model:
  hf_id: Qwen/Qwen3-1.7B
  local_path: /shared-ro/models/Qwen3-1.7B
  dtype: bfloat16
  attn_impl: flash_attention_2
  max_model_len: 4096
data:
  name: safety_aug302
  loader: load_safetybench_distilled
  kwargs:
    cache: /scratch/sft_aug302.jsonl
    limit: null
training:
  recipe: sft
  epochs: 5
  per_device_train_batch_size: 1
  gradient_accumulation_steps: 16
  learning_rate: 5.0e-06
  warmup_ratio: 0.05
  lr_scheduler_type: cosine
  gradient_checkpointing: true
  bf16: true
  max_seq_length: 4096
  logging_steps: 10
  save_steps: 500
  save_total_limit: 2
  report_to: wandb
  deepspeed: configs/deepspeed_zero2.json
  use_peft: true
  lora_r: 16
  lora_alpha: 32
  lora_target_modules: q_proj,v_proj,k_proj,o_proj
  lora_dropout: 0.05
domain:
  name: safety
  thinking_mode: true
  benchmark: safety
seed: 42
run_id: ${now:%Y%m%d_%H%M%S}
output_dir: /scratch/checkpoints/safety/sft/20260521_aug302
wandb:
  project: flab-cs552
  entity: cs-552-2026-flab
  mode: null
  tags: []
  notes: null
eval:
  model_dir: null
  'n': 8
  temperature: 0.7
  top_p: 0.95
  max_tokens: 4096
  samples_dir: null
push:
  model_dir: null
es:
  base_model_dir: null
  data: null
hf_push:
  enabled: false
  org: cs-552-2026-flab
  repo: null
  private: false
  vllm_validate: true
  commit_message: null
validate_after_train: false
vllm_validate: true