mini_lora.yaml

# Config for multi-device LoRA finetuning in lora_finetune_distributed.py
# using a Phi3 mini (3.8B) model
#
# This config assumes that you've run the following command before launching
# this run:
#   tune download microsoft/Phi-3-mini-4k-instruct --output-dir /tmp/Phi-3-mini-4k-instruct --ignore-patterns None --hf-token <HF_TOKEN>
#
# To launch on 2 devices, run the following command from root:
#   tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config phi3/mini_lora
#
# You can add specific overrides through the command line. For example
# to override the checkpointer directory while launching training
# you can run:
#   tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config phi3/mini_lora checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
#
# This config works best when the model is being fine-tuned on 2+ GPUs.
# For single device LoRA finetuning please use mini_lora_single_device.yaml
# or mini_qlora_single_device.yaml

# Model arguments
model:
  _component_: torchtune.models.phi3.lora_phi3_mini
  lora_attn_modules: ['q_proj', 'v_proj']
  apply_lora_to_mlp: False
  apply_lora_to_output: False
  lora_rank: 8
  lora_alpha: 16
  lora_dropout: 0.0

# Tokenizer
tokenizer:
  _component_: torchtune.models.phi3.phi3_mini_tokenizer
  path: /tmp/Phi-3-mini-4k-instruct/tokenizer.model
  max_seq_len: null

# Checkpointer
checkpointer:
  _component_: torchtune.training.FullModelHFCheckpointer
  checkpoint_dir: /tmp/Phi-3-mini-4k-instruct
  checkpoint_files: [
    model-00001-of-00002.safetensors,
    model-00002-of-00002.safetensors
  ]
  recipe_checkpoint: null
  output_dir: /tmp/Phi-3-mini-4k-instruct
  model_type: PHI3_MINI
resume_from_checkpoint: False
save_adapter_weights_only: False

# Dataset
dataset:
  _component_: torchtune.datasets.alpaca_cleaned_dataset
seed: null
shuffle: True

# Fine-tuning arguments
epochs: 1
max_steps_per_epoch: null
batch_size: 2
gradient_accumulation_steps: 16
optimizer:
  _component_: torch.optim.AdamW
  fused: True
  weight_decay: 0.01
  lr: 3e-4
lr_scheduler:
  _component_: torchtune.modules.get_cosine_schedule_with_warmup
  num_warmup_steps: 100
loss:
  _component_: torchtune.modules.loss.CEWithChunkedOutputLoss

# Training env
device: cuda

# Memory management
enable_activation_checkpointing: False
dtype: bf16

# Logging
output_dir: /tmp/phi3_lora_finetune_output
metric_logger:
  _component_: torchtune.training.metric_logging.DiskLogger
  log_dir: /tmp/Phi-3-mini-4k-instruct/logs
log_every_n_steps: 1
log_peak_memory_stats: False