example_training_config_for_finetuning_twllm.yaml

base_model: yentinglin/Llama-3-Taiwan-70B-Instruct
trust_remote_code: true
load_in_8bit: false
load_in_4bit: false
strict: false

hub_model_id: YOUR_USER_NAME/Llama-3-Taiwan-70B-Instruct-MY_FINETUNING
hub_strategy: end
wandb_name: twllm 70b sft

dataset_processes: 16
datasets:
  - path: yentinglin/TaiwanChat
    type: sharegpt
    conversation: llama3
chat_template: llama3

dataset_prepared_path: last_run_prepared
val_set_size: 0
output_dir: ./output/tw-llm/70b/sft/

save_safetensors: true
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project: twllm
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 5e-5

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: 100
evals_per_epoch: 0
eval_table_size:
saves_per_epoch: 1
save_total_limit: 10
save_steps:
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.05
fsdp:
fsdp_config: