# conda環境の構築
wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"  

# このコマンドではいくつか質問があるので答えて下さい。おそらくインストール先のデフォルトは/root/miniforge3かと思います  
bash Miniforge3-$(uname)-$(uname -m).sh  

# 以下、インストール先が/root/miniforge3であることを前提とします  
export PATH=/root/miniforge3/bin:$PATH  
conda init  

# ここで一度、terminalを立ち上げ直す必要があります。  
# 以下のリンク先に従い環境を作ります。  
# https://docs.unsloth.ai/get-started/installation/conda-install  
conda create --name unsloth_env python=3.10 pytorch-cuda=12.1 pytorch cudatoolkit xformers -c pytorch -c nvidia -c xformers -y  
conda activate unsloth_env  
pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"  
pip install --no-deps "trl<0.9.0" peft accelerate bitsandbytes  

# jupyter notebook用のセットアップ。  
conda install -c conda-forge ipykernel  
python -m ipykernel install --user --name=unsloth_env --display-name "Python (unsloth_env)"  

Follow these steps, run in the notebook:

load model

shell

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

python

from unsloth import FastLanguageModel
import torch
import json

model_name = "tomofusa/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 = "hf-token", # In the Google Colab case, it call from ENV. If you want to write the token directly, please comment it out.
)
FastLanguageModel.for_inference(model)

Set up datasets and run inference.

Upload elyza-tasks-100-TV_0.jsonl to your workspace in manual.

python

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 = ""

python

from tqdm import tqdm

# inference
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})

Save results to jsonl.

python

file_name = model_name.replace("/", "_") + "_output.jsonl"
with open(f"./{file_name}", 'w', encoding='utf-8') as f:
    for result in results:
        json.dump(result, f, ensure_ascii=False)
        f.write('\n')