minillama3_8b

Train Config

base_model: allganize/Llama-3-Alpha-Ko-8B-Instruct model_type: AutoModelForCausalLM tokenizer_type: AutoTokenizer

load_in_8bit: false load_in_4bit: true strict: false

datasets:

• path: ? type: alpaca dataset_prepared_path: val_set_size: 0 output_dir: ./outputs/qlora-out

adapter: qlora lora_model_dir:

sequence_len: 2048 sample_packing: true pad_to_sequence_len: true

lora_r: 32 lora_alpha: 16 lora_dropout: 0.05 lora_target_modules:

• q_proj
• v_proj

lora_target_linear: true lora_fan_in_fan_out: lora_modules_to_save:

• embed_tokens
• lm_head

wandb_project: wandb_entity: wandb_watch: wandb_name: wandb_log_model:

gradient_accumulation_steps: 4 micro_batch_size: 2 num_epochs: 3 optimizer: paged_adamw_32bit lr_scheduler: cosine learning_rate: 0.0002

train_on_inputs: false group_by_length: false bf16: auto fp16: tf32: false

gradient_checkpointing: true early_stopping_patience: resume_from_checkpoint: local_rank: logging_steps: 100 xformers_attention: flash_attention: true

warmup_steps: 10 evals_per_epoch: 4 eval_table_size: saves_per_epoch: 1 debug: deepspeed: weight_decay: 0.01 neftune_noise_alpha: 5 fsdp: fsdp_config: special_tokens: pad_token: :"<|end_of_text|>"