log_dir: ${oc.env:ROOT,.}/logs

training:
  max_round: 1000000
  max_stage: 1
  hf_push_frequency: 1
  num_generations: 1
  num_transplant_trees: 1
  seed: 42
  fp16: false # set this line to true if your hardware supports fp16

blockchain:
  alchemy_url: "https://gensyn-testnet.g.alchemy.com/public"
  contract_address: ${oc.env:SWARM_CONTRACT,null} # This is set by modal-login in run_rl_swarm.sh
  org_id: ${oc.env:ORG_ID,null} # This is set by modal-login in run_rl_swarm.sh
  mainnet_chain_id: 685685 # currently unused, will be used with WalletSwarmCoordinator
  modal_proxy_url: "http://localhost:3000/api/"

communications:
  initial_peers:
    - '/ip4/38.101.215.15/tcp/30011/p2p/QmQ2gEXoPJg6iMBSUFWGzAabS2VhnzuS782Y637hGjfsRJ'
    - '/ip4/38.101.215.15/tcp/30012/p2p/QmWhiaLrx3HRZfgXc2i7KW5nMUNK7P9tRc71yFJdGEZKkC'
    - '/ip4/38.101.215.15/tcp/30013/p2p/QmQa1SCfYTxx7RvU7qJJRo79Zm1RAwPpkeLueDVJuBBmFp'

eval:
  judge_base_url: https://swarm-judge-102957787771.us-east1.run.app

hydra:
  run:
    dir: ${log_dir}

game_manager:
  _target_: genrl_swarm.game.game_manager.SwarmGameManager
  max_stage: ${training.max_stage}
  max_round: ${training.max_round}
  log_dir: ${log_dir}
  hf_token: ${oc.env:HUGGINGFACE_ACCESS_TOKEN,null}
  hf_push_frequency: ${training.hf_push_frequency}
  run_mode: "train_and_evaluate"
  bootnodes: ${communications.initial_peers}
  game_state:
    _target_: genrl_swarm.state.game_state.GameState
    round: 0
    stage: 0
  reward_manager:
    _target_: genrl_swarm.rewards.DefaultRewardManager
    reward_fn_store:
      _target_: genrl_swarm.rewards.reward_store.RewardFnStore
      max_rounds: ${training.max_round}
      reward_fn_stores:
        - _target_: genrl_swarm.rewards.reward_store.RoundRewardFnStore
          num_stages: ${training.max_stage}
          reward_fns:
            - _target_: genrl_swarm.examples.rgym.rewards.RGRewards
  trainer:
    _target_: genrl_swarm.examples.rgym.trainer.GRPOTrainerModule
    models:
      - _target_: transformers.AutoModelForCausalLM.from_pretrained
        pretrained_model_name_or_path: ${oc.env:MODEL_NAME, ${gpu_model_choice:${default_large_model_pool},${default_small_model_pool}}}
    config:
      _target_: trl.trainer.GRPOConfig
      logging_dir: ${log_dir}
      fp16: ${training.fp16}
    log_with: wandb
    log_dir: ${log_dir}
    epsilon: 0.2
    epsilon_high: 0.28
    num_generations: ${training.num_generations}
    judge_base_url: ${eval.judge_base_url}
  data_manager:
    _target_: genrl_swarm.examples.rgym.data.ReasoningGymDataManager
    yaml_config_path: "genrl-swarm/src/genrl_swarm/examples/rgym/datasets.yaml"
    num_train_samples: 1
    num_evaluation_samples: 0
    system_prompt_id: 'default'
    seed: ${training.seed}
    num_transplant_trees: ${training.num_transplant_trees}
  communication:
    _target_: genrl_swarm.communication.hivemind.hivemind_backend.HivemindBackend
    initial_peers: ${communications.initial_peers}
    identity_path: ${oc.env:IDENTITY_PATH,null}
  coordinator:
    _target_: genrl_swarm.blockchain.coordinator.ModalSwarmCoordinator
    web3_url: ${blockchain.alchemy_url}
    contract_address: ${blockchain.contract_address}
    org_id: ${blockchain.org_id}
    modal_proxy_url: ${blockchain.modal_proxy_url}

default_large_model_pool:
  - nvidia/AceInstruct-1.5B
  - dnotitia/Smoothie-Qwen3-1.7B
  - Gensyn/Qwen2.5-1.5B-Instruct

default_small_model_pool:
  - Gensyn/Qwen2.5-0.5B-Instruct
  - Qwen/Qwen3-0.6B