llm-jp-3-13b-finetune-3

LLM-JP-3-13B ファインチューニングモデル

使用例

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel, LoraConfig
import torch
from tqdm import tqdm

HF_TOKEN = "Your HF token"
BASE_MODEL_ID = "llm-jp/llm-jp-3-13b"
ADAPTER_ID = "Enverop3/llm-jp-3-13b-finetune" 

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

# モデル読み込み
model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL_ID,
    quantization_config=bnb_config,
    device_map="auto",
    token = HF_TOKEN
)

# Tokenizer読み込み
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID, trust_remote_code=True, token=HF_TOKEN)

# FTしたアダプタを接続
model = PeftModel.from_pretrained(model, ADAPTER_ID, token=HF_TOKEN)

データセットを使用した推論例

results = []
for data in tqdm(datasets):
    input = data["input"]
    prompt = (
        '### 指示\n'
        f'{input}\n'
        '\n'
        '### 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})