fft-ds-zero3.yaml

base_model: LnL-AI/dbrx-base-converted-v2
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

trust_remote_code: true

datasets:
  - path: tatsu-lab/alpaca
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./outputs/out

sequence_len: 512
sample_packing: false
pad_to_sequence_len: false

unfrozen_parameters:
  - transformer.blocks.[0-7].

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002

bf16: auto
tf32: false

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

warmup_steps: 10
evals_per_epoch:
saves_per_epoch: 1

weight_decay: 0.0
deepspeed: deepspeed_configs/zero3_bf16.json