---
title: Unichat-qwen2.5-32B-c1
canonical_url: "https://www.modelscope.cn/models/UnicomAI/Unichat-qwen2.5-32B-c1"
md_url: "https://www.modelscope.cn/models/UnicomAI/Unichat-qwen2.5-32B-c1.md"
repository: UnicomAI/Unichat-qwen2.5-32B-c1
last_updated: 2025-01-18
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - qwen2
architectures:
  - Qwen2ForCausalLM
parameters: 32.8B
tensor_type:
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - Pytorch
inference_backends:
  - "deploy_task text/emb"
  - "lmdeploy 0.9.1"
  - "lmdeploy_turbomind 0.9.1"
  - "sglang 0.5.2"
  - "vllm 0.9.2"
downloads: 585
stars: 4
---

# Unichat-qwen2.5-32B-c1

> Unichat-qwen2.5-32B-c1 - UnicomAI 在 ModelScope 开源的模型。Unichat-32B-c1 --- GitHub

UnicomAI/Unichat-qwen2.5-32B-c1 是 ModelScope 魔搭社区上的 32.8B 参数text-generation模型，可用 deploy_task text/emb、lmdeploy 0.9.1、lmdeploy_turbomind 0.9.1 部署。

- **Repository**: UnicomAI/Unichat-qwen2.5-32B-c1
- **Tasks**: text-generation
- **Parameters**: 32.8B
- **Inference backends**: deploy_task text/emb, lmdeploy 0.9.1, lmdeploy_turbomind 0.9.1, sglang 0.5.2, vllm 0.9.2
- **Downloads**: 585
- **Stars**: 4
- **Last updated**: 2025-01-18

Source: https://www.modelscope.cn/models/UnicomAI/Unichat-qwen2.5-32B-c1

---

Unichat-32B-c1
---
[GitHub](https://github.com/UnicomAI/Unichat-32B-c1) 

介绍
---
元景思维链模型，目前我们发布基于Qwen2.5-32B-Instruct实现的版本，后续将开源基于元景34B模型（UniChat 34B）的版本并公开我们的技术报告。

测评结果
---

| Model                | GSM8K | MATH500 | OlympidiaBench   | AIME2024 | AMC23 |
|-----------------------|---------------------|--------|-------|------------|------------|
| GPT-o1 mini   | 96.5                    | 93.7   | 78.8  | 66.7       | 92.5 |
|   GPT4o       | 90.4                    | 79.3   | 48.6  | 20.0       | 62.5 |
|  Deepseek V3  | 95.8                    | 90.2   | 50.1  | 40.0       | 80.0 |
| Qwen2.5-MATH-72B | 95.8                 | 85.9   | 49.0  | 30.0       | 70.0 |
|  Qwen-QwQ  | 95.6                       | 90.0   | 57.3  | 40.0       | 85.0 |
|  Unichat-qwen2.5-32B-c1   | 95.8                | 90.6   | 59.6  | 43.3       | 90.0 |

测评脚本使用[Qwen2.5-Math](https://github.com/QwenLM/Qwen2.5-Math)，推理长度设为12288

快速开始
---
这里提供代码片段来使用模型进行推理。
```python
from modelscope import AutoModelForCausalLM, AutoTokenizer

model_name = "UnicomAI/Unichat-32B-c1"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "9.8和9.11哪个数比较大？"
messages = [
    {"role": "system", "content": "请一步一步推理, 并把最终答案放在 \\boxed{{}} 里。"},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=8192
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)

```