davidkim205's picture
Create README.md
0a95e86
|
raw
history blame
7.15 kB
metadata
language:
  - en
  - ko
pipeline_tag: text-generation
inference: false
tags:
  - facebook
  - meta
  - pytorch
  - llama
  - llama-2
  - llama-2-chat
library_name: peft

komt : korean multi task instruction tuning model

multi task instruction tuning.jpg

Recently, due to the success of ChatGPT, numerous large language models have emerged in an attempt to catch up with ChatGPT's capabilities. However, when it comes to Korean language performance, it has been observed that many models still struggle to provide accurate answers or generate Korean text effectively. This study addresses these challenges by introducing a multi-task instruction technique that leverages supervised datasets from various tasks to create training data for Large Language Models (LLMs).

Model Details

  • Model Developers : davidkim(changyeon kim)
  • Repository : https://github.com/davidkim205/komt
  • Model Architecture : The komt-mistral-7b-v1-dpo is is a fine-tuned version of the komt-mistral-7b-v1(original model : Mistral-7B-Instruct-v0.1).

Dataset

Hardware and Software

  • nvidia driver : 535.54.03
  • CUDA Version: 12.2

Training

Refer https://github.com/davidkim205/komt

Prompt template: Mistral

<s>[INST] {prompt} [/INST]</s>

Usage

import torch

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import PeftModel, PeftConfig
from transformers import TextStreamer, GenerationConfig


model='davidkim205/komt-mistral-7b-v1'
peft_model_name = 'davidkim205/komt-mistral-7b-v1-dpo'
config = PeftConfig.from_pretrained(peft_model_name)
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)
config.base_model_name_or_path =model
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, quantization_config=bnb_config, device_map="auto")
model = PeftModel.from_pretrained(model, peft_model_name)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
streamer = TextStreamer(tokenizer)

def gen(x):
    generation_config = GenerationConfig(
        temperature=0.8,
        top_p=0.8,
        top_k=100,
        max_new_tokens=1024,
        early_stopping=True,
        do_sample=True,
    )
    q = f"[INST]{x} [/INST]"
    gened = model.generate(
        **tokenizer(
            q,
            return_tensors='pt',
            return_token_type_ids=False
        ).to('cuda'),
        generation_config=generation_config,
        pad_token_id=tokenizer.eos_token_id,
        eos_token_id=tokenizer.eos_token_id,
        streamer=streamer,
    )
    result_str = tokenizer.decode(gened[0])

    start_tag = f"[/INST]"
    start_index = result_str.find(start_tag)

    if start_index != -1:
        result_str = result_str[start_index + len(start_tag):].strip()
    return result_str

result = gen('μ œμ£Όλ„λ₯Ό 1λ°•2일둜 혼자 μ—¬ν–‰ν•˜λ €κ³  ν•˜λŠ”λ° μ—¬ν–‰ μ½”μŠ€λ₯Ό λ§Œλ“€μ–΄μ€˜')

print('##########')
print(result)

output

<s> [INST]μ œμ£Όλ„λ₯Ό 1λ°•2일둜 혼자 μ—¬ν–‰ν•˜λ €κ³  ν•˜λŠ”λ° μ—¬ν–‰ μ½”μŠ€λ₯Ό λ§Œλ“€μ–΄μ€˜ [/INST]μ œμ£Όλ„μ˜ μ•„λ¦„λ‹€μš΄ μžμ—° κ²½κ΄€κ³Ό λ¬Έν™”μœ μ‚°μ„ 느끼고 μ‹Άλ‹€λ©΄ λ‹€μŒκ³Ό 같은 μ½”μŠ€λ₯Ό μΆ”μ²œν•œλ‹€.

첫 번째 μ½”μŠ€λŠ” μ•„λ¦„λ‹€μš΄ ν•΄μ•ˆ κ²½κ΄€κ³Ό ν•΄μ•ˆμ‚¬κ΅¬λ₯Ό κ°μƒν•˜κΈ° μœ„ν•΄ 제주 μ„œλΆ€ ν•΄μ•ˆμœΌλ‘œ μ΄λ™ν•˜λŠ” 것이닀. 제주 μ‹œλ‚΄μ—μ„œ μ™Όμͺ½ λ°©ν–₯으둜 νƒλ‚˜λ©΄ ν•œλ¦Όν•΄μˆ˜μš•μž₯, μ„±μ‚°ν•΄μˆ˜μš•μž₯, λ΄‰λ™ν•΄μˆ˜μš•μž₯ λ“± 유λͺ…ν•œ ν•΄μˆ˜μš•μž₯을 κ²½μœ ν•  수 μžˆλ‹€. 이 지역은 맑은 바닀와 넓은 ν•΄μ•ˆμ—μ„œ ν•΄μˆ˜μš•μ„ 즐길 수 있으며, ν•΄μˆ˜μš•μž₯ μ£Όλ³€μ—λŠ” λ§Žμ€ μŒμ‹μ μ΄ μžˆμ–΄ 배식을 즐길 수 μžˆλ‹€. μ„œμͺ½ ν•΄μ•ˆμœΌλ‘œ μ΄λ™ν•˜λŠ” λ™μ•ˆ 제주 λŒ€ν‘œ μ‚¬κ³„μ ˆ 맛집인 ν—ˆλΈŒ μˆ˜ν”„ 및 μ†ŒλΌλΉ„ λ“± λ§›μžˆλŠ” μŒμ‹μ„ 맛볼 수 μžˆλ‹€. μ„œλΆ€ ν•΄μ•ˆμ„ λŒμ•„ λ‹€μ‹œ 제주 μ‹œλ‚΄λ‘œ λŒμ•„μ˜€λŠ” λ™μ•ˆ 제주 νŠΉμ‚°ν’ˆ μ‹œμž₯μ—μ„œ 제주 νŠΉμ‚°ν’ˆμ„ μ‚΄ 수 μžˆλ‹€.

두 번째 μ½”μŠ€λŠ” 동뢀 ν•΄μ•ˆμ„ λŒμ•„λ³΄λŠ” 것이닀. 제주 μ‹œλ‚΄μ—μ„œ 였λ₯Έμͺ½ λ°©ν–₯으둜 νƒλ‚˜λ©΄ μ•„μ΄μŠ€ν¬λ¦Ό 거리인 ν•œλ¦Όν•΄μˆ˜μš•μž₯, μ„±μ‚°ν•΄μˆ˜μš•μž₯, λ΄‰λ™ν•΄μˆ˜μš•μž₯ λ“± λ‹€μ‹œ ν•œ 번 유λͺ…ν•œ ν•΄μˆ˜μš•μž₯을 κ²½μœ ν•  수 μžˆλ‹€. 이 지역은 ν•΄μˆ˜μš•μž₯ μ£Όλ³€μ—λŠ” λ§Žμ€ μŒμ‹μ μ΄ μžˆμ–΄ 배식을 즐길 수 μžˆλ‹€. 동뢀 ν•΄μ•ˆμ„ λŒμ•„ λ‹€μ‹œ 제주 μ‹œλ‚΄λ‘œ λŒμ•„μ˜€λŠ” λ™μ•ˆ 제주 νŠΉμ‚°ν’ˆ μ‹œμž₯μ—μ„œ 제주 νŠΉμ‚°ν’ˆμ„ μ‚΄ 수 μžˆλ‹€. 이 μ§€μ—­μ—λŠ” λ§Žμ€ μŒμ‹μ μ΄ μžˆμ–΄ λ§›μžˆλŠ” μŒμ‹μ„ 맛볼 수 μžˆλ‹€.

μ„Έ 번째 μ½”μŠ€λŠ” 제주 λ‚¨λΆ€λ‘œ μ΄λ™ν•˜λŠ” 것이닀. 제주 μ‹œλ‚΄μ—μ„œ 였λ₯Έμͺ½ λ°©ν–₯으둜 νƒλ‚˜λ©΄ 제주 λ‚¨λΆ€λ‘œ 이동할 수 μžˆλ‹€. 이 지역은 ν•œλΌμ‚° ꡭ립곡원이 μœ„μΉ˜ν•΄ μžˆμ–΄ μžμ—° 경관을 감상할 수 μžˆλ‹€. ν•œλΌμ‚° ꡭ립곡원 λ‚΄μ—λŠ” λ‹€μ–‘ν•œ μžμ—° κ²½κ΄€κ³Ό μ‚°μ•… 경둜λ₯Ό 즐길 수 μžˆλŠ” 탐방 μ½”μŠ€κ°€ μžˆλ‹€. λ˜ν•œ, 제주 λ‚¨λΆ€λŠ” λ§Žμ€ ν•΄μˆ˜μš•μž₯κ³Ό 골프μž₯이 μœ„μΉ˜ν•΄ μžˆμ–΄ ν•΄μˆ˜μš•κ³Ό 골프λ₯Ό 즐길 수 μžˆλ‹€. λ‚¨λΆ€λ‘œ μ΄λ™ν•˜λŠ” λ™μ•ˆ 제주 νŠΉμ‚°ν’ˆ μ‹œμž₯μ—μ„œ 제주 νŠΉμ‚°ν’ˆμ„ μ‚΄ 수 μžˆλ‹€.

Evaluation

For objective model evaluation, we initially used EleutherAI's lm-evaluation-harness but obtained unsatisfactory results. Consequently, we conducted evaluations using ChatGPT, a widely used model, as described in Self-Alignment with Instruction Backtranslation and Three Ways of Using Large Language Models to Evaluate Chat .

model score average(0~5) percentage
gpt-3.5-turbo(close) 147 3.97 79.45%
naver Cue(close) 140 3.78 75.67%
clova X(close) 136 3.67 73.51%
WizardLM-13B-V1.2(open) 96 2.59 51.89%
Llama-2-7b-chat-hf(open) 67 1.81 36.21%
Llama-2-13b-chat-hf(open) 73 1.91 38.37%
nlpai-lab/kullm-polyglot-12.8b-v2(open) 70 1.89 37.83%
kfkas/Llama-2-ko-7b-Chat(open) 96 2.59 51.89%
beomi/KoAlpaca-Polyglot-12.8B(open) 100 2.70 54.05%
komt-llama2-7b-v1 (open)(ours) 117 3.16 63.24%
komt-llama2-13b-v1 (open)(ours) 129 3.48 69.72%
komt-llama-30b-v1 (open)(ours) 129 3.16 63.24%
komt-mistral-7b-v1 (open)(ours) 131 3.54 70.81%
komt-mistral-7b-v1-dpo (open)(ours) 142 3.83 76.75%