# ============================================================
# PatchHawk Training & Pipeline Configuration
# All tunable hyperparameters live here
# ============================================================

data_generation:
  num_samples: 10
  output_format: "json"
  benign_dir: "patchhawk/data/benign/"
  scenarios_output: "patchhawk/data/scenarios.json"
  sdk_config: "patchhawk/data/sdk_config.yaml"

training:
  learning_rate: 0.000001    # 1e-6
  group_size: 4              # GRPO group size
  max_seq_len: 1024
  max_steps: 100
  gradient_accumulation_steps: 4
  ppo_clip_eps: 0.2
  lora_r: 16
  lora_alpha: 16
  lora_dropout: 0
  output_dir: "grpo_lora"

environment:
  max_steps: 5
  use_docker: false
  sandbox_timeout_sec: 5

docker:
  image: "patchhawk-sandbox:latest"
  network: "none"
  memory: "256m"
  cpus: "0.5"

a2a:
  host: "0.0.0.0"
  port: 8000

huggingface:
  repo: "ramprasathk07/patchhawk"
  push_after_training: true