# Model defaults for unsloth/Pixtral-12B-2409
# Based on Pixtral_(12B)-Vision.ipynb
# Also applies to: unsloth/Pixtral-12B-2409-unsloth-bnb-4bit, mistralai/Pixtral-12B-2409, unsloth/Pixtral-12B-2409-bnb-4bit
# added inference parameters from unsloth notebook

training:
  trust_remote_code: false
  max_seq_length: 2048
  # num_epochs: 4
  num_epochs: 0
  learning_rate: 2e-4
  batch_size: 1
  gradient_accumulation_steps: 4
  warmup_steps: 5
  max_steps: 30
  save_steps: 30
  weight_decay: 0.001
  random_seed: 3407
  packing: false
  train_on_completions: true
  gradient_checkpointing: "unsloth"
  optim: "paged_adamw_8bit"
  lr_scheduler_type: "linear"

lora:
  lora_r: 8
  lora_alpha: 8
  lora_dropout: 0.0
  target_modules:
    - "all-linear"
  use_rslora: false
  use_loftq: false
  finetune_vision_layers: true
  finetune_language_layers: true
  finetune_attention_modules: false
  finetune_mlp_modules: true

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

inference:
  trust_remote_code: false
  temperature: 1.5
  min_p: 0.1