Views
No views yet
kaonai/kaon-c-mistral-12b-v0.1-dpo-v0.1进行GPT-4数据优化和DPO训练的完整合并模型。1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3
4# 加载完整模型（无需额外LoRA加载）
5model = AutoModelForCausalLM.from_pretrained(
6 kaonai/kaon-w-personal-dpo-v1.2-pro,
7 torch_dtype=torch.bfloat16,
8 device_map=auto,
9 trust_remote_code=True
10)
11
12tokenizer = AutoTokenizer.from_pretrained(kaonai/kaon-w-personal-dpo-v1.2-pro)
13
14# 对话示例
15messages = [
16 {role: system, content: 你是一个有用的AI助手。},
17 {role: user, content: 请介绍一下你自己}
18]
19
20text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
21inputs = tokenizer(text, return_tensors=pt)
22
23with torch.no_grad():
24 outputs = model.generate(
25 **inputs,
26 max_new_tokens=512,
27 temperature=0.7,
28 do_sample=True,
29 top_p=0.9,
30 pad_token_id=tokenizer.eos_token_id
31 )
32
33response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
34print(response)