big_fut_final / README.md
mintaeng's picture
Update README.md
0c7e800 verified
metadata
library_name: transformers
tags:
  - unsloth
  - trl
  - sft
datasets:
  - mintaeng/llm_futsaldata_yo
license: apache-2.0
language:
  - ko

FUT FUT CHAT BOT

  • μ˜€ν”ˆμ†ŒμŠ€ λͺ¨λΈμ— LLM fine tuning κ³Ό RAG λ₯Ό μ μš©ν•œ μƒμ„±ν˜• AI
  • 풋살에 λŒ€ν•œ 관심이 λ†’μ•„μ§€λ©΄μ„œ μˆ˜μš” λŒ€λΉ„ μž…λ¬Έμžλ₯Ό μœ„ν•œ 정보 제곡 μ„œλΉ„μŠ€κ°€ ν•„μš”ν•˜λ‹€κ³  느껴 μ œμž‘ν•˜κ²Œ 됨
  • ν’‹μ‚΄ ν”Œλž«νΌμ— μ‚¬μš©λ˜λŠ” ν’‹μ‚΄ 정보 λ„μš°λ―Έ 챗봇
  • 'ν•΄μš”μ²΄'둜 λ‹΅ν•˜λ©° λ¬Έμž₯ 끝에 'μ–Όλ§ˆλ“ μ§€ λ¬Όμ–΄λ³΄μ„Έμš”~ ν’‹ν’‹~!' 을 좜λ ₯함
  • train for 7h23m

HOW TO USE

  
  #!pip install transformers==4.40.0 accelerate
  import os
  import torch
  from transformers import AutoTokenizer, AutoModelForCausalLM
  
  model_id = 'Dongwookss/small_fut_final'
  
  tokenizer = AutoTokenizer.from_pretrained(model_id)
  model = AutoModelForCausalLM.from_pretrained(
      model_id,
      torch_dtype=torch.bfloat16,
      device_map="auto",
  )
  model.eval()

Query

from transformers import TextStreamer
PROMPT = '''Below is an instruction that describes a task. Write a response that appropriately completes the request.
μ œμ‹œν•˜λŠ” contextμ—μ„œλ§Œ λŒ€λ‹΅ν•˜κ³  context에 μ—†λŠ” λ‚΄μš©μ€ λͺ¨λ₯΄κ² λ‹€κ³  λŒ€λ‹΅ν•΄'''

messages = [
    {"role": "system", "content": f"{PROMPT}"},
    {"role": "user", "content": f"{instruction}"}
    ]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

text_streamer = TextStreamer(tokenizer)
_ = model.generate(
    input_ids,
    max_new_tokens=4096,
    eos_token_id=terminators,
    do_sample=True,
    streamer = text_streamer,
    temperature=0.6,
    top_p=0.9,
    repetition_penalty = 1.1
)

Model Details

Model Description

This is the model card of a πŸ€— transformers model that has been pushed on the Hub. This model card has been automatically generated.

  • Developed by: Dongwookss
  • Model type: text generation
  • Language(s) (NLP): Korean
  • Finetuned from model : HuggingFaceH4/zephyr-7b-beta

Data

https://huggingface.co/datasets/mintaeng/llm_futsaldata_yo

ν•™μŠ΅ 데이터셋은 beomi/KoAlpaca-v1.1a λ₯Ό 베이슀둜 μΆ”κ°€, ꡬ좕, μ „μ²˜λ¦¬ μ§„ν–‰ν•œ 23.5k λ°μ΄ν„°λ‘œ νŠœλ‹ν•˜μ˜€μŠ΅λ‹ˆλ‹€. 데이터셋은 instruction, input, output 으둜 κ΅¬μ„±λ˜μ–΄ 있으며 tuning λͺ©ν‘œμ— 맞게 말투 μˆ˜μ •ν•˜μ˜€μŠ΅λ‹ˆλ‹€. 도메인 정보에 λŒ€ν•œ 데이터 μΆ”κ°€ν•˜μ˜€μŠ΅λ‹ˆλ‹€.

Training & Result

Training Procedure

LoRA와 SFT Trainer 방식을 μ‚¬μš©ν•˜μ˜€μŠ΅λ‹ˆλ‹€.

Training Hyperparameters

  • Training regime: bf16 mixed precision
  r=32,
  lora_alpha=64,  #  QLoRA : alpha = r/2 // LoRA : alpha =r*2
  lora_dropout=0.05, 
  target_modules=[
  "q_proj",
  "k_proj",
  "v_proj",
  "o_proj",
  "gate_proj",
  "up_proj",
  "down_proj",
  ],  # νƒ€κ²Ÿ λͺ¨λ“ˆ

Result

https://github.com/lucide99/Chatbot_FutFut

Environment

L4 GPU