model:
  target_model_path: Qwen/Qwen3-8B
  draft_model_config: configs/qwen3-8b-dflash.json
  target_backend: sglang
data:
  train_data_path: ./cache/dataset/sharegpt_train.jsonl
  max_length: 3072
  chat_template: qwen
  cache_dir: cache
  build_dataset_num_proc: 32
training:
  strategy: dflash
  num_epochs: 6
  max_steps: 10000
  batch_size: 4
  learning_rate: 0.0006
  warmup_ratio: 0.04
  max_grad_norm: 1.0
  num_anchors: 512
  loss_decay_gamma: 7.0
  save_interval: 1000
  dist_timeout: 30
  seed: 42
tracking:
  report_to: none
  wandb_project: qwen3-8b-dflash-disagg
  wandb_name: qwen3-8b-dflash-disagg-dp4
run_id: qwen3-8b-dflash-disaggregated
output_dir: outputs/qwen3-8b-dflash-disaggregated

deployment:
  mode: disaggregated
  trainer:
    nnodes: 1
    nproc_per_node: 4
  disaggregated:
    control_dir: outputs/qwen3-8b-dflash-disaggregated/control
    consumer_state_dir: outputs/qwen3-8b-dflash-disaggregated/consumer-state
    backend: mooncake
    server_urls:
      - http://127.0.0.1:30000
    mooncake_metadata_server: http://127.0.0.1:35880/metadata
    mooncake_master_server_addr: 127.0.0.1:35551
    mooncake_protocol: tcp