MLP-KTLim/llama-3-Korean-Bllossom-8B model fine tuning

(TREX-Lab at Seoul Cyber University)

Summary

  • Base Model : MLP-KTLim/llama-3-Korean-Bllossom-8B
  • Dataset : heegyu/open-korean-instructions (10%)
  • Tuning Method
    • PEFT(Parameter Efficient Fine-Tuning)
    • LoRA(Low-Rank Adaptation of Large Language Models)
  • Related Articles : https://arxiv.org/abs/2106.09685, https://arxiv.org/pdf/2403.10882
  • Fine-tuning the Base Model with a random 10% of Korean chatbot data (open Korean instructions)
  • Test whether fine tuning of a large language model is possible on A30 GPU*1 (successful)
  • Developed by: [TREX-Lab at Seoul Cyber University]
  • Language(s) (NLP): [Korean]
  • Finetuned from model : [MLP-KTLim/llama-3-Korean-Bllossom-8B]

Fine Tuning Detail

  • alpha value 16
  • r value 64 (it seems a bit big...@@)
peft_config = LoraConfig(
    lora_alpha=16,
    lora_dropout=0.1,
    r=64,
    bias='none',
    task_type='CAUSAL_LM'
)
  • Mixed precision : 4bit (bnb_4bit_use_double_quant)
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type='nf4',
    bnb_4bit_compute_dtype='float16',
)
trainer = SFTTrainer(
    model=peft_model,
    train_dataset=dataset,
    dataset_text_field='text',
    max_seq_length=min(tokenizer.model_max_length, 2048),
    tokenizer=tokenizer,
    packing=True,
    args=training_args
)

Train Result

time taken : executed in 21h 45m 55s
TrainOutput(global_step=816, training_loss=1.718194248045192,
            metrics={'train_runtime': 78354.6002,
                     'train_samples_per_second': 0.083,
                     'train_steps_per_second': 0.01,
                     'train_loss': 1.718194248045192,
                     'epoch': 2.99})
Downloads last month
23
Safetensors
Model size
8.03B params
Tensor type
FP16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train LEESM/llama-3-Korean-Bllossom-8B-trexlab-oki10p