Llama-2-7BをQLoRAファインチューニングしてござる口調にする

!python qlora.py \
    --model_name meta-llama/Llama-2-7b-hf \
    --output_dir "./output/test_peft" \
    --dataset "alpaca" \
    --max_steps 1000 \
    --use_auth \
    --logging_steps 10 \
    --save_strategy steps \
    --data_seed 42 \
    --save_steps 50 \
    --save_total_limit 40 \
    --max_new_tokens 32 \
    --dataloader_num_workers 1 \
    --group_by_length \
    --logging_strategy steps \
    --remove_unused_columns False \
    --do_train \
    --lora_r 64 \
    --lora_alpha 16 \
    --lora_modules all \
    --double_quant \
    --quant_type nf4 \
    --bf16 \
    --bits 4 \
    --warmup_ratio 0.03 \
    --lr_scheduler_type constant \
    --gradient_checkpointing \
    --source_max_len 16 \
    --target_max_len 512 \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 16 \
    --eval_steps 187 \
    --learning_rate 0.0002 \
    --adam_beta2 0.999 \
    --max_grad_norm 0.3 \
    --lora_dropout 0.1 \
    --weight_decay 0.0 \
    --seed 0 \
    --load_in_4bit \
    --use_peft \
    --batch_size 4 \
    --gradient_accumulation_steps 2

# 推論
import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

# ベースモデルをロード
model_id = "meta-llama/Llama-2-7b-hf"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_use_double_quant=True,
        bnb_4bit_quant_type="nf4",
        bnb_4bit_compute_dtype=torch.bfloat16
    ),
    device_map={"":0}
)

# LoRAのロード
model = PeftModel.from_pretrained(
    model,
    "./output/test_peft/checkpoint-1000/adapter_model/",
    device_map={"":0}
)
model.eval()

# リポジトリでチューニングに使われているプロンプトと同じもの
# https://github.com/artidoro/qlora/blob/main/qlora.py#L517-L528
prompt = """Below is an instruction that describes a task.
Write a response that appropriately completes the request.

### Instruction:
富士山とは？

### Response: """

<s> Below is an instruction that describes a task.
Write a response that appropriately completes the request.

### Instruction:
富士山とは？

### Response: 
富士山は、日本の山でござる。知らんけど。</s>

<s> Below is an instruction that describes a task.
Write a response that appropriately completes the request.

### Instruction:
侍とは？

### Response: 
侍は、戦士や武士として、独特の価値観と規範を持つ人間の集団でござる。知らんけど。</s>

<s> Below is an instruction that describes a task.
Write a response that appropriately completes the request.

### Instruction:
人工知能とは？

### Response: 
人工知能は、人間が持つ知識、感情、行動を模倣したコンピュータプログラムでござる。知らんけど。</s>