0.5B_full_single_device.yaml

# Config for single device full finetuning in full_finetune_single_device.py
# using a Qwen2 0.5B
#
# This config assumes that you've run the following command before launching
# this run:
#   tune download Qwen/Qwen2-0.5B-Instruct --output-dir /tmp/Qwen2-0.5B-Instruct --ignore-patterns None
#
# To launch on a single device, run the following command from root:
#   tune run full_finetune_single_device --config qwen2/0.5B_full_single_device
#
# You can add specific overrides through the command line. For example
# to override the checkpointer directory while launching training
# you can run:
#   tune run full_finetune_single_device --config qwen2/0.5B_full_single_device checkpointer.checkpoint_dir=<YOUR_CHECKPOINT_DIR>
#
# This config works only for training on single device.

# Tokenizer
tokenizer:
  _component_: torchtune.models.qwen2.qwen2_tokenizer
  path: /tmp/Qwen2-0.5B-Instruct/vocab.json
  merges_file: /tmp/Qwen2-0.5B-Instruct/merges.txt

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

# Model Arguments
model:
  _component_: torchtune.models.qwen2.qwen2_0_5b

checkpointer:
  _component_: torchtune.utils.FullModelHFCheckpointer
  checkpoint_dir: /tmp/Qwen2-0.5B-Instruct
  checkpoint_files: [
    model.safetensors
  ]
  recipe_checkpoint: null
  output_dir: /tmp/Qwen2-0.5B-Instruct-finetune
  model_type: QWEN2
resume_from_checkpoint: False

# Fine-tuning arguments
batch_size: 2
epochs: 1
optimizer:
  _component_: torch.optim.AdamW
  lr: 2e-5

loss:
  _component_: torch.nn.CrossEntropyLoss
optimizer_in_bwd: False

max_steps_per_epoch: null
gradient_accumulation_steps: 8
compile: False

# Training environment
device: cuda

# Memory management
enable_activation_checkpointing: True

# Reduced precision
dtype: bf16

# Logging
metric_logger:
  _component_: torchtune.utils.metric_logging.DiskLogger
  log_dir: ${output_dir}
output_dir: /tmp/Qwen2-0.5B-Instruct-finetune
log_every_n_steps: 1
log_peak_memory_stats: False