84basi/llm-jp-3-13b-finetune-2.1

Uploaded model

• Developed by: 84basi
• License: apache-2.0
• Finetuned from model : llm-jp/llm-jp-3-13b

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

%%capture
!pip install unsloth
!pip uninstall unsloth -y && pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"

from unsloth import FastLanguageModel
import torch
import json

model_name = "84basi/llm-jp-3-13b-finetune-2.1"
token = "Hugging Face Token" #@param {type:"string"}

max_seq_length = 2048
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = model_name,
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    token = token,
)
FastLanguageModel.for_inference(model)

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 = ""

from tqdm import tqdm

results = []
for dt in tqdm(datasets):
  input = dt["input"]
  prompt = f"""### 指示\n{input}\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')[-1]
  results.append({"task_id": dt["task_id"], "input": input, "output": prediction})

with open(f"/content/llm-jp-3-13b-finetune-2.1_output-2.jsonl", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')

!pip install python-docx

import json

from docx import Document  # pip install python-docxでインストールする
from docx.shared import Inches, Pt, RGBColor
from docx.enum.text import WD_ALIGN_PARAGRAPH


def read_jsonl_data(jsonl_path):
    """
    提出用jsonlを読み、json形式で返す

    Args:
        jsonl_path (str): 提出用jsonlへのパス

    Returns:
        jsonデータ (list of dict)
    """
    results = []
    with open(jsonl_path, 'r', encoding='utf-8') as f:
        for line in f:
            line = line.strip()
            if line:
                try:
                    results.append(json.loads(line))
                except json.JSONDecodeError as e:
                    print(f"JSONデコードエラー（行内容を確認してください）: {e}")
    return results


def json_to_word(json_data, output_file):
    """
    JSONデータをWord文書に変換する

    Args:
        json_data (list of dict): JSONデータのリスト
        output_file (str): 出力するWordファイルの名前
    """
    doc = Document()

    title = doc.add_heading('LLM Output Analysis', 0)
    title.alignment = WD_ALIGN_PARAGRAPH.CENTER

    for item in json_data:
        task_id = item.get("task_id", "No Task ID")
        doc.add_heading(f'Task ID: {task_id}', level=1)

        doc.add_heading('Input:', level=2)
        input_text = item.get("input", "No Input")
        input_para = doc.add_paragraph()
        input_para.add_run(input_text).bold = False

        doc.add_heading('Output:', level=2)
        output_text = item.get("output", "No Output")
        output_para = doc.add_paragraph()
        output_para.add_run(output_text).bold = False

        doc.add_paragraph('=' * 50)
    doc.save(output_file)

jsonl_path = '/content/llm-jp-3-13b-finetune-2.1_output-2.jsonl'
output_file = '/content/llm-jp-3-13b-finetune-2.1_output-2.docx'
jsonl_data = read_jsonl_data(jsonl_path)
json_to_word(jsonl_data, output_file)