File size: 4,363 Bytes
93fca56
 
 
 
 
999ab47
93fca56
 
 
 
 
5cc4626
93fca56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
efe9263
 
93fca56
 
999ab47
 
26a49b4
 
93fca56
 
999ab47
93fca56
999ab47
93fca56
 
5cc4626
93fca56
 
999ab47
 
1108a17
93fca56
 
 
999ab47
 
93fca56
 
 
 
999ab47
93fca56
 
5cc4626
999ab47
26a49b4
93fca56
 
5cc4626
efe9263
 
93fca56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5cc4626
93fca56
 
 
 
 
 
 
 
 
 
 
 
 
5cc4626
93fca56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5cc4626
93fca56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5cc4626
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
---
license: mit
library_name: transformers
pipeline_tag: image-text-to-text
---
# Skywork-R1V3-38B-AWQ

<div align="center">   
  <img src="skywork-logo.png" alt="Introduction Image" width="500" height="400"> 
</div>

## 📖 [R1V3 Report](https://arxiv.org/abs/2507.06167) | 💻 [GitHub](https://github.com/SkyworkAI/Skywork-R1V) | 🌐 [ModelScope](https://modelscope.cn/models/Skywork/Skywork-R1V3-38B) 

<div align="center">

[![GitHub Stars](https://img.shields.io/github/stars/SkyworkAI/Skywork-R1V)](https://github.com/SkyworkAI/Skywork-R1V/stargazers)[![GitHub Forks](https://img.shields.io/github/forks/SkyworkAI/Skywork-R1V)](https://github.com/SkyworkAI/Skywork-R1V/fork)

</div>


## Evaluation

<div align="center">
  <b>Comprehensive performance comparison across text and multimodal reasoning benchmarks.</b>
</div>
<table align="center" border="1" style="border-collapse: collapse; width: 100%;">
  <thead>
    <tr>
      <th>Model</th>
      <th align="center">MMMU</th>
      <th align="center">MathVista</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td colspan="3" align="center"><i
>Proprietary Models</i></td>
    </tr>
    <tr>
      <td>Claude-3.7-Sonnet</td>
      <td align="center">75.0</td>
      <td
 align="center">66.8</td>
    </tr>
    <tr>
      <td>OpenAI-4o</td>
      <td align="center">70.7</td>
      <td align="center">62.9</td>
    </tr>
    <tr>
      <td colspan="3" align="center"><i>Open-Source Models</i></td>
    </tr>
    <tr>
      <td>InternVL3-78B</td>
      <td align="center">72.2</td>
      <td align="center">72.2</td>
    </tr>
    <tr>
      <td>Qwen2.5-VL-72B</td>
      <td align="center">70.3</td>
      <td align="center">74.8</td>
    </tr>
    <tr>
      <td>QvQ-Preview-72B</td>
      <td align="center">70.3</td>
      <td align="center">71.4</td>
    </tr>
    <tr>
      <td>Skywork-R1V3</td>
      <td align="center"><b>76.0</b></td>
      <td align="center"><b>77.1</b></td>
    </tr>
    <tr>
      <td>Skywork-R1V3-AWQ</td>
      <td align="center">66.7</td>
      <td align="center">70.5</td>
    </tr>
  </tbody>
</table>

## Usage
You can use the quantized model with different inference frameworks:
### Using VLLM


#### Python API

```python
import os
from vllm import LLM, SamplingParams
from vllm.entrypoints.chat_utils import load_chat_template
model_name = "Skywork/Skywork-R1V3-38B-AWQ"  # or local path
llm = LLM(model_name, 
          dtype='float16', 
          quantization="awq", 
          gpu_memory_utilization=0.9,
          max_model_len=4096,
          trust_remote_code=True,
         )
# Add your inference code here
```

#### OpenAI-compatible API Server

```bash
MODEL_ID="Skywork/Skywork-R1V3-38B-AWQ"  # or local path
CUDA_VISIBLE_DEVICES=0 \
    python -m vllm.entrypoints.openai.api_server \
    --model $MODEL_ID \
    --dtype float16 \
    --quantization awq \
    --port 23334 \
    --max-model-len 12000 \
    --gpu-memory-utilization 0.9 \
    --trust-remote-code
```

### Using LMDeploy

```python
import os
from lmdeploy import pipeline, TurbomindEngineConfig, ChatTemplateConfig
from lmdeploy.vl import load_image
model_path = "Skywork/Skywork-R1V3-38B-AWQ"  # or local path
engine_config = TurbomindEngineConfig(cache_max_entry_count=0.75) 
chat_template_config = ChatTemplateConfig(model_name=model_path)
pipe = pipeline(model_path, 
                backend_config=engine_config, 
                chat_template_config=chat_template_config,
               )
# Example: Multimodal inference
image = load_image('table.jpg')
response = pipe(('Describe this image?', image))
print(response.text)
```

## Hardware Requirements

The AWQ quantization reduces the memory footprint compared to the original FP16 model. We recommend:

- At least one GPU with 30GB+ VRAM for inference
- For optimal performance with longer contexts, 40GB+ VRAM is recommended

## Citation

If you use this model in your research, please cite:


```bibtex
@misc{shen2025skyworkr1v3technicalreport,
      title={Skywork-R1V3 Technical Report}, 
      author={Wei Shen and Jiangbo Pei and Yi Peng and Xuchen Song and Yang Liu and Jian Peng and Haofeng Sun and Yunzhuo Hao and Peiyu Wang and Jianhao Zhang and Yahui Zhou},
      year={2025},
      eprint={2507.06167},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2507.06167}, 
}
```