# Model defaults for unsloth/functiongemma-270m-it
# Based on FunctionGemma_(270M).ipynb
# Also applies to: unsloth/functiongemma-270m-it-unsloth-bnb-4bit, google/functiongemma-270m-it, unsloth/functiongemma-270m-it-unsloth-bnb-4bit
# added inference parameters from unsloth guides

training:
  trust_remote_code: false
  max_seq_length: 4096
  # num_epochs: 4
  num_epochs: 0
  learning_rate: 2e-4
  batch_size: 4
  gradient_accumulation_steps: 2
  warmup_steps: 10
  max_steps: 30
  save_steps: 30
  weight_decay: 0.001
  random_seed: 3407
  packing: false
  train_on_completions: true
  gradient_checkpointing: "unsloth"
  optim: "adamw_8bit"
  lr_scheduler_type: "linear"

lora:
  lora_r: 128
  lora_alpha: 256
  lora_dropout: 0.0
  target_modules:
    - "q_proj"
    - "k_proj"
    - "v_proj"
    - "o_proj"
    - "gate_proj"
    - "up_proj"
    - "down_proj"
  use_rslora: false
  use_loftq: false

logging:
  enable_wandb: false
  wandb_project: "llm-finetuning"
  enable_tensorboard: false
  tensorboard_dir: "runs"
  log_frequency: 10

inference:
  trust_remote_code: false
  temperature: 1.0
  top_k: 64
  top_p: 0.95
  min_p: 0.0