model:
  target_model_path: "meta-llama/Llama-3.1-8B-Instruct"
  draft_model_config: "configs/llama3-8B-eagle3.json"
  target_backend: "sglang"
  embedding_key: "model.embed_tokens.weight"
  vocab_mapping_path: "./cache/vocab_mapping/llama3.1-8b-eagle3.pt"
  torch_dtype: "bfloat16"
data:
  hidden_states_path: "./cache/hidden_states/sharegpt_train_Llama-3.1-8B-Instruct"
  max_length: 4096
  chat_template: "llama3"
  cache_dir: "./cache"
training:
  strategy: "eagle3"
  num_epochs: 10
  batch_size: 1
  learning_rate: 0.0001
  max_grad_norm: 0.5
  ttt_length: 7
  attention_backend: "sdpa"
  save_interval: 1000
  log_interval: 50
  dist_timeout: 20
  seed: 0
run_id: "llama3.1-8b-eagle3-offline"
output_dir: "./outputs/llama3.1-8b-eagle3-offline"

deployment:
  mode: local_colocated
  trainer:
    nnodes: 1
    nproc_per_node: 1