ikedachin commited on
Commit
c4e2a6c
1 Parent(s): 3fdb260

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +86 -0
README.md CHANGED
@@ -9,6 +9,7 @@ tags:
9
  license: apache-2.0
10
  language:
11
  - en
 
12
  ---
13
 
14
  # Uploaded model
@@ -20,3 +21,88 @@ language:
20
  This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
21
 
22
  [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9
  license: apache-2.0
10
  language:
11
  - en
12
+ - ja
13
  ---
14
 
15
  # Uploaded model
 
21
  This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
22
 
23
  [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
24
+
25
+
26
+ ## 学習データ
27
+
28
+ 使用したSupervised fien-tune用dataset:下記からランダムに20000データを抽出
29
+ DeL-TaiseiOzaki/Tengentoppa-sft-v1.0
30
+ 🌾 ランダムに20000データを取り出して学習
31
+
32
+ SFTに用いた継続事前学習モデル
33
+ ikedachin/llm-jp-3-13b-october-news-e1-all-3-5
34
+
35
+
36
+
37
+ ### 実行コード
38
+
39
+ ```:Python
40
+ # import libraries
41
+ import re
42
+ import json
43
+
44
+ import torch
45
+
46
+ from peft import PeftModel
47
+ from tqdm import tqdm
48
+ from unsloth import FastLanguageModel
49
+
50
+ # define base model_id and peft model_id
51
+ model_id = "llm-jp/llm-jp-3-13b"
52
+ adapter_id = "ikedachin/llm-jp-3-13b-october-news-e1-all-3-5-sft-ozaki-30000"
53
+
54
+ dtype = None
55
+ load_in_4bit
56
+
57
+ # down load base model
58
+ model, tokenizer = FastLanguageModel.from_pretrained(
59
+ model_name=model_id,
60
+ dtype=dtype,
61
+ load_in_4bit=load_in_4bit,
62
+ trust_remote_code=True,
63
+ )
64
+
65
+ # adapt peft model
66
+ model = PeftModel.from_pretrained(model, adapter_id, token = HF_TOKEN)
67
+
68
+
69
+ # prepare dataset elyza-tasks-100-TV_0.jsonl
70
+ datasets_elyza = []
71
+
72
+ with open('./elyza-tasks-100-TV_0.jsonl', 'r') as f:
73
+ item = ""
74
+ for line in f:
75
+ line = line.strip()
76
+ item += line
77
+ if item.endswith("}"):
78
+ datasets_elyza.append(json.loads(item))
79
+ item = ""
80
+
81
+
82
+ # change mode for inference
83
+ FastLanguageModel.for_inference(model)
84
+
85
+ # inferrence
86
+ results = []
87
+
88
+ for dt in tqdm(datasets):
89
+ input = dt["input"]
90
+
91
+ prompt = f"""### 指示\n{input}\n### 回答\n"""
92
+
93
+ inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)
94
+
95
+ outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
96
+ prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答')[-1]
97
+
98
+ results.append({"task_id": dt["task_id"], "input": input, "output": prediction})
99
+
100
+ # create result file as jsonl type
101
+ json_file_id = re.sub(".*/", adapter_id)
102
+ with open(f"/content/{json_file_id}_output.jsonl", 'w', encoding='utf-8') as f:
103
+ for result in results:
104
+ json.dump(result, f, ensure_ascii=False)
105
+ f.write('\n')
106
+
107
+
108
+ ```