model_name: "meta-llama/Llama-3.2-1B"
dataset_path: "data/artisanal_training_sets/artisanal_data_snc_v2.jsonl"
output_dir: "experimental/slm_distillation/adapters/v1"
lora_config:
  r: 16
  lora_alpha: 32
  target_modules: ["q_proj", "v_proj"]
  lora_dropout: 0.05
training_args:
  per_device_train_batch_size: 4
  gradient_accumulation_steps: 4
  learning_rate: 2e-4
  num_train_epochs: 3
  logging_steps: 10
  save_strategy: "epoch"
distillation:
  teacher_model: "Qwen/Qwen2.5-72B-Instruct"
  temperature: 2.0
  alpha: 0.5  # Weight for distillation loss vs student loss