taskinui/llm-jp-3-13b-finetune-2

Uploaded model

• Developed by: onewan
• 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.

Sample Use

Jupyter notebook、特にGoogle Colaboratoryで動作させることを想定しています

まずは以下のとおりインストールを実行してください

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

• 以下のコードを実行する前に、
  • コードの中の「--Input your own Hugging Face Token--」の部分にご自身のHugging FaceのTokenを入力してください
    • Access Tokens > Create new tokenで、Token typeは「Finegrained」として、Read, Writeをチェックして、取得するとよいでしょう
  • "elyza-tasks-100-TV_0.jsonl"を同じフォルダに入れてください
    • Colaboratoryの場合は、左のフォルダマークを押して、直下にファイルをドラッグ＆ドロップしてください

import torch
if torch.cuda.get_device_capability()[0] >= 8:
    !pip install --no-deps packaging ninja einops "flash-attn>=2.6.3"

from unsloth import FastLanguageModel
import json
from tqdm import tqdm
import datetime
import pytz

model_name = "onewan/llm-jp-3-13b-finetune-2"
new_model_id = "llm-jp-3-13b-finetune-2"

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 = "--Input your own Hugging Face 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 = ""

FastLanguageModel.for_inference(model)

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})

now = datetime.datetime.now(pytz.timezone('Asia/Tokyo')).strftime("%Y%m%d-%H%M%S")

with open(f"{new_model_id}_output_{now}.jsonl", 'w', encoding='utf-8') as f:
  for result in results:
      json.dump(result, f, ensure_ascii=False)
      f.write('\n')