komt : korean multi task instruction tuning model
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
- Lora target modules : q_proj, o_proj, v_proj, gate_proj, down_proj, k_proj, up_proj
- Model Size : 84MB
- Model Architecture : The komt-mistral-7b-v1 is is a fine-tuned version of the Mistral-7B-Instruct-v0.1.
Dataset
korean multi-task instruction 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-lora'
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 μ£Όλλ₯Ό 1λ°2μΌλ‘ νΌμ μ¬ννλ €κ³ νλλ° μ¬ν μ½μ€λ₯Ό λ§λ€μ΄μ€')
print('##########')
print(result)
output
<s> [INST]μ μ£Όλλ₯Ό 1λ°2μΌλ‘ νΌμ μ¬ννλ €κ³ νλλ° μ¬ν μ½μ€λ₯Ό λ§λ€μ΄μ€ [/INST]μ μ£Όλμ μλ¦λ€μ΄ μμ° κ²½κ΄κ³Ό λ¬Ένμ μ°μ λλΌκ³ μΆλ€λ©΄ λ€μκ³Ό κ°μ μ½μ€λ₯Ό μΆμ²νλ€.
첫 λ²μ§Έ μ½μ€λ μλ¦λ€μ΄ ν΄μ κ²½κ΄κ³Ό ν΄μμ¬κ΅¬λ₯Ό κ°μνκΈ° μν΄ μ μ£Ό μλΆ ν΄μμΌλ‘ μ΄λνλ κ²μ΄λ€. μ μ£Ό μλ΄μμ μΌμͺ½ λ°©ν₯μΌλ‘ νλλ©΄ νλ¦Όν΄μμμ₯, μ±μ°ν΄μμμ₯, λ΄λν΄μμμ₯ λ± μ λͺ
ν ν΄μμμ₯μ κ²½μ ν μ μλ€. μ΄ μ§μμ λ§μ λ°λ€μ λμ ν΄μμμ ν΄μμμ μ¦κΈΈ μ μμΌλ©°, ν΄μμμ₯ μ£Όλ³μλ λ§μ μμμ μ΄ μμ΄ λ°°μμ μ¦κΈΈ μ μλ€. μμͺ½ ν΄μμΌλ‘ μ΄λνλ λμ μ μ£Ό λν μ¬κ³μ λ§μ§μΈ νλΈ μν λ° μλΌλΉ λ± λ§μλ μμμ λ§λ³Ό μ μλ€. μλΆ ν΄μμ λμ λ€μ μ μ£Ό μλ΄λ‘ λμμ€λ λμ μ μ£Ό νΉμ°ν μμ₯μμ μ μ£Ό νΉμ°νμ μ΄ μ μλ€.
λ λ²μ§Έ μ½μ€λ λλΆ ν΄μμ λμ보λ κ²μ΄λ€. μ μ£Ό μλ΄μμ μ€λ₯Έμͺ½ λ°©ν₯μΌλ‘ νλλ©΄ μμ΄μ€ν¬λ¦Ό κ±°λ¦¬μΈ νλ¦Όν΄μμμ₯, μ±μ°ν΄μμμ₯, λ΄λν΄μμμ₯ λ± λ€μ ν λ² μ λͺ
ν ν΄μμμ₯μ κ²½μ ν μ μλ€. μ΄ μ§μμ ν΄μμμ₯ μ£Όλ³μλ λ§μ μμμ μ΄ μμ΄ λ°°μμ μ¦κΈΈ μ μλ€. λλΆ ν΄μμ λμ λ€μ μ μ£Ό μλ΄λ‘ λμμ€λ λμ μ μ£Ό νΉμ°ν μμ₯μμ μ μ£Ό νΉμ°νμ μ΄ μ μλ€. μ΄ μ§μμλ λ§μ μμμ μ΄ μμ΄ λ§μλ μμμ λ§λ³Ό μ μλ€.
μΈ λ²μ§Έ μ½μ€λ μ μ£Ό λ¨λΆλ‘ μ΄λνλ κ²μ΄λ€. μ μ£Ό μλ΄μμ μ€λ₯Έμͺ½ λ°©ν₯μΌλ‘ νλλ©΄ μ μ£Ό λ¨λΆλ‘ μ΄λν μ μλ€. μ΄ μ§μμ νλΌμ° κ΅λ¦½κ³΅μμ΄ μμΉν΄ μμ΄ μμ° κ²½κ΄μ κ°μν μ μλ€. νλΌμ° κ΅λ¦½κ³΅μ λ΄μλ λ€μν μμ° κ²½κ΄κ³Ό μ°μ
κ²½λ‘λ₯Ό μ¦κΈΈ μ μλ νλ°© μ½μ€κ° μλ€. λν, μ μ£Ό λ¨λΆλ λ§μ ν΄μμμ₯κ³Ό 골νμ₯μ΄ μμΉν΄ μμ΄ ν΄μμκ³Ό 골νλ₯Ό μ¦κΈΈ μ μλ€. λ¨λΆλ‘ μ΄λνλ λμ μ μ£Ό νΉμ°ν μμ₯μμ μ μ£Ό νΉμ°νμ μ΄ μ μλ€.
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% |
- Downloads last month
- 2