run:
  name: qwen35-exp2-canary-smoke
  output_root: data/runs/eval_suite
runner: auto
runners:
  h100:
    tensor_parallel: 1
    gpu_memory_utilization: 0.9
    max_num_seqs: 64
    port: 8000
  h200:
    tensor_parallel: 1
    gpu_memory_utilization: 0.92
    max_num_seqs: 128
    port: 8000
  two_h100:
    tensor_parallel: 2
    gpu_memory_utilization: 0.9
    max_num_seqs: 96
    port: 8000
  auto:
    tensor_parallel: auto
    gpu_memory_utilization: 0.92
    max_num_seqs: 128
    port: 8000
model:
  id: Qwen/Qwen3.5-9B
  path: null
  substrate: exp2
  max_model_len: 32768
  max_lora_rank: 128
  max_loras: auto
  reasoning_parser: qwen3
  enable_thinking: true
  chat_template: auto
judge: anthropic/claude-sonnet-4-6
max_connections: 64
keep_serving: false
arms: []
suites:
  interface_canary:
    type: inspect
    tasks:
    - name: interface_canary
      task: why_gen/inspect_tasks/interface_canary.py@interface_canary
      limit: 1
      temperature: 0.0
      max_tokens: 512