Views
No views yet
1from transformers import AutoTokenizer
2import transformers
3import torch
4
5model_name = "opencsg/OpenCSG-R1-Qwen2.5-Math-3B-V1"
6model = AutoModelForCausalLM.from_pretrained(
7 model_name,
8 torch_dtype="auto",
9 device_map="auto"
10)
11tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
12
13messages = [
14 {
15 "role": "user",
16 "content": f"请你帮我用因式分解拆解123958102这个数字。在 <think> </think> 标签中输出思考过程，并在 <answer> </answer> 标签中返回最终结果，例如 <answer> (1 + 2) / 3 </answer>。在 <think> 标签中逐步思考。",
17 },
18 {
19 "role": "assistant",
20 "content": "让我们逐步解决这个问题。\n<think>",
21 },
22]
23text = tokenizer.apply_chat_template(
24 messages,
25 tokenize=False,
26 continue_final_message=True,
27 # add_generation_prompt=True
28)
29model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
30
31generated_ids = model.generate(
32 **model_inputs,
33 max_new_tokens=512,
34 temperature=0.6
35)
36generated_ids = [
37 output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
38]
39
40response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]1from transformers import AutoTokenizer
2import transformers
3import torch
4
5model_name = "opencsg/OpenCSG-R1-Qwen2.5-Math-3B-V1"
6model = AutoModelForCausalLM.from_pretrained(
7 model_name,
8 torch_dtype="auto",
9 device_map="auto"
10)
11tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
12
13messages = [
14 {
15 "role": "user",
16 "content": f"请你帮我用因式分解拆解123958102这个数字。在 <think> </think> 标签中输出思考过程，并在 <answer> </answer> 标签中返回最终结果，例如 <answer> (1 + 2) / 3 </answer>。在 <think> 标签中逐步思考。",
17 },
18 {
19 "role": "assistant",
20 "content": "让我们逐步解决这个问题。\n<think>",
21 },
22]
23text = tokenizer.apply_chat_template(
24 messages,
25 tokenize=False,
26 continue_final_message=True,
27 # add_generation_prompt=True
28)
29model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
30
31generated_ids = model.generate(
32 **model_inputs,
33 max_new_tokens=512,
34 temperature=0.6
35)
36generated_ids = [
37 output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
38]
39
40response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]