File size: 2,977 Bytes
a108004
 
 
 
4e5f73a
 
 
a108004
 
4e5f73a
 
 
a108004
 
4e5f73a
 
a108004
 
635bd2a
 
325b999
635bd2a
4e5f73a
 
325b999
 
 
 
 
 
 
4e5f73a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0451c76
4e5f73a
 
 
 
0451c76
4e5f73a
 
 
e781167
4e5f73a
 
 
 
 
 
 
 
 
 
 
a108004
4e5f73a
325b999
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
---
base_model: llm-jp/llm-jp-3-13b
license: apache-2.0
language:
- ja
datasets:
- kajuma/dpo_1
---

# Model Card for JunichiroMorita/llm-jp-3-13b-it_lora_20241216

## Model Details

- **Developed by:** JunichiroMorita
- **Language(s) (NLP):** Japanese
- **License:** Apache license 2.0
- **Finetuned from model :** llm-jp/llm-jp-3-13b

## Description

This model was developed for use in a competition, specifically for [松尾研大規模言語モデル講座2024](https://weblab.t.u-tokyo.ac.jp/lecture/course-list/large-language-model/).

## Uses

```python
!pip install unsloth
!pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
!pip install -U torch
!pip install -U peft
```

```python
from unsloth import FastLanguageModel
from peft import PeftModel
import torch
import json
from tqdm import tqdm
import re

model_id = "llm-jp/llm-jp-3-13b"
adapter_id = f"JunichiroMorita/llm-jp-3-13b-it_lora_20241216"

HF_TOKEN = 'your_hugging_face_token'

dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=model_id,
    dtype=dtype,
    load_in_4bit=load_in_4bit,
    trust_remote_code=True,
)

model = PeftModel.from_pretrained(model, adapter_id, token=HF_TOKEN)

datasets = []
with open("./elyza-tasks-100-TV_0.jsonl", "r") as f:
    item = ""
    for line in f:
      line = line.strip()
      item += line
      if item.endswith("}"):
        datasets.append(json.loads(item))
        item = ""

FastLanguageModel.for_inference(model)

results = []
for dt in tqdm(datasets):
  input = dt["input"]

  prompt = f"""### 指示\n{input}\n\n### 回答\n"""

  inputs = tokenizer([prompt], return_tensors = "pt").to(model.device)

  outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True, do_sample=False, repetition_penalty=1.2)
  prediction = tokenizer.decode(outputs[0], skip_special_tokens=True).split('\n### 回答\n')[-1]

  results.append({"task_id": dt["task_id"], "input": input, "output": prediction})

with open(f'./llm-jp-3-13b-it_lora_20241216_output.jsonl', 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')
```

## Training Details

### Training Data
- [kajuma/dpo_1](https://huggingface.co/datasets/kajuma/dpo_1)

### Training Procedure

This model was fine-tuned using LoRA (Low-Rank Adaptation) to optimize training efficiency and minimize computational overhead while maintaining performance. The fine-tuning process utilized Japanese instruction data specifically designed for LLMs to enhance its capabilities in understanding and generating Japanese-language instructions.

This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)