model:
  target_model_path: "meta-llama/Llama-3.1-8B-Instruct"
  draft_model_config: "configs/llama3-8B-eagle3.json"
  vocab_mapping_path: cache/vocab_mapping/llama3.1-8b-eagle3.pt
  target_backend: "sglang"
  embedding_key: "model.embed_tokens.weight"
  torch_dtype: "bfloat16"
  sglang_mem_fraction_static: 0.25
data:
  train_data_path: "./cache/dataset/sharegpt_train.jsonl"
  max_length: 4096
  chat_template: "llama3"
  build_dataset_num_proc: 64
  cache_dir: "./cache"
training:
  strategy: "eagle3"
  num_epochs: 10
  max_steps: 10000
  batch_size: 1
  learning_rate: 0.0001
  max_grad_norm: 0.5
  ttt_length: 7
  attention_backend: "sdpa"
  save_interval: 1000
  log_interval: 10
  dist_timeout: 20
  seed: 0
run_id: "llama3.1-8b-eagle3-online"
output_dir: "./outputs/llama3.1-8b-eagle3-online"

deployment:
  mode: disaggregated
  trainer:
    nnodes: 1
    nproc_per_node: 1
  disaggregated:
    control_dir: outputs/llama3.1-8b-eagle3-online/control
    consumer_state_dir: outputs/llama3.1-8b-eagle3-online/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