model_config:
  name: Qwen/Qwen3-32B
  num_mtp_tokens: 0
  quantize_linear_action: W8A8_DYNAMIC
  quantize_lmhead: False
  mxfp4_group_size: 32
  quantize_attention_action: DISABLED
  do_compile: False
  allow_graph_break: False
  dump_input_shapes: False
  chrome_trace: null
  enable_multi_process: False
  num_processes: 10
  predict_steps: 20
  enable_interpolate: True
  interpolation_seed: 1234
  enable_preprocessing_modeling: True
  enable_kv_transfer_modeling: True

load_gen:
  load_gen_type: fixed_length
  num_requests: 500
  num_input_tokens: 3500
  num_output_tokens: 1500
  request_rate: 2.0

serving_config:
  max_concurrency: 100
  block_size: 128
  max_tokens_budget: 8192