LGAI-EXAONE/EXAONE-Deep-32B
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Update LM-Studio guide

Browse Source
This commit is contained in:
Cherrytest
2025-03-18 05:36:41 +00:00
parent ae1d939204
commit eb16a8aa4f
28 changed files with 315059 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

317
README.md
View File

@@ -1,47 +1,282 @@
---
license: Apache License 2.0
#model-type:
##如 gpt、phi、llama、chatglm、baichuan 等
#- gpt
#domain:
##如 nlp、cv、audio、multi-modal
#- nlp
#language:
##语言代码列表 https://help.aliyun.com/document_detail/215387.html?spm=a2c4g.11186623.0.0.9f8d7467kni6Aa
#- cn
#metrics:
##如 CIDEr、Blue、ROUGE 等
#- CIDEr
#tags:
##各种自定义，包括 pretrained、fine-tuned、instruction-tuned、RL-tuned 等训练方法和其他
#- pretrained
#tools:
##如 vllm、fastchat、llamacpp、AdaSeq 等
#- vllm
base_model: LGAI-EXAONE/EXAONE-3.5-32B-Instruct
base_model_relation: finetune
license: other
license_name: exaone
license_link: LICENSE
language:
- en
- ko
tags:
- lg-ai
- exaone
- exaone-deep
pipeline_tag: text-generation
library_name: transformers
---
### 当前模型的贡献者未提供更加详细的模型介绍。模型文件和权重，可浏览“模型文件”页面获取。
#### 您可以通过如下git clone命令，或者ModelScope SDK来下载模型
SDK下载
```bash
#安装ModelScope
pip install modelscope
```
<p align="center">
<img src="assets/EXAONE_Symbol+BI_3d.png", width="300", style="margin: 40 auto;">
<br>
# EXAONE-Deep-32B
## Introduction
We introduce EXAONE Deep, which exhibits superior capabilities in various reasoning tasks including math and coding benchmarks, ranging from 2.4B to 32B parameters developed and released by LG AI Research. Evaluation results show that 1) EXAONE Deep **2.4B** outperforms other models of comparable size, 2) EXAONE Deep **7.8B** outperforms not only open-weight models of comparable scale but also a proprietary reasoning model OpenAI o1-mini, and 3) EXAONE Deep **32B** demonstrates competitive performance against leading open-weight models.
For more details, please refer to our [documentation](https://arxiv.org/abs/2503.12524), [blog](https://www.lgresearch.ai/news/view?seq=543) and [GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep).
<p align="center">
<img src="assets/exaone_deep_overall_performance.png", width="100%", style="margin: 40 auto;">
This repository contains the reasoning 32B language model with the following features:
- Number of Parameters (without embeddings): 30.95B
- Number of Layers: 64
- Number of Attention Heads: GQA with 40 Q-heads and 8 KV-heads
- Vocab Size: 102,400
- Context Length: 32,768 tokens
## Quickstart
We recommend to use `transformers` v4.43.1 or later.
Here is the code snippet to run conversational inference with the model:
```python
#SDK模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('LGAI-EXAONE/EXAONE-Deep-32B')
```
Git下载
```
#Git模型下载
git clone https://www.modelscope.cn/LGAI-EXAONE/EXAONE-Deep-32B.git
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread
model_name = "LGAI-EXAONE/EXAONE-Deep-32B"
streaming = True # choose the streaming option
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.bfloat16,
trust_remote_code=True,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
messages = [
{"role": "user", "content": "How many golf balls can fit in a school bus?"}
]
input_ids = tokenizer.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_tensors="pt"
)
if streaming:
streamer = TextIteratorStreamer(tokenizer)
thread = Thread(target=model.generate, kwargs=dict(
input_ids=input_ids.to("cuda"),
eos_token_id=tokenizer.eos_token_id,
max_new_tokens=32768,
do_sample=True,
temperature=0.6,
top_p=0.95,
streamer=streamer
))
thread.start()
for text in streamer:
print(text, end="", flush=True)
else:
output = model.generate(
input_ids.to("cuda"),
eos_token_id=tokenizer.eos_token_id,
max_new_tokens=32768,
do_sample=True,
temperature=0.6,
top_p=0.95,
)
print(tokenizer.decode(output[0]))
```
<p style="color: lightgrey;">如果您是本模型的贡献者，我们邀请您根据<a href="https://modelscope.cn/docs/ModelScope%E6%A8%A1%E5%9E%8B%E6%8E%A5%E5%85%A5%E6%B5%81%E7%A8%8B%E6%A6%82%E8%A7%88" style="color: lightgrey; text-decoration: underline;">模型贡献文档</a>，及时完善模型卡片内容。</p>
> ### Note
> The EXAONE Deep models are trained with an optimized configuration,
> so we recommend following the [Usage Guideline](#usage-guideline) section to achieve optimal performance.
## Evaluation
The following table shows the evaluation results of reasoning tasks such as math and coding. The full evaluation results can be found in the [documentation](https://arxiv.org/abs/2503.12524).
<table>
<tr>
<th>Models</th>
<th>MATH-500 (pass@1)</th>
<th>AIME 2024 (pass@1 / cons@64)</th>
<th>AIME 2025 (pass@1 / cons@64)</th>
<th>CSAT Math 2025 (pass@1)</th>
<th>GPQA Diamond (pass@1)</th>
<th>Live Code Bench (pass@1)</th>
</tr>
<tr>
<td>EXAONE Deep 32B</td>
<td>95.7</td>
<td>72.1 / <strong>90.0</strong></td>
<td>65.8 / <strong>80.0</strong></td>
<td><strong>94.5</strong></td>
<td>66.1</td>
<td>59.5</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-32B</td>
<td>94.3</td>
<td>72.6 / 83.3</td>
<td>55.2 / 73.3</td>
<td>84.1</td>
<td>62.1</td>
<td>57.2</td>
</tr>
<tr>
<td>QwQ-32B</td>
<td>95.5</td>
<td><strong>79.5</strong> / 86.7</td>
<td><strong>67.1</strong> / 76.7</td>
<td>94.4</td>
<td>63.3</td>
<td>63.4</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Llama-70B</td>
<td>94.5</td>
<td>70.0 / 86.7</td>
<td>53.9 / 66.7</td>
<td>88.8</td>
<td>65.2</td>
<td>57.5</td>
</tr>
<tr>
<td>DeepSeek-R1 (671B)</td>
<td><strong>97.3</strong></td>
<td>79.8 / 86.7</td>
<td>66.8 / <strong>80.0</strong></td>
<td>89.9</td>
<td><strong>71.5</strong></td>
<td><strong>65.9</strong></td>
</tr>
<tr>
<th colspan="7" height="30px"></th>
</tr>
<tr>
<td>EXAONE Deep 7.8B</td>
<td><strong>94.8</strong></td>
<td><strong>70.0</strong> / <strong>83.3</strong></td>
<td><strong>59.6</strong> / <strong>76.7</strong></td>
<td><strong>89.9</strong></td>
<td><strong>62.6</strong></td>
<td><strong>55.2</strong></td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-7B</td>
<td>92.8</td>
<td>55.5 / <strong>83.3</strong></td>
<td>38.5 / 56.7</td>
<td>79.7</td>
<td>49.1</td>
<td>37.6</td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Llama-8B</td>
<td>89.1</td>
<td>50.4 / 80.0</td>
<td>33.6 / 53.3</td>
<td>74.1</td>
<td>49.0</td>
<td>39.6</td>
</tr>
<tr>
<td>OpenAI o1-mini</td>
<td>90.0</td>
<td>63.6 / 80.0</td>
<td>54.8 / 66.7</td>
<td>84.4</td>
<td>60.0</td>
<td>53.8</td>
</tr>
<tr>
<th colspan="7" height="30px"></th>
</tr>
<tr>
<td>EXAONE Deep 2.4B</td>
<td><strong>92.3</strong></td>
<td><strong>52.5</strong> / <strong>76.7</strong></td>
<td><strong>47.9</strong> / <strong>73.3</strong></td>
<td><strong>79.2</strong></td>
<td><strong>54.3</strong></td>
<td><strong>46.6</strong></td>
</tr>
<tr>
<td>DeepSeek-R1-Distill-Qwen-1.5B</td>
<td>83.9</td>
<td>28.9 / 52.7</td>
<td>23.9 / 36.7</td>
<td>65.6</td>
<td>33.8</td>
<td>16.9</td>
</tr>
</table>
## Deployment
EXAONE Deep models can be inferred in the various frameworks, such as:
- `TensorRT-LLM`
- `vLLM`
- `SGLang`
- `llama.cpp`
- `Ollama`
- `LM-Studio`
Please refer to our [EXAONE Deep GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep) for more details about the inference frameworks.
## Quantization
We are working on quantized versions of EXAONE Deep models in both **AWQ** and **GGUF** formats. We will update this section with detailed instructions upon release.
## Usage Guideline
To achieve the expected performance, we recommend using the following configurations:
1. Ensure the model starts with `<thought>\n` for reasoning steps. The model's output quality may be degraded when you omit it. You can easily apply this feature by using `tokenizer.apply_chat_template()` with `add_generation_prompt=True`. Please check the example code on [Quickstart](#quickstart) section.
2. The reasoning steps of EXAONE Deep models enclosed by `<thought>\n...\n</thought>` usually have lots of tokens, so previous reasoning steps may be necessary to be removed in multi-turn situation. The provided tokenizer handles this automatically.
3. Avoid using system prompt, and build the instruction on the user prompt.
4. When it comes to math problems, include **"Please reason step by step, and put your final answer within \boxed{}."** in your prompt.
5. In our evaluation, we use `temperature=0.6` and `top_p=0.95` for generation.
6. When evaluating the models, it is recommended to test multiple times to assess the expected performance accurately.
## Limitation
The EXAONE language model has certain limitations and may occasionally generate inappropriate responses. The language model generates responses based on the output probability of tokens, and it is determined during learning from training data. While we have made every effort to exclude personal, harmful, and biased information from the training data, some problematic content may still be included, potentially leading to undesirable responses. Please note that the text generated by EXAONE language model does not reflects the views of LG AI Research.
- Inappropriate answers may be generated, which contain personal, harmful or other inappropriate information.
- Biased responses may be generated, which are associated with age, gender, race, and so on.
- The generated responses rely heavily on statistics from the training data, which can result in the generation of
semantically or syntactically incorrect sentences.
- Since the model does not reflect the latest information, the responses may be false or contradictory.
LG AI Research strives to reduce potential risks that may arise from EXAONE language models. Users are not allowed
to engage in any malicious activities (e.g., keying in illegal information) that may induce the creation of inappropriate
outputs violating LG AI’s ethical principles when using EXAONE language models.
## License
The model is licensed under [EXAONE AI Model License Agreement 1.1 - NC](./LICENSE)
## Citation
```
@article{exaone-deep,
title={EXAONE Deep: Reasoning Enhanced Language Models},
author={{LG AI Research}},
journal={arXiv preprint arXiv:2503.12524},
year={2025}
}
```
## Contact
LG AI Research Technical Support: contact_us@lgresearch.ai