nakashi104/llm-jp-3-13b-it_lora

Uploaded model

• Developed by: shiki07
• 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.

how to use

本アダプタを用いて，ELYZA-tasks-100-TVの出力を得る推論コードです．Jupyter Notebook環境を想定しています．

使用ライブラリのインストール

!pip install -U bitsandbytes
!pip install -U transformers
!pip install -U accelerate
!pip install -U datasets
!pip install -U peft

準備

from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    BitsAndBytesConfig,
)
from peft import PeftModel
import torch
from tqdm import tqdm
import json

HF_TOKEN = "Hugging Face Token" #Write権限のHFトークンを設定
base_model_id = "llm-jp/llm-jp-3-13b" 
adapter_id = "shiki07/llm-jp-3-13b-it_lora"
eval_data_path = "./elyza-tasks-100-TV_0.jsonl" # elyza-tasks-100-TVのパスを指定

時間がかかります．

# QLoRA config
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)

# Load model
model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    quantization_config=bnb_config,
    device_map="auto",
    token = HF_TOKEN
)

tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True, token = HF_TOKEN)
model = PeftModel.from_pretrained(model, adapter_id, token = HF_TOKEN)

データ読み込みと推論

# データセットの読み込み。
datasets = []
with open(eval_data_path, "r") as f:
    item = ""
    for line in f:
      line = line.strip()
      item += line
      if item.endswith("}"):
        datasets.append(json.loads(item))
        item = ""

# llmjp
results = []
for data in tqdm(datasets):

  input = data["input"]

  prompt = f"""### 指示
  {input}
  ### 回答
  """

  tokenized_input = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt").to(model.device)
  attention_mask = torch.ones_like(tokenized_input)
  with torch.no_grad():
      outputs = model.generate(
          tokenized_input,
          attention_mask=attention_mask,
          max_new_tokens=100,
          do_sample=False,
          repetition_penalty=1.2,
          pad_token_id=tokenizer.eos_token_id
      )[0]
  output = tokenizer.decode(outputs[tokenized_input.size(1):], skip_special_tokens=True)

  results.append({"task_id": data["task_id"], "input": input, "output": output})

import re
jsonl_id = re.sub(".*/", "", adapter_id)
with open(f"./{jsonl_id}-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')

以上です．
.jsonlファイルが推論結果のファイルになります．

Instruction tuning

The models have been fine-tuned on the following datasets.
日本語インストラクションデータ：ichikara-instruction