Views
No views yet
pip install -U bitsandbytes transformers accelerate datasets peft1from transformers import (
2 AutoModelForCausalLM,
3 AutoTokenizer,
4 BitsAndBytesConfig,
5)
6from peft import PeftModel
7import torch
8from tqdm import tqdm
9import json
10
11model_id = "models/models--llm-jp--llm-jp-3-13b/snapshots/cd3823f4c1fcbb0ad2e2af46036ab1b0ca13192a"
12# model_id = "llm-jp/llm-jp-3-13b"
13adapter_id = "ken0x0a/llm-jp-3-13b-finetune"
14out_filename = "./llm-jp-3-13b-outputs.jsonl"
15
16# QLoRA config
17bnb_config = BitsAndBytesConfig(
18 load_in_4bit=True,
19 bnb_4bit_quant_type="nf4",
20 bnb_4bit_compute_dtype=torch.bfloat16,
21)
22
23# Load model
24model = AutoModelForCausalLM.from_pretrained(
25 model_id,
26 quantization_config=bnb_config,
27 device_map="auto",
28)
29
30tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
31
32model = PeftModel.from_pretrained(model, adapter_id)
33
34# 実行
35datasets = []
36with open("./elyza-tasks-100-TV_0.jsonl", "r") as f:
37 item = ""
38 for line in f:
39 line = line.strip()
40 item += line
41 if item.endswith("}"):
42 datasets.append(json.loads(item))
43 item = ""
44
45results = []
46for data in tqdm(datasets):
47 input = data["input"]
48
49 prompt = f"""### 指示
50 {input}
51 ### 回答
52 """
53
54 tokenized_input = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt").to(model.device)
55 attention_mask = torch.ones_like(tokenized_input)
56 with torch.no_grad():
57 outputs = model.generate(
58 tokenized_input,
59 attention_mask=attention_mask,
60 max_new_tokens=100,
61 do_sample=False,
62 repetition_penalty=1.2,
63 pad_token_id=tokenizer.eos_token_id
64 )[0]
65 output = tokenizer.decode(outputs[tokenized_input.size(1):], skip_special_tokens=True)
66 results.append({"task_id": data["task_id"], "output": output})
67
68with open(out_filename, 'w', encoding='utf-8') as f:
69 for result in results:
70 json.dump(result, f, ensure_ascii=False)
71 f.write('\n')