llm-jp-3-13b-ftELZ-2

Uploaded model

• Developed by: satoyutaka
• 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.

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

sample of use(python)

from transformers import (

AutoModelForCausalLM,

AutoTokenizer,

BitsAndBytesConfig,

)

モデルのロード

import torch from tqdm import tqdm import json

HF_TOKEN = "Hugging Face Token"　#Hugging Face のAPIキーを入力（read)

model_name = "satoyutaka/llm-jp-3-13b-ftELZ-2"　#作成したモデル名

量子化パラメータの設定

bnb_config = BitsAndBytesConfig(

load_in_4bit=True,

bnb_4bit_quant_type="nf4",

bnb_4bit_compute_dtype=torch.bfloat16,

bnb_4bit_use_double_quant=False,

)

問題文の読み込み

datasets = []

with open("elyza-tasks-100-TV_0.jsonl", "r") as f:　#ファイルを格納したパスに書き換えてください。

item = ""

for line in f:

  line = line.strip()

  item += line

  if item.endswith("}"):

    datasets.append(json.loads(item))

    item = ""

results = []

推論

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})

提出ファイルの作成

import re

model_name = re.sub(".*/", "", model_name)

with open(f"./{model_name}-outputs.jsonl", 'w', encoding='utf-8') as f:

for result in results:


    json.dump(result, f, ensure_ascii=False)  # ensure_ascii=False for handling non-ASCII characters


    f.write('\n')