base_model: /raid/HUB_LLM/080225_vi_test_llama33_70b_instruct/checkpoint-2568/dpo32_epoch2/
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

dataset_processes: 128
datasets:
  - path: json
    data_files:      
      - /raid/instruction_tuning_060225/data/data_to_train/0103_136k.jsonl
      
    type: chat_template
    field_messages: messages
    message_field_role: role
    message_field_content: content

val_set_size: 0.0
output_dir: /raid/HUB_LLM/010325_VAI_v27_70b_instruct

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

gradient_accumulation_steps: 16
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: cosine_with_min_lr
learning_rate: 3e-6
lr_scheduler_kwargs:
    min_lr: 0.0000007

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 0
evals_per_epoch: 1
eval_table_size:
saves_per_epoch: 2
save_total_limit: 10
debug:
deepspeed: /raid/instruction_tuning_060225/training/accelerate_config/zero3_bf16_cpuoffload_params.json
weight_decay: 0.1

max_grad_norm: 1.0
special_tokens:
  pad_token: <|end_of_text|>