Unichat-qwen2.5-32B-c1

---

GitHub | ModelScope | WiseModel

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，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。

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)