Uploaded model

Developed by: CiderRoad
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.

使用方法
松尾研大規模言語モデル講座2024のコンペのタスクの推論方法を以下に記載します。

0.事前準備
実行するコードファイルと同じ場所に「lyza-tasks-100-TV_0.jsonl」を配置する。

1.環境セットアップ
pip install unsloth
pip uninstall unsloth -y && pip install --upgrade --no-cache-dir "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"

2.推論コード
from unsloth import FastLanguageModel
import torch
import json
from tqdm import tqdm

モデルの読み込み

model_name = "CiderRoad/llm-jp-3-13b-finetune-2"
max_seq_length = 2048
dtype = None
load_in_4bit = True

モデルとトークナイザーの読み込み

model, tokenizer = FastLanguageModel.from_pretrained(
model_name=model_name,
max_seq_length=max_seq_length,
dtype=dtype,
load_in_4bit=load_in_4bit,
token="YOUR_HUGGINGFACE_TOKEN" # Hugging Faceのアクセストークンを記入
)

FastLanguageModel.for_inference(model)

データセットの読み込み

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 = ""

推論の実行

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})

結果の書き出し

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