Fine-tuning Commands

Basic Usage

ideaweaver finetune [OPTIONS] COMMAND [ARGS]

Available Methods

• full: Full parameter fine-tuning
• lora: LoRA fine-tuning
• qlora: QLoRA fine-tuning

Full Fine-tuning Example

ideaweaver finetune full \
  --model microsoft/DialoGPT-small \
  --dataset datasets/instruction_following_sample.json \
  --output-dir ./test_full_basic \
  --epochs 5 \
  --batch-size 2 \
  --gradient-accumulation-steps 2 \
  --learning-rate 5e-5 \
  --max-seq-length 256 \
  --gradient-checkpointing \
  --verbose

Common Options

Option	Description	Default
--model	Base model name or path	Required
--dataset	Path to dataset file	Required
--output-dir	Directory to save model	Required
--epochs	Number of training epochs	3
--batch-size	Training batch size	8
--learning-rate	Learning rate	2e-5
--max-seq-length	Maximum sequence length	512
--gradient-checkpointing	Enable gradient checkpointing	False
--gradient-accumulation-steps	Gradient accumulation steps	1
--verbose	Enable verbose output	False

LoRA Fine-tuning Example

ideaweaver finetune lora \
  --model microsoft/DialoGPT-small \
  --dataset datasets/instruction_following_sample.json \
  --output-dir ./test_lora \
  --epochs 3 \
  --batch-size 4 \
  --learning-rate 1e-4 \
  --lora-rank 8 \
  --lora-alpha 16 \
  --verbose

QLoRA Fine-tuning Example

ideaweaver finetune qlora \
  --model microsoft/DialoGPT-small \
  --dataset datasets/instruction_following_sample.json \
  --output-dir ./test_qlora \
  --epochs 3 \
  --batch-size 4 \
  --learning-rate 1e-4 \
  --lora-rank 8 \
  --lora-alpha 16 \
  --bits 4 \
  --verbose

Additional Options

Experiment Tracking

# With MLflow
ideaweaver finetune full \
  --model microsoft/DialoGPT-small \
  --dataset datasets/instruction_following_sample.json \
  --output-dir ./test_full_basic \
  --track-experiments \
  --mlflow-uri http://127.0.0.1:5000 \
  --mlflow-experiment "MyExperiment"

# With Weights & Biases
ideaweaver finetune full \
  --model microsoft/DialoGPT-small \
  --dataset datasets/instruction_following_sample.json \
  --output-dir ./test_full_basic \
  --track-experiments \
  --wandb-project "MyProject" \
  --wandb-entity "MyEntity"

For more examples and use cases, see the Examples page.