsqueeze(0)
|
| 113 |
+
|
| 114 |
+
# Upcast to float32 on CPU because antialias is not supported for bfloat16/float16 on CPU
|
| 115 |
+
if positional_embeddings.device.type == "cpu":
|
| 116 |
+
positional_embeddings = positional_embeddings.to(torch.float32)
|
| 117 |
+
|
| 118 |
+
for i in range(batch_size):
|
| 119 |
+
# (1, dim, height, width) -> (1, dim, target_height, target_width)
|
| 120 |
+
height, width = spatial_shapes[i]
|
| 121 |
+
resized_embeddings = F.interpolate(
|
| 122 |
+
positional_embeddings,
|
| 123 |
+
size=(height, width),
|
| 124 |
+
mode="bilinear",
|
| 125 |
+
align_corners=False,
|
| 126 |
+
antialias=True,
|
| 127 |
+
)
|
| 128 |
+
|
| 129 |
+
# (1, dim, target_height, target_width) -> (target_height * target_width, dim)
|
| 130 |
+
resized_embeddings = resized_embeddings.reshape(embed_dim, height * width).transpose(0, 1)
|
| 131 |
+
|
| 132 |
+
# Cast to original dtype
|
| 133 |
+
resized_embeddings = resized_embeddings.to(source_dtype)
|
| 134 |
+
|
| 135 |
+
resulted_positional_embeddings[i, : height * width] = resized_embeddings
|
| 136 |
+
resulted_positional_embeddings[i, height * width :] = resized_embeddings[0]
|
| 137 |
+
|
| 138 |
+
return resulted_positional_embeddings
|
| 139 |
+
|
| 140 |
+
def forward(self, pixel_values: torch.FloatTensor,
|
| 141 |
+
grid_thws: Optional[torch.LongTensor] = None) -> torch.Tensor:
|
| 142 |
+
"""
|
| 143 |
+
Args:
|
| 144 |
+
pixel_values (`torch.FloatTensor`):
|
| 145 |
+
Pixel values of shape (num_patches, num_channels * temporal_patch_size * patch_size * patch_size)
|
| 146 |
+
grid_thws: (`torch.LongTensor`):
|
| 147 |
+
grid shape (num_patches, 3)
|
| 148 |
+
"""
|
| 149 |
+
|
| 150 |
+
# Apply patch embeddings to already patchified pixel values
|
| 151 |
+
target_dtype = self.patch_embedding.weight.dtype
|
| 152 |
+
if isinstance(self.patch_embedding, nn.Linear):
|
| 153 |
+
patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))
|
| 154 |
+
elif isinstance(self.patch_embedding, nn.Conv2d):
|
| 155 |
+
pixel_values = pixel_values.view(-1, self.config.num_channels * self.config.temporal_patch_size, self.patch_size,
|
| 156 |
+
self.patch_size)
|
| 157 |
+
patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype))
|
| 158 |
+
patch_embeds = patch_embeds.reshape(-1, self.embed_dim)
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
if self.preserve_original_pe:
|
| 162 |
+
assert grid_thws is not None
|
| 163 |
+
pos_embed_new = torch.zeros_like(patch_embeds)
|
| 164 |
+
ori_h = ori_w = self.position_embedding_size
|
| 165 |
+
positional_embeddings = self.position_embedding.weight.reshape(
|
| 166 |
+
self.position_embedding_size, self.position_embedding_size, -1
|
| 167 |
+
).unsqueeze(0).permute(0,3,1,2)
|
| 168 |
+
# pos_embed = self.pos_embed.reshape(1, ori_h, ori_w, -1).permute(0, 3, 1, 2)
|
| 169 |
+
cnt = 0
|
| 170 |
+
for t, h, w in grid_thws:
|
| 171 |
+
thw = t * h * w
|
| 172 |
+
pe = F.interpolate(positional_embeddings, size=(h, w), mode='bicubic', align_corners=False)
|
| 173 |
+
pe = pe.permute(0, 2, 3, 1).reshape(1, h * w, -1)
|
| 174 |
+
pe = pe[0].repeat(t, 1)
|
| 175 |
+
pe = pe.reshape(t, h // self.hidden_stride, self.hidden_stride, w // self.hidden_stride,
|
| 176 |
+
self.hidden_stride, -1)
|
| 177 |
+
pe = pe.permute(0, 1, 3, 2, 4, 5).reshape(thw, -1)
|
| 178 |
+
pos_embed_new[cnt:cnt + thw] = pe
|
| 179 |
+
cnt += thw
|
| 180 |
+
patch_embeds = patch_embeds + pos_embed_new
|
| 181 |
+
|
| 182 |
+
return patch_embeds
|
| 183 |
+
|
| 184 |
+
|
| 185 |
+
# copied from qwen2.5-vl
|
| 186 |
+
def apply_rotary_pos_emb_flashatt(
|
| 187 |
+
q: torch.Tensor, k: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor
|
| 188 |
+
) -> Tuple[torch.Tensor, torch.Tensor]:
|
| 189 |
+
cos = cos.chunk(2, dim=-1)[0].contiguous()
|
| 190 |
+
sin = sin.chunk(2, dim=-1)[0].contiguous()
|
| 191 |
+
q_embed = apply_rotary_emb(q.float(), cos.float(), sin.float()).type_as(q)
|
| 192 |
+
k_embed = apply_rotary_emb(k.float(), cos.float(), sin.float()).type_as(k)
|
| 193 |
+
return q_embed, k_embed
|
| 194 |
+
|
| 195 |
+
|
| 196 |
+
class Siglip2Attention(nn.Module):
|
| 197 |
+
"""Multi-headed attention from 'Attention Is All You Need' paper"""
|
| 198 |
+
|
| 199 |
+
def __init__(self, config):
|
| 200 |
+
super().__init__()
|
| 201 |
+
self.config = config
|
| 202 |
+
self.embed_dim = config.hidden_size
|
| 203 |
+
self.num_heads = config.num_attention_heads
|
| 204 |
+
self.head_dim = self.embed_dim // self.num_heads
|
| 205 |
+
if self.head_dim * self.num_heads != self.embed_dim:
|
| 206 |
+
raise ValueError(
|
| 207 |
+
f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
|
| 208 |
+
f" {self.num_heads})."
|
| 209 |
+
)
|
| 210 |
+
self.scale = self.head_dim**-0.5
|
| 211 |
+
self.dropout = config.attention_dropout
|
| 212 |
+
self.is_causal = False
|
| 213 |
+
|
| 214 |
+
self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
| 215 |
+
self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
| 216 |
+
self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
| 217 |
+
self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)
|
| 218 |
+
|
| 219 |
+
self.use_rope = config.use_rope
|
| 220 |
+
|
| 221 |
+
def forward(
|
| 222 |
+
self,
|
| 223 |
+
hidden_states: torch.Tensor,
|
| 224 |
+
cu_seqlens: torch.Tensor,
|
| 225 |
+
position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
|
| 226 |
+
) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
|
| 227 |
+
"""Input shape: Batch x Time x Channel"""
|
| 228 |
+
|
| 229 |
+
seq_length, embed_dim = hidden_states.shape
|
| 230 |
+
|
| 231 |
+
queries = self.q_proj(hidden_states)
|
| 232 |
+
keys = self.k_proj(hidden_states)
|
| 233 |
+
values = self.v_proj(hidden_states)
|
| 234 |
+
|
| 235 |
+
queries = queries.view(seq_length, self.num_heads, self.head_dim)
|
| 236 |
+
keys = keys.view(seq_length, self.num_heads, self.head_dim)
|
| 237 |
+
values = values.view(seq_length, self.num_heads, self.head_dim)
|
| 238 |
+
|
| 239 |
+
if self.use_rope:
|
| 240 |
+
cos, sin = position_embeddings
|
| 241 |
+
queries, keys = apply_rotary_pos_emb_flashatt(queries.unsqueeze(0), keys.unsqueeze(0), cos, sin)
|
| 242 |
+
queries = queries.squeeze(0)
|
| 243 |
+
keys = keys.squeeze(0)
|
| 244 |
+
|
| 245 |
+
max_seqlen = (cu_seqlens[1:] - cu_seqlens[:-1]).max().item()
|
| 246 |
+
attn_output = flash_attn_varlen_func(queries, keys, values, cu_seqlens, cu_seqlens, max_seqlen, max_seqlen).reshape(
|
| 247 |
+
seq_length, -1
|
| 248 |
+
)
|
| 249 |
+
attn_output = self.out_proj(attn_output)
|
| 250 |
+
return attn_output
|
| 251 |
+
|
| 252 |
+
class Siglip2MLP(nn.Module):
|
| 253 |
+
def __init__(self, config):
|
| 254 |
+
super().__init__()
|
| 255 |
+
self.config = config
|
| 256 |
+
self.activation_fn = ACT2FN[config.hidden_act]
|
| 257 |
+
self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
|
| 258 |
+
self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
|
| 259 |
+
|
| 260 |
+
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
|
| 261 |
+
hidden_states = self.fc1(hidden_states)
|
| 262 |
+
hidden_states = self.activation_fn(hidden_states)
|
| 263 |
+
hidden_states = self.fc2(hidden_states)
|
| 264 |
+
return hidden_states
|
| 265 |
+
|
| 266 |
+
|
| 267 |
+
class Siglip2EncoderLayer(nn.Module):
|
| 268 |
+
def __init__(self, config: Siglip2NavitConfig):
|
| 269 |
+
super().__init__()
|
| 270 |
+
self.embed_dim = config.hidden_size
|
| 271 |
+
self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
|
| 272 |
+
self.self_attn = Siglip2Attention(config)
|
| 273 |
+
self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
|
| 274 |
+
self.mlp = Siglip2MLP(config)
|
| 275 |
+
|
| 276 |
+
def forward(
|
| 277 |
+
self,
|
| 278 |
+
hidden_states: torch.Tensor,
|
| 279 |
+
cu_seqlens: torch.Tensor,
|
| 280 |
+
position_embeddings: torch.Tensor
|
| 281 |
+
) -> tuple[torch.FloatTensor]:
|
| 282 |
+
"""
|
| 283 |
+
Args:
|
| 284 |
+
hidden_states (`torch.FloatTensor`):
|
| 285 |
+
Input to the layer of shape `(batch, seq_len, embed_dim)`.
|
| 286 |
+
attention_mask (`torch.FloatTensor`):
|
| 287 |
+
Attention mask of shape `(batch, 1, q_len, k_v_seq_len)` where padding elements are indicated by very large negative values.
|
| 288 |
+
output_attentions (`bool`, *optional*, defaults to `False`):
|
| 289 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
| 290 |
+
returned tensors for more detail.
|
| 291 |
+
"""
|
| 292 |
+
residual = hidden_states
|
| 293 |
+
|
| 294 |
+
hidden_states = self.layer_norm1(hidden_states)
|
| 295 |
+
hidden_states = self.self_attn(
|
| 296 |
+
hidden_states=hidden_states,
|
| 297 |
+
cu_seqlens=cu_seqlens,
|
| 298 |
+
position_embeddings=position_embeddings
|
| 299 |
+
)
|
| 300 |
+
hidden_states = residual + hidden_states
|
| 301 |
+
|
| 302 |
+
residual = hidden_states
|
| 303 |
+
hidden_states = self.layer_norm2(hidden_states)
|
| 304 |
+
hidden_states = self.mlp(hidden_states)
|
| 305 |
+
hidden_states = residual + hidden_states
|
| 306 |
+
|
| 307 |
+
return hidden_states
|
| 308 |
+
|
| 309 |
+
class Siglip2Encoder(nn.Module):
|
| 310 |
+
"""
|
| 311 |
+
Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
|
| 312 |
+
[`Siglip2EncoderLayer`].
|
| 313 |
+
|
| 314 |
+
Args:
|
| 315 |
+
config: Siglip2NavitConfig
|
| 316 |
+
"""
|
| 317 |
+
|
| 318 |
+
def __init__(self, config: Siglip2NavitConfig):
|
| 319 |
+
super().__init__()
|
| 320 |
+
self.config = config
|
| 321 |
+
self.layers = nn.ModuleList([Siglip2EncoderLayer(config) for _ in range(config.num_hidden_layers)])
|
| 322 |
+
self.gradient_checkpointing = False
|
| 323 |
+
|
| 324 |
+
self.rotary_pos_emb = VisionRotaryEmbedding(config.hidden_size // config.num_attention_heads // 2)
|
| 325 |
+
self.patch_size = config.patch_size
|
| 326 |
+
self.hidden_stride = config.hidden_stride
|
| 327 |
+
self.window_size = config.window_size
|
| 328 |
+
self.spatial_merge_unit = config.hidden_stride * config.hidden_stride
|
| 329 |
+
self.fullatt_block_indexes = None if config.fullatt_block_indexes is None else [int(i) for i in config.fullatt_block_indexes.split('|')]
|
| 330 |
+
|
| 331 |
+
|
| 332 |
+
# copied from qwen2.5_vl
|
| 333 |
+
def rot_pos_emb(self, grid_thw):
|
| 334 |
+
pos_ids = []
|
| 335 |
+
for t, h, w in grid_thw:
|
| 336 |
+
hpos_ids = torch.arange(h).unsqueeze(1).expand(-1, w)
|
| 337 |
+
hpos_ids = hpos_ids.reshape(
|
| 338 |
+
h // self.hidden_stride,
|
| 339 |
+
self.hidden_stride,
|
| 340 |
+
w // self.hidden_stride,
|
| 341 |
+
self.hidden_stride,
|
| 342 |
+
)
|
| 343 |
+
hpos_ids = hpos_ids.permute(0, 2, 1, 3)
|
| 344 |
+
hpos_ids = hpos_ids.flatten()
|
| 345 |
+
|
| 346 |
+
wpos_ids = torch.arange(w).unsqueeze(0).expand(h, -1)
|
| 347 |
+
wpos_ids = wpos_ids.reshape(
|
| 348 |
+
h // self.hidden_stride,
|
| 349 |
+
self.hidden_stride,
|
| 350 |
+
w // self.hidden_stride,
|
| 351 |
+
self.hidden_stride,
|
| 352 |
+
)
|
| 353 |
+
wpos_ids = wpos_ids.permute(0, 2, 1, 3)
|
| 354 |
+
wpos_ids = wpos_ids.flatten()
|
| 355 |
+
pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1))
|
| 356 |
+
pos_ids = torch.cat(pos_ids, dim=0)
|
| 357 |
+
max_grid_size = grid_thw[:, 1:].max()
|
| 358 |
+
rotary_pos_emb_full = self.rotary_pos_emb(max_grid_size)
|
| 359 |
+
rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
|
| 360 |
+
return rotary_pos_emb
|
| 361 |
+
|
| 362 |
+
def get_window_index(self, grid_thw):
|
| 363 |
+
window_index: list = []
|
| 364 |
+
cu_window_seqlens: list = [0]
|
| 365 |
+
window_index_id = 0
|
| 366 |
+
vit_merger_window_size = self.window_size // self.hidden_stride // self.patch_size # patch (after merge) number in each window
|
| 367 |
+
|
| 368 |
+
for grid_t, grid_h, grid_w in grid_thw:
|
| 369 |
+
llm_grid_h, llm_grid_w = (
|
| 370 |
+
grid_h // self.hidden_stride, # number of patch after merge
|
| 371 |
+
grid_w // self.hidden_stride,
|
| 372 |
+
)
|
| 373 |
+
index = torch.arange(grid_t * llm_grid_h * llm_grid_w).reshape(grid_t, llm_grid_h, llm_grid_w)
|
| 374 |
+
pad_h = vit_merger_window_size - llm_grid_h % vit_merger_window_size
|
| 375 |
+
pad_w = vit_merger_window_size - llm_grid_w % vit_merger_window_size
|
| 376 |
+
num_windows_h = (llm_grid_h + pad_h) // vit_merger_window_size
|
| 377 |
+
num_windows_w = (llm_grid_w + pad_w) // vit_merger_window_size
|
| 378 |
+
index_padded = F.pad(index, (0, pad_w, 0, pad_h), "constant", -100)
|
| 379 |
+
index_padded = index_padded.reshape(
|
| 380 |
+
grid_t,
|
| 381 |
+
num_windows_h,
|
| 382 |
+
vit_merger_window_size,
|
| 383 |
+
num_windows_w,
|
| 384 |
+
vit_merger_window_size,
|
| 385 |
+
)
|
| 386 |
+
index_padded = index_padded.permute(0, 1, 3, 2, 4).reshape(
|
| 387 |
+
grid_t,
|
| 388 |
+
num_windows_h * num_windows_w,
|
| 389 |
+
vit_merger_window_size,
|
| 390 |
+
vit_merger_window_size,
|
| 391 |
+
)
|
| 392 |
+
seqlens = (index_padded != -100).sum([2, 3]).reshape(-1)
|
| 393 |
+
index_padded = index_padded.reshape(-1)
|
| 394 |
+
index_new = index_padded[index_padded != -100]
|
| 395 |
+
window_index.append(index_new + window_index_id)
|
| 396 |
+
cu_seqlens_tmp = seqlens.cumsum(0) * self.spatial_merge_unit + cu_window_seqlens[-1]
|
| 397 |
+
cu_window_seqlens.extend(cu_seqlens_tmp.tolist())
|
| 398 |
+
window_index_id += (grid_t * llm_grid_h * llm_grid_w).item()
|
| 399 |
+
window_index = torch.cat(window_index, dim=0)
|
| 400 |
+
|
| 401 |
+
return window_index, cu_window_seqlens
|
| 402 |
+
|
| 403 |
+
# Ignore copy
|
| 404 |
+
def forward(
|
| 405 |
+
self,
|
| 406 |
+
inputs_embeds,
|
| 407 |
+
grid_thws: torch.Tensor,
|
| 408 |
+
output_hidden_states: bool = False,
|
| 409 |
+
) -> Tuple[torch.Tensor, Optional[Tuple[torch.Tensor, ...]]]:
|
| 410 |
+
r"""
|
| 411 |
+
Args:
|
| 412 |
+
inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
|
| 413 |
+
Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
|
| 414 |
+
This is useful if you want more control over how to convert `input_ids` indices into associated vectors
|
| 415 |
+
than the model's internal embedding lookup matrix.
|
| 416 |
+
attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
|
| 417 |
+
Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
|
| 418 |
+
|
| 419 |
+
- 1 for tokens that are **not masked**,
|
| 420 |
+
- 0 for tokens that are **masked**.
|
| 421 |
+
|
| 422 |
+
[What are attention masks?](../glossary#attention-mask)
|
| 423 |
+
output_attentions (`bool`, *optional*):
|
| 424 |
+
Whether or not to return the attentions tensors of all attention layers. See `attentions` under
|
| 425 |
+
returned tensors for more detail.
|
| 426 |
+
output_hidden_states (`bool`, *optional*):
|
| 427 |
+
Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
|
| 428 |
+
for more detail.
|
| 429 |
+
return_dict (`bool`, *optional*):
|
| 430 |
+
Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
|
| 431 |
+
"""
|
| 432 |
+
|
| 433 |
+
rotary_pos_emb = self.rot_pos_emb(grid_thws)
|
| 434 |
+
window_index, cu_window_seqlens = self.get_window_index(grid_thws)
|
| 435 |
+
cu_window_seqlens = torch.tensor(
|
| 436 |
+
cu_window_seqlens,
|
| 437 |
+
device=inputs_embeds.device,
|
| 438 |
+
dtype=grid_thws.dtype if torch.jit.is_tracing() else torch.int32,
|
| 439 |
+
)
|
| 440 |
+
cu_window_seqlens = torch.unique_consecutive(cu_window_seqlens)
|
| 441 |
+
|
| 442 |
+
seq_len, _ = inputs_embeds.size()
|
| 443 |
+
inputs_embeds = inputs_embeds.reshape(seq_len // self.spatial_merge_unit, self.spatial_merge_unit, -1)
|
| 444 |
+
inputs_embeds = inputs_embeds[window_index, :, :]
|
| 445 |
+
inputs_embeds = inputs_embeds.reshape(seq_len, -1)
|
| 446 |
+
rotary_pos_emb = rotary_pos_emb.reshape(seq_len // self.spatial_merge_unit, self.spatial_merge_unit, -1)
|
| 447 |
+
rotary_pos_emb = rotary_pos_emb[window_index, :, :]
|
| 448 |
+
rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1)
|
| 449 |
+
emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
|
| 450 |
+
position_embeddings = (emb.cos(), emb.sin())
|
| 451 |
+
|
| 452 |
+
cu_seqlens = torch.repeat_interleave(grid_thws[:, 1] * grid_thws[:, 2], grid_thws[:, 0]).cumsum(
|
| 453 |
+
dim=0,
|
| 454 |
+
# Select dtype based on the following factors:
|
| 455 |
+
# - FA2 requires that cu_seqlens_q must have dtype int32
|
| 456 |
+
# - torch.onnx.export requires that cu_seqlens_q must have same dtype as grid_thw
|
| 457 |
+
# See https://github.com/huggingface/transformers/pull/34852 for more information
|
| 458 |
+
dtype=grid_thws.dtype if torch.jit.is_tracing() else torch.int32,
|
| 459 |
+
)
|
| 460 |
+
cu_seqlens = F.pad(cu_seqlens, (1, 0), value=0)
|
| 461 |
+
|
| 462 |
+
reverse_indices = torch.argsort(window_index)
|
| 463 |
+
encoder_states = () if output_hidden_states else None
|
| 464 |
+
|
| 465 |
+
hidden_states = inputs_embeds
|
| 466 |
+
for index, block in enumerate(self.layers):
|
| 467 |
+
if self.fullatt_block_indexes is None or index in self.fullatt_block_indexes:
|
| 468 |
+
cu_seqlens_tmp = cu_seqlens
|
| 469 |
+
else:
|
| 470 |
+
cu_seqlens_tmp = cu_window_seqlens
|
| 471 |
+
if self.gradient_checkpointing and self.training:
|
| 472 |
+
hidden_states = self._gradient_checkpointing_func(block.__call__, hidden_states, cu_seqlens_tmp, position_embeddings)
|
| 473 |
+
else:
|
| 474 |
+
hidden_states = block(hidden_states, cu_seqlens_tmp, position_embeddings)
|
| 475 |
+
if output_hidden_states:
|
| 476 |
+
hidden_states_ = hidden_states.reshape(seq_len // self.spatial_merge_unit, self.spatial_merge_unit, -1)
|
| 477 |
+
encoder_states += (hidden_states_[reverse_indices, :].reshape(seq_len, -1),)
|
| 478 |
+
# tokens = self.post_trunk_norm(tokens)
|
| 479 |
+
hidden_states = hidden_states.reshape(seq_len // self.spatial_merge_unit, self.spatial_merge_unit, -1)
|
| 480 |
+
hidden_states = hidden_states[reverse_indices, :].reshape(seq_len, -1)
|
| 481 |
+
|
| 482 |
+
return hidden_states, encoder_states
|
| 483 |
+
|
| 484 |
+
class Siglip2VisionTransformer(nn.Module):
|
| 485 |
+
def __init__(self, config: Siglip2NavitConfig):
|
| 486 |
+
super().__init__()
|
| 487 |
+
self.config = config
|
| 488 |
+
embed_dim = config.hidden_size
|
| 489 |
+
|
| 490 |
+
self.embeddings = Siglip2VisionEmbeddings(config)
|
| 491 |
+
self.encoder = Siglip2Encoder(config)
|
| 492 |
+
self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)
|
| 493 |
+
self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
|
| 494 |
+
|
| 495 |
+
def forward(
|
| 496 |
+
self,
|
| 497 |
+
pixel_values: torch.FloatTensor,
|
| 498 |
+
grid_thws: torch.LongTensor,
|
| 499 |
+
output_hidden_states: Optional[bool] = True,
|
| 500 |
+
return_dict: Optional[bool] = True,
|
| 501 |
+
) -> Union[
|
| 502 |
+
Tuple[torch.Tensor],
|
| 503 |
+
Tuple[torch.Tensor, Tuple[torch.Tensor, ...]],
|
| 504 |
+
BaseModelOutputWithNoAttention,
|
| 505 |
+
]:
|
| 506 |
+
r"""
|
| 507 |
+
spatial_shapes (`torch.LongTensor` of shape `(batch_size, 2)`):
|
| 508 |
+
Tensor containing the spatial dimensions (height, width) of the input images.
|
| 509 |
+
"""
|
| 510 |
+
# output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
|
| 511 |
+
# output_hidden_states = (
|
| 512 |
+
# output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
| 513 |
+
# )
|
| 514 |
+
|
| 515 |
+
hidden_states = self.embeddings(pixel_values, grid_thws)
|
| 516 |
+
|
| 517 |
+
last_hidden_state, hidden_states = self.encoder(hidden_states, grid_thws, output_hidden_states)
|
| 518 |
+
last_hidden_state = self.post_layernorm(last_hidden_state)
|
| 519 |
+
|
| 520 |
+
if not return_dict:
|
| 521 |
+
output = (last_hidden_state,)
|
| 522 |
+
output += (hidden_states,) if output_hidden_states else ()
|
| 523 |
+
return output
|
| 524 |
+
|
| 525 |
+
return BaseModelOutputWithNoAttention(
|
| 526 |
+
last_hidden_state=last_hidden_state,
|
| 527 |
+
hidden_states=hidden_states
|
| 528 |
+
)
|
| 529 |
+
|
| 530 |
+
class Siglip2PreTrainedModel(PreTrainedModel):
|
| 531 |
+
config_class = Siglip2NavitConfig
|
| 532 |
+
base_model_prefix = "siglip2_navit"
|
| 533 |
+
supports_gradient_checkpointing = True
|
| 534 |
+
|
| 535 |
+
_no_split_modules = [
|
| 536 |
+
"Siglip2VisionEmbeddings",
|
| 537 |
+
"Siglip2EncoderLayer",
|
| 538 |
+
]
|
| 539 |
+
_supports_flash_attn_2 = True
|
| 540 |
+
_supports_sdpa = False
|
| 541 |
+
_supports_flex_attn = False
|
| 542 |
+
_supports_attention_backend = True
|
| 543 |
+
|
| 544 |
+
|
| 545 |
+
class Siglip2NavitModel(Siglip2PreTrainedModel):
|
| 546 |
+
config_class = Siglip2NavitConfig
|
| 547 |
+
main_input_name = "pixel_values"
|
| 548 |
+
|
| 549 |
+
def __init__(self, config: Siglip2NavitConfig):
|
| 550 |
+
super().__init__(config)
|
| 551 |
+
|
| 552 |
+
self.vision_model = Siglip2VisionTransformer(config)
|
| 553 |
+
|
| 554 |
+
def get_input_embeddings(self) -> nn.Module:
|
| 555 |
+
return self.vision_model.embeddings.patch_embedding
|
| 556 |
+
|
| 557 |
+
def forward(
|
| 558 |
+
self,
|
| 559 |
+
pixel_values: torch.FloatTensor,
|
| 560 |
+
grid_thws: torch.LongTensor,
|
| 561 |
+
output_hidden_states: Optional[bool] = None,
|
| 562 |
+
return_dict: Optional[bool] = None,
|
| 563 |
+
) -> Union[
|
| 564 |
+
Tuple[torch.Tensor],
|
| 565 |
+
Tuple[torch.Tensor, Tuple[torch.Tensor, ...]],
|
| 566 |
+
BaseModelOutputWithNoAttention,
|
| 567 |
+
]:
|
| 568 |
+
|
| 569 |
+
if output_hidden_states is None:
|
| 570 |
+
output_hidden_states = self.config.output_hidden_states
|
| 571 |
+
if return_dict is None:
|
| 572 |
+
return_dict = self.config.use_return_dict
|
| 573 |
+
|
| 574 |
+
return self.vision_model(
|
| 575 |
+
pixel_values=pixel_values,
|
| 576 |
+
grid_thws=grid_thws,
|
| 577 |
+
output_hidden_states=output_hidden_states,
|
| 578 |
+
return_dict=return_dict,
|
| 579 |
+
)
|
| 580 |
+
|
| 581 |
+
class VisualEmbedding(torch.nn.Embedding):
|
| 582 |
+
"""
|
| 583 |
+
A visual embedding layer that can handle both discrete token IDs (long) and continuous
|
| 584 |
+
soft-token probabilities (float).
|
| 585 |
+
"""
|
| 586 |
+
|
| 587 |
+
def forward(self, visual_tokens: Tensor) -> Tensor:
|
| 588 |
+
if visual_tokens.dtype in [torch.int8, torch.int16, torch.int32, torch.int64, torch.long]:
|
| 589 |
+
return super().forward(visual_tokens)
|
| 590 |
+
# Handle soft tokens (probabilities) by matrix multiplication with the embedding weight
|
| 591 |
+
return torch.matmul(visual_tokens, self.weight)
|
| 592 |
+
|
| 593 |
+
|
| 594 |
+
class VisualTokenizer(torch.nn.Module):
|
| 595 |
+
"""
|
| 596 |
+
Tokenizes images or videos into a sequence of continuous visual tokens.
|
| 597 |
+
"""
|
| 598 |
+
|
| 599 |
+
def __init__(self, vit, visual_vocab_size, image_processor_name_or_path, *args, **kwargs):
|
| 600 |
+
super().__init__(*args, **kwargs)
|
| 601 |
+
self.vit = vit
|
| 602 |
+
self.image_processor = AutoImageProcessor.from_pretrained(image_processor_name_or_path, do_center_crop=False)
|
| 603 |
+
head_dim = visual_vocab_size - len(INDICATOR_IDS)
|
| 604 |
+
self.head = torch.nn.Sequential(
|
| 605 |
+
torch.nn.Linear(self.vit.config.hidden_size * self.vit.config.hidden_stride ** 2, head_dim, bias=False),
|
| 606 |
+
torch.nn.LayerNorm(head_dim)
|
| 607 |
+
)
|
| 608 |
+
|
| 609 |
+
def _encode(self, pixel_values, grid_thws):
|
| 610 |
+
output = self.vit(pixel_values, grid_thws, output_hidden_states=True, return_dict=True)
|
| 611 |
+
features = output.hidden_states[-1]
|
| 612 |
+
seq_len, _ = features.shape
|
| 613 |
+
features = features.reshape(seq_len // (self.vit.config.hidden_stride ** 2), -1)
|
| 614 |
+
return features
|
| 615 |
+
|
| 616 |
+
# Adapted from qwen2_vl
|
| 617 |
+
@staticmethod
|
| 618 |
+
def smart_resize(
|
| 619 |
+
height: int, width: int, factor: int = 28, min_pixels: int = 448 * 448, max_pixels: int = 1344 * 1792
|
| 620 |
+
):
|
| 621 |
+
"""Rescales the image so that the following conditions are met:
|
| 622 |
+
1. Both dimensions are divisible by 'factor'.
|
| 623 |
+
2. The total number of pixels is within ['min_pixels', 'max_pixels'].
|
| 624 |
+
3. The aspect ratio is maintained as closely as possible.
|
| 625 |
+
"""
|
| 626 |
+
if height < factor or width < factor:
|
| 627 |
+
if height < width:
|
| 628 |
+
width = round(factor / height * width)
|
| 629 |
+
height = factor
|
| 630 |
+
else:
|
| 631 |
+
height = round(factor / width * height)
|
| 632 |
+
width = factor
|
| 633 |
+
|
| 634 |
+
elif max(height, width) / min(height, width) > 200:
|
| 635 |
+
if height > width:
|
| 636 |
+
height = 200 * width
|
| 637 |
+
else:
|
| 638 |
+
width = 200 * height
|
| 639 |
+
|
| 640 |
+
h_bar = round(height / factor) * factor
|
| 641 |
+
w_bar = round(width / factor) * factor
|
| 642 |
+
if h_bar * w_bar > max_pixels:
|
| 643 |
+
beta = math.sqrt((height * width) / max_pixels)
|
| 644 |
+
h_bar = math.floor(height / beta / factor) * factor
|
| 645 |
+
w_bar = math.floor(width / beta / factor) * factor
|
| 646 |
+
elif h_bar * w_bar < min_pixels:
|
| 647 |
+
beta = math.sqrt(min_pixels / (height * width))
|
| 648 |
+
h_bar = math.ceil(height * beta / factor) * factor
|
| 649 |
+
w_bar = math.ceil(width * beta / factor) * factor
|
| 650 |
+
return h_bar, w_bar
|
| 651 |
+
|
| 652 |
+
def preprocess(
|
| 653 |
+
self,
|
| 654 |
+
image: Optional[PIL.Image.Image] = None,
|
| 655 |
+
video: Optional[List[PIL.Image.Image]] = None,
|
| 656 |
+
min_pixels: Optional[int] = None,
|
| 657 |
+
max_pixels: Optional[int] = None
|
| 658 |
+
):
|
| 659 |
+
patch_size = self.vit.config.patch_size
|
| 660 |
+
temporal_patch_size = self.vit.config.temporal_patch_size
|
| 661 |
+
hidden_stride = self.vit.config.hidden_stride
|
| 662 |
+
assert (image is None) ^ (video is None), "Invalid input: expect either image or video"
|
| 663 |
+
if image is not None:
|
| 664 |
+
images = [image]
|
| 665 |
+
else:
|
| 666 |
+
images = video
|
| 667 |
+
images = [image.convert("RGB") if image.mode != 'RGB' else image for image in images]
|
| 668 |
+
width, height = images[0].size
|
| 669 |
+
processed_images = []
|
| 670 |
+
for image in images:
|
| 671 |
+
resized_height, resized_width = self.smart_resize(
|
| 672 |
+
height,
|
| 673 |
+
width,
|
| 674 |
+
factor=patch_size * hidden_stride,
|
| 675 |
+
min_pixels=min_pixels,
|
| 676 |
+
max_pixels=max_pixels,
|
| 677 |
+
)
|
| 678 |
+
new_size = dict(height=resized_height, width=resized_width)
|
| 679 |
+
new_image = self.image_processor.preprocess(image, size=new_size, return_tensors="np")['pixel_values'][0]
|
| 680 |
+
processed_images.append(new_image)
|
| 681 |
+
|
| 682 |
+
patches = np.array(processed_images)
|
| 683 |
+
if patches.shape[0] % temporal_patch_size != 0:
|
| 684 |
+
repeats = np.repeat(patches[-1][np.newaxis], temporal_patch_size - 1, axis=0)
|
| 685 |
+
patches = np.concatenate([patches, repeats], axis=0)
|
| 686 |
+
channel = patches.shape[1]
|
| 687 |
+
grid_t = patches.shape[0] // temporal_patch_size
|
| 688 |
+
grid_h, grid_w = resized_height // patch_size, resized_width // patch_size
|
| 689 |
+
grid_thw = torch.tensor([[grid_t, grid_h, grid_w]])
|
| 690 |
+
|
| 691 |
+
patches = patches.reshape(
|
| 692 |
+
grid_t, temporal_patch_size, channel,
|
| 693 |
+
grid_h // hidden_stride, hidden_stride, patch_size,
|
| 694 |
+
grid_w // hidden_stride, hidden_stride, patch_size,
|
| 695 |
+
)
|
| 696 |
+
patches = patches.transpose(0, 3, 6, 4, 7, 2, 1, 5, 8)
|
| 697 |
+
flatten_patches = patches.reshape(
|
| 698 |
+
grid_t * grid_h * grid_w, channel * temporal_patch_size * patch_size * patch_size
|
| 699 |
+
)
|
| 700 |
+
flatten_patches = torch.tensor(flatten_patches)
|
| 701 |
+
|
| 702 |
+
return flatten_patches, grid_thw
|
| 703 |
+
|
| 704 |
+
def forward(
|
| 705 |
+
self, pixel_values, grid_thws
|
| 706 |
+
) -> torch.Tensor: # [BatchSize, ImageShape] -> [BatchSize, #Token, VocabSize]
|
| 707 |
+
features = self._encode(pixel_values, grid_thws)
|
| 708 |
+
logits = self.head(features)
|
| 709 |
+
tokens = torch.softmax(logits, dim=-1, dtype=torch.float32).to(logits.dtype)
|
| 710 |
+
|
| 711 |
+
token_len, _ = tokens.shape
|
| 712 |
+
padding_tensor = torch.zeros(size=(token_len, len(INDICATOR_IDS)),
|
| 713 |
+
dtype=tokens.dtype,
|
| 714 |
+
device=tokens.device,
|
| 715 |
+
layout=tokens.layout,
|
| 716 |
+
requires_grad=False)
|
| 717 |
+
tokens = torch.cat((tokens, padding_tensor), dim=1)
|
| 718 |
+
return tokens
|
| 719 |
+
|
| 720 |
+
|
| 721 |
+
class OvisPreTrainedModel(PreTrainedModel):
|
| 722 |
+
config_class = Ovis2_5_Config
|
| 723 |
+
base_model_prefix = "ovis2_5"
|
| 724 |
+
|
| 725 |
+
|
| 726 |
+
class Ovis2_5(OvisPreTrainedModel):
|
| 727 |
+
_supports_flash_attn_2 = True
|
| 728 |
+
|
| 729 |
+
def __init__(self, config: Ovis2_5_Config, *inputs, **kwargs):
|
| 730 |
+
super().__init__(config, *inputs, **kwargs)
|
| 731 |
+
|
| 732 |
+
self.llm = AutoModelForCausalLM.from_config(self.config.llm_config)
|
| 733 |
+
assert self.config.hidden_size == self.llm.config.hidden_size, "hidden size mismatch"
|
| 734 |
+
self.text_tokenizer = AutoTokenizer.from_pretrained(self.config.name_or_path)
|
| 735 |
+
self.visual_tokenizer = VisualTokenizer(vit=AutoModel.from_config(self.config.vit_config),
|
| 736 |
+
visual_vocab_size=self.config.visual_vocab_size,
|
| 737 |
+
image_processor_name_or_path=self.config.name_or_path)
|
| 738 |
+
|
| 739 |
+
self.vte = VisualEmbedding(self.config.visual_vocab_size, self.config.hidden_size,
|
| 740 |
+
device=self.visual_tokenizer.vit.device, dtype=self.visual_tokenizer.vit.dtype)
|
| 741 |
+
indicator_token_indices = torch.arange(
|
| 742 |
+
self.config.visual_vocab_size - len(INDICATOR_IDS),
|
| 743 |
+
self.config.visual_vocab_size,
|
| 744 |
+
dtype=torch.long
|
| 745 |
+
)
|
| 746 |
+
self.register_buffer("indicator_token_indices", indicator_token_indices, persistent=False)
|
| 747 |
+
|
| 748 |
+
def _merge_modules(modules_list: tuple):
|
| 749 |
+
merged_modules = []
|
| 750 |
+
for modules in modules_list:
|
| 751 |
+
merged_modules.extend(modules if modules else [])
|
| 752 |
+
return merged_modules
|
| 753 |
+
|
| 754 |
+
# Standard model configurations for parallelism and device placement
|
| 755 |
+
self._no_split_modules = _merge_modules(
|
| 756 |
+
(self.llm._no_split_modules, self.visual_tokenizer.vit._no_split_modules))
|
| 757 |
+
self._skip_keys_device_placement = self.llm._skip_keys_device_placement
|
| 758 |
+
self._keep_in_fp32_modules = _merge_modules(
|
| 759 |
+
(self.llm._keep_in_fp32_modules, self.visual_tokenizer.vit._keep_in_fp32_modules))
|
| 760 |
+
self.is_parallelizable = all((self.llm.is_parallelizable, self.visual_tokenizer.vit.is_parallelizable))
|
| 761 |
+
self.supports_gradient_checkpointing = True
|
| 762 |
+
|
| 763 |
+
def tie_weights(self):
|
| 764 |
+
self.llm.tie_weights()
|
| 765 |
+
|
| 766 |
+
def get_wte(self):
|
| 767 |
+
return self.llm.get_input_embeddings()
|
| 768 |
+
|
| 769 |
+
def forward(
|
| 770 |
+
self,
|
| 771 |
+
input_ids: torch.Tensor,
|
| 772 |
+
attention_mask: torch.Tensor,
|
| 773 |
+
pixel_values: Optional[torch.Tensor],
|
| 774 |
+
grid_thws: Optional[torch.Tensor],
|
| 775 |
+
labels: Optional[torch.Tensor] = None,
|
| 776 |
+
**kwargs
|
| 777 |
+
):
|
| 778 |
+
inputs_embeds = self.merge_multimodal(
|
| 779 |
+
input_ids=input_ids,
|
| 780 |
+
pixel_values=pixel_values,
|
| 781 |
+
grid_thws=grid_thws,
|
| 782 |
+
)
|
| 783 |
+
return self.llm(inputs_embeds=inputs_embeds, attention_mask=attention_mask, labels=labels, **kwargs)
|
| 784 |
+
|
| 785 |
+
def merge_multimodal(
|
| 786 |
+
self,
|
| 787 |
+
input_ids: torch.Tensor,
|
| 788 |
+
pixel_values: Optional[torch.Tensor],
|
| 789 |
+
grid_thws: Optional[torch.Tensor],
|
| 790 |
+
):
|
| 791 |
+
placeholder_token_mask = torch.lt(input_ids, 0)
|
| 792 |
+
multimodal_embeds = self.get_wte()(torch.masked_fill(input_ids, placeholder_token_mask, 0))
|
| 793 |
+
|
| 794 |
+
if pixel_values is not None:
|
| 795 |
+
visual_indicator_embeds = self.vte(self.indicator_token_indices).to(
|
| 796 |
+
dtype=multimodal_embeds.dtype, device=multimodal_embeds.device
|
| 797 |
+
)
|
| 798 |
+
visual_tokens = self.visual_tokenizer(pixel_values, grid_thws)
|
| 799 |
+
visual_embeds = self.vte(visual_tokens).to(dtype=multimodal_embeds.dtype, device=multimodal_embeds.device)
|
| 800 |
+
|
| 801 |
+
for i, indicator_id in enumerate(INDICATOR_IDS):
|
| 802 |
+
multimodal_embeds[input_ids == indicator_id] = visual_indicator_embeds[i]
|
| 803 |
+
multimodal_embeds[input_ids == VISUAL_ATOM_ID] = visual_embeds
|
| 804 |
+
|
| 805 |
+
return multimodal_embeds
|
| 806 |
+
|
| 807 |
+
def _merge_inputs(
|
| 808 |
+
self, raw_input_ids, placeholder_id, grid_thws, indicator_begin_id, indicator_end_id
|
| 809 |
+
):
|
| 810 |
+
input_ids = []
|
| 811 |
+
prev_index = 0
|
| 812 |
+
placeholder_indexes = [i for i, v in enumerate(raw_input_ids) if v == placeholder_id]
|
| 813 |
+
for placeholder_index, grid_thw in zip(placeholder_indexes, grid_thws):
|
| 814 |
+
input_ids.extend(raw_input_ids[prev_index:placeholder_index])
|
| 815 |
+
num_image_atoms = grid_thw.prod().item()
|
| 816 |
+
num_image_atoms //= self.visual_tokenizer.vit.config.hidden_stride ** 2
|
| 817 |
+
num_image_atoms //= self.visual_tokenizer.vit.config.temporal_patch_size
|
| 818 |
+
input_ids.extend([indicator_begin_id] + [VISUAL_ATOM_ID] * num_image_atoms + [indicator_end_id])
|
| 819 |
+
prev_index = placeholder_index + 1
|
| 820 |
+
input_ids.extend(raw_input_ids[prev_index:])
|
| 821 |
+
return input_ids
|
| 822 |
+
|
| 823 |
+
def _tokenize_with_visual_placeholder(self, text):
|
| 824 |
+
placeholder = VIDEO_PLACEHOLDER if VIDEO_PLACEHOLDER in text else IMAGE_PLACEHOLDER
|
| 825 |
+
placeholder_id = VIDEO_PLACEHOLDER_ID if VIDEO_PLACEHOLDER in text else IMAGE_PLACEHOLDER_ID
|
| 826 |
+
chunks = [self.text_tokenizer(chunk, add_special_tokens=False).input_ids for chunk in text.split(placeholder)]
|
| 827 |
+
input_ids = chunks[0]
|
| 828 |
+
for chunk in chunks[1:]:
|
| 829 |
+
input_ids.append(placeholder_id)
|
| 830 |
+
input_ids.extend(chunk)
|
| 831 |
+
return input_ids
|
| 832 |
+
|
| 833 |
+
def preprocess_inputs(
|
| 834 |
+
self,
|
| 835 |
+
messages: List[Union[str, Dict]],
|
| 836 |
+
min_pixels=448 * 448,
|
| 837 |
+
max_pixels=1344 * 1792,
|
| 838 |
+
add_generation_prompt=True,
|
| 839 |
+
enable_thinking=False
|
| 840 |
+
):
|
| 841 |
+
text = self.text_tokenizer.apply_chat_template(
|
| 842 |
+
messages,
|
| 843 |
+
tokenize=False,
|
| 844 |
+
add_generation_prompt=add_generation_prompt,
|
| 845 |
+
enable_thinking=enable_thinking
|
| 846 |
+
)
|
| 847 |
+
input_ids = self._tokenize_with_visual_placeholder(text)
|
| 848 |
+
images = []
|
| 849 |
+
videos = []
|
| 850 |
+
for message in messages:
|
| 851 |
+
content = message["content"]
|
| 852 |
+
if isinstance(content, list):
|
| 853 |
+
images.extend([item["image"] for item in content if item.get("image") is not None])
|
| 854 |
+
videos.extend([item["video"] for item in content if item.get("video") is not None])
|
| 855 |
+
if images and videos:
|
| 856 |
+
raise ValueError(
|
| 857 |
+
"Multiple visual input data types detected (both image and video provided). "
|
| 858 |
+
"This model supports only one type of visual input data at a time. "
|
| 859 |
+
"Please provide either image or video, but not both."
|
| 860 |
+
)
|
| 861 |
+
|
| 862 |
+
pixel_values, grid_thws = None, None
|
| 863 |
+
if images:
|
| 864 |
+
pixel_values, grid_thws = zip(
|
| 865 |
+
*(self.visual_tokenizer.preprocess(image=image, min_pixels=min_pixels, max_pixels=max_pixels)
|
| 866 |
+
for image in images)
|
| 867 |
+
)
|
| 868 |
+
input_ids = self._merge_inputs(
|
| 869 |
+
input_ids, IMAGE_PLACEHOLDER_ID, grid_thws, INDICATOR_IDS[0], INDICATOR_IDS[1]
|
| 870 |
+
)
|
| 871 |
+
pixel_values = torch.cat(pixel_values, dim=0)
|
| 872 |
+
grid_thws = torch.cat(grid_thws, dim=0)
|
| 873 |
+
elif videos:
|
| 874 |
+
assert len(videos) == 1, "only support single video"
|
| 875 |
+
pixel_values, grid_thws = self.visual_tokenizer.preprocess(
|
| 876 |
+
video=videos[0], min_pixels=min_pixels, max_pixels=max_pixels
|
| 877 |
+
)
|
| 878 |
+
input_ids = self._merge_inputs(
|
| 879 |
+
input_ids, VIDEO_PLACEHOLDER_ID, grid_thws, INDICATOR_IDS[2], INDICATOR_IDS[3]
|
| 880 |
+
)
|
| 881 |
+
|
| 882 |
+
input_ids = torch.tensor(input_ids, dtype=torch.long).unsqueeze(0)
|
| 883 |
+
|
| 884 |
+
return input_ids, pixel_values, grid_thws
|
| 885 |
+
|
| 886 |
+
def generate(
|
| 887 |
+
self,
|
| 888 |
+
inputs: Optional[torch.Tensor] = None,
|
| 889 |
+
**kwargs,
|
| 890 |
+
) -> Union[GenerateOutput, torch.LongTensor]:
|
| 891 |
+
attention_mask = torch.ne(inputs, self.text_tokenizer.pad_token_id).to(device=inputs.device)
|
| 892 |
+
inputs_embeds = self.merge_multimodal(
|
| 893 |
+
input_ids=inputs,
|
| 894 |
+
pixel_values=kwargs.pop('pixel_values', None),
|
| 895 |
+
grid_thws=kwargs.pop('grid_thws', None)
|
| 896 |
+
)
|
| 897 |
+
return self.llm.generate(inputs=None, inputs_embeds=inputs_embeds, attention_mask=attention_mask, **kwargs)
|
| 898 |
+
|
| 899 |
+
|
| 900 |
+
AutoConfig.register('siglip2_navit', Siglip2NavitConfig)
|
| 901 |
+
AutoModel.register(Siglip2NavitConfig, Siglip2NavitModel)
|
| 902 |
+
AutoConfig.register("ovis2_5", Ovis2_5_Config)
|
| 903 |
+
AutoModelForCausalLM.register(Ovis2_5_Config, Ovis2_5)
|
preprocessor_config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"do_convert_rgb": null,
|
| 3 |
+
"do_normalize": true,
|
| 4 |
+
"do_rescale": true,
|
| 5 |
+
"do_resize": true,
|
| 6 |
+
"image_mean": [
|
| 7 |
+
0.5,
|
| 8 |
+
0.5,
|
| 9 |
+
0.5
|
| 10 |
+
],
|
| 11 |
+
"image_processor_type": "SiglipImageProcessor",
|
| 12 |
+
"image_std": [
|
| 13 |
+
0.5,
|
| 14 |
+
0.5,
|
| 15 |
+
0.5
|
| 16 |
+
],
|
| 17 |
+
"processor_class": "SiglipProcessor",
|
| 18 |
+
"resample": 2,
|
| 19 |
+
"rescale_factor": 0.00392156862745098,
|
| 20 |
+
"size": {
|
| 21 |
+
"height": 512,
|
| 22 |
+
"width": 512
|
| 23 |
+
}
|
| 24 |
+
}
|
special_tokens_map.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"additional_special_tokens": [
|
| 3 |
+
"<|im_start|>",
|
| 4 |
+
"<|im_end|>",
|
| 5 |
+
"<|object_ref_start|>",
|
| 6 |
+
"<|object_ref_end|>",
|
| 7 |
+
"<|box_start|>",
|
| 8 |
+
"<|box_end|>",
|
| 9 |
+
"<|quad_start|>",
|
| 10 |
+
"<|quad_end|>",
|
| 11 |
+
"<|vision_start|>",
|
| 12 |
+
"<|vision_end|>",
|
| 13 |
+
"<|vision_pad|>",
|
| 14 |
+
"<|image_pad|>",
|
| 15 |
+
"<|video_pad|>"
|
| 16 |
+
],
|
| 17 |
+
"eos_token": {
|
| 18 |
+
"content": "<|im_end|>",
|
| 19 |
+
"lstrip": false,
|
| 20 |
+
"normalized": false,
|
| 21 |
+
"rstrip": false,
|
| 22 |
+
"single_word": false
|
| 23 |
+
},
|
| 24 |
+
"pad_token": {
|
| 25 |
+
"content": "<|endoftext|>",
|
| 26 |
+
"lstrip": false,
|
| 27 |
+
"normalized": false,
|
| 28 |
+
"rstrip": false,
|
| 29 |
+
"single_word": false
|
| 30 |
+
}
|
| 31 |
+
}
|
tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
|
| 3 |
+
size 11422654
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,240 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_bos_token": false,
|
| 3 |
+
"add_prefix_space": false,
|
| 4 |
+
"added_tokens_decoder": {
|
| 5 |
+
"151643": {
|
| 6 |
+
"content": "<|endoftext|>",
|
| 7 |
+
"lstrip": false,
|
| 8 |
+
"normalized": false,
|
| 9 |
+
"rstrip": false,
|
| 10 |
+
"single_word": false,
|
| 11 |
+
"special": true
|
| 12 |
+
},
|
| 13 |
+
"151644": {
|
| 14 |
+
"content": "<|im_start|>",
|
| 15 |
+
"lstrip": false,
|
| 16 |
+
"normalized": false,
|
| 17 |
+
"rstrip": false,
|
| 18 |
+
"single_word": false,
|
| 19 |
+
"special": true
|
| 20 |
+
},
|
| 21 |
+
"151645": {
|
| 22 |
+
"content": "<|im_end|>",
|
| 23 |
+
"lstrip": false,
|
| 24 |
+
"normalized": false,
|
| 25 |
+
"rstrip": false,
|
| 26 |
+
"single_word": false,
|
| 27 |
+
"special": true
|
| 28 |
+
},
|
| 29 |
+
"151646": {
|
| 30 |
+
"content": "<|object_ref_start|>",
|
| 31 |
+
"lstrip": false,
|
| 32 |
+
"normalized": false,
|
| 33 |
+
"rstrip": false,
|
| 34 |
+
"single_word": false,
|
| 35 |
+
"special": true
|
| 36 |
+
},
|
| 37 |
+
"151647": {
|
| 38 |
+
"content": "<|object_ref_end|>",
|
| 39 |
+
"lstrip": false,
|
| 40 |
+
"normalized": false,
|
| 41 |
+
"rstrip": false,
|
| 42 |
+
"single_word": false,
|
| 43 |
+
"special": true
|
| 44 |
+
},
|
| 45 |
+
"151648": {
|
| 46 |
+
"content": "<|box_start|>",
|
| 47 |
+
"lstrip": false,
|
| 48 |
+
"normalized": false,
|
| 49 |
+
"rstrip": false,
|
| 50 |
+
"single_word": false,
|
| 51 |
+
"special": true
|
| 52 |
+
},
|
| 53 |
+
"151649": {
|
| 54 |
+
"content": "<|box_end|>",
|
| 55 |
+
"lstrip": false,
|
| 56 |
+
"normalized": false,
|
| 57 |
+
"rstrip": false,
|
| 58 |
+
"single_word": false,
|
| 59 |
+
"special": true
|
| 60 |
+
},
|
| 61 |
+
"151650": {
|
| 62 |
+
"content": "<|quad_start|>",
|
| 63 |
+
"lstrip": false,
|
| 64 |
+
"normalized": false,
|
| 65 |
+
"rstrip": false,
|
| 66 |
+
"single_word": false,
|
| 67 |
+
"special": true
|
| 68 |
+
},
|
| 69 |
+
"151651": {
|
| 70 |
+
"content": "<|quad_end|>",
|
| 71 |
+
"lstrip": false,
|
| 72 |
+
"normalized": false,
|
| 73 |
+
"rstrip": false,
|
| 74 |
+
"single_word": false,
|
| 75 |
+
"special": true
|
| 76 |
+
},
|
| 77 |
+
"151652": {
|
| 78 |
+
"content": "<|vision_start|>",
|
| 79 |
+
"lstrip": false,
|
| 80 |
+
"normalized": false,
|
| 81 |
+
"rstrip": false,
|
| 82 |
+
"single_word": false,
|
| 83 |
+
"special": true
|
| 84 |
+
},
|
| 85 |
+
"151653": {
|
| 86 |
+
"content": "<|vision_end|>",
|
| 87 |
+
"lstrip": false,
|
| 88 |
+
"normalized": false,
|
| 89 |
+
"rstrip": false,
|
| 90 |
+
"single_word": false,
|
| 91 |
+
"special": true
|
| 92 |
+
},
|
| 93 |
+
"151654": {
|
| 94 |
+
"content": "<|vision_pad|>",
|
| 95 |
+
"lstrip": false,
|
| 96 |
+
"normalized": false,
|
| 97 |
+
"rstrip": false,
|
| 98 |
+
"single_word": false,
|
| 99 |
+
"special": true
|
| 100 |
+
},
|
| 101 |
+
"151655": {
|
| 102 |
+
"content": "<|image_pad|>",
|
| 103 |
+
"lstrip": false,
|
| 104 |
+
"normalized": false,
|
| 105 |
+
"rstrip": false,
|
| 106 |
+
"single_word": false,
|
| 107 |
+
"special": true
|
| 108 |
+
},
|
| 109 |
+
"151656": {
|
| 110 |
+
"content": "<|video_pad|>",
|
| 111 |
+
"lstrip": false,
|
| 112 |
+
"normalized": false,
|
| 113 |
+
"rstrip": false,
|
| 114 |
+
"single_word": false,
|
| 115 |
+
"special": true
|
| 116 |
+
},
|
| 117 |
+
"151657": {
|
| 118 |
+
"content": "<tool_call>",
|
| 119 |
+
"lstrip": false,
|
| 120 |
+
"normalized": false,
|
| 121 |
+
"rstrip": false,
|
| 122 |
+
"single_word": false,
|
| 123 |
+
"special": false
|
| 124 |
+
},
|
| 125 |
+
"151658": {
|
| 126 |
+
"content": "</tool_call>",
|
| 127 |
+
"lstrip": false,
|
| 128 |
+
"normalized": false,
|
| 129 |
+
"rstrip": false,
|
| 130 |
+
"single_word": false,
|
| 131 |
+
"special": false
|
| 132 |
+
},
|
| 133 |
+
"151659": {
|
| 134 |
+
"content": "<|fim_prefix|>",
|
| 135 |
+
"lstrip": false,
|
| 136 |
+
"normalized": false,
|
| 137 |
+
"rstrip": false,
|
| 138 |
+
"single_word": false,
|
| 139 |
+
"special": false
|
| 140 |
+
},
|
| 141 |
+
"151660": {
|
| 142 |
+
"content": "<|fim_middle|>",
|
| 143 |
+
"lstrip": false,
|
| 144 |
+
"normalized": false,
|
| 145 |
+
"rstrip": false,
|
| 146 |
+
"single_word": false,
|
| 147 |
+
"special": false
|
| 148 |
+
},
|
| 149 |
+
"151661": {
|
| 150 |
+
"content": "<|fim_suffix|>",
|
| 151 |
+
"lstrip": false,
|
| 152 |
+
"normalized": false,
|
| 153 |
+
"rstrip": false,
|
| 154 |
+
"single_word": false,
|
| 155 |
+
"special": false
|
| 156 |
+
},
|
| 157 |
+
"151662": {
|
| 158 |
+
"content": "<|fim_pad|>",
|
| 159 |
+
"lstrip": false,
|
| 160 |
+
"normalized": false,
|
| 161 |
+
"rstrip": false,
|
| 162 |
+
"single_word": false,
|
| 163 |
+
"special": false
|
| 164 |
+
},
|
| 165 |
+
"151663": {
|
| 166 |
+
"content": "<|repo_name|>",
|
| 167 |
+
"lstrip": false,
|
| 168 |
+
"normalized": false,
|
| 169 |
+
"rstrip": false,
|
| 170 |
+
"single_word": false,
|
| 171 |
+
"special": false
|
| 172 |
+
},
|
| 173 |
+
"151664": {
|
| 174 |
+
"content": "<|file_sep|>",
|
| 175 |
+
"lstrip": false,
|
| 176 |
+
"normalized": false,
|
| 177 |
+
"rstrip": false,
|
| 178 |
+
"single_word": false,
|
| 179 |
+
"special": false
|
| 180 |
+
},
|
| 181 |
+
"151665": {
|
| 182 |
+
"content": "<tool_response>",
|
| 183 |
+
"lstrip": false,
|
| 184 |
+
"normalized": false,
|
| 185 |
+
"rstrip": false,
|
| 186 |
+
"single_word": false,
|
| 187 |
+
"special": false
|
| 188 |
+
},
|
| 189 |
+
"151666": {
|
| 190 |
+
"content": "</tool_response>",
|
| 191 |
+
"lstrip": false,
|
| 192 |
+
"normalized": false,
|
| 193 |
+
"rstrip": false,
|
| 194 |
+
"single_word": false,
|
| 195 |
+
"special": false
|
| 196 |
+
},
|
| 197 |
+
"151667": {
|
| 198 |
+
"content": "<think>",
|
| 199 |
+
"lstrip": false,
|
| 200 |
+
"normalized": false,
|
| 201 |
+
"rstrip": false,
|
| 202 |
+
"single_word": false,
|
| 203 |
+
"special": false
|
| 204 |
+
},
|
| 205 |
+
"151668": {
|
| 206 |
+
"content": "</think>",
|
| 207 |
+
"lstrip": false,
|
| 208 |
+
"normalized": false,
|
| 209 |
+
"rstrip": false,
|
| 210 |
+
"single_word": false,
|
| 211 |
+
"special": false
|
| 212 |
+
}
|
| 213 |
+
},
|
| 214 |
+
"additional_special_tokens": [
|
| 215 |
+
"<|im_start|>",
|
| 216 |
+
"<|im_end|>",
|
| 217 |
+
"<|object_ref_start|>",
|
| 218 |
+
"<|object_ref_end|>",
|
| 219 |
+
"<|box_start|>",
|
| 220 |
+
"<|box_end|>",
|
| 221 |
+
"<|quad_start|>",
|
| 222 |
+
"<|quad_end|>",
|
| 223 |
+
"<|vision_start|>",
|
| 224 |
+
"<|vision_end|>",
|
| 225 |
+
"<|vision_pad|>",
|
| 226 |
+
"<|image_pad|>",
|
| 227 |
+
"<|video_pad|>"
|
| 228 |
+
],
|
| 229 |
+
"bos_token": null,
|
| 230 |
+
"chat_template": "{%- for message in messages %}{{- '<|im_start|>' + message.role + '\n'}}{%- if message.role == 'system' or message.role == 'user' %}{%- if message.content is string %}{{- message.content | replace('<image>', '') | replace('<video>', '') }}{%- else %}{%- for item in message.content %}{%- if item.type == 'text' and 'text' in item %}{{- item.text | replace('<image>', '') | replace('<video>', '') }}{%- elif item.type == 'image' and 'image' in item %}{{- '<image>'}}{%- elif item.type == 'video' and 'video' in item %}{{- '<video>'}}{%- else %}{{- raise_exception('Invalid content type. Supported types for system and user are text, image, video.')}}{%- endif %}{%- if not loop.last %}{{- '\n'}}{%- endif %}{%- endfor %}{%- endif %}{%- elif message.role == 'assistant' %}{%- set content = '' %}{%- if message.content is string %}{%- set content = message.content | replace('<image>', '') | replace('<video>', '') %}{%- else %}{%- for item in message.content %}{%- if item.type == 'text' and 'text' in item %}{%- set content = content ~ (item.text | replace('<image>', '') | replace('<video>', '')) %}{%- else %}{{- raise_exception('Invalid content type. Supported type for assistant is text.')}}{%- endif %}{%- endfor %}{%- endif %}{%- set content = content.split('</think>')[-1].lstrip('\n') %}{{- content }}{%- else %}{{- raise_exception('Invalid role. Supported roles are system, user, assistant.')}}{%- endif %}{{- '<|im_end|>\n'}}{%- endfor %}{%- if add_generation_prompt %}{{- '<|im_start|>assistant\n' }}{%- if enable_thinking is defined and enable_thinking is false %}{{- '<think>\n\n</think>\n\n' }}{%- endif %}{%- endif %}",
|
| 231 |
+
"clean_up_tokenization_spaces": false,
|
| 232 |
+
"eos_token": "<|im_end|>",
|
| 233 |
+
"errors": "replace",
|
| 234 |
+
"extra_special_tokens": {},
|
| 235 |
+
"model_max_length": 131072,
|
| 236 |
+
"pad_token": "<|endoftext|>",
|
| 237 |
+
"split_special_tokens": false,
|
| 238 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 239 |
+
"unk_token": null
|
| 240 |
+
}
|
vocab.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|