model_name: "Tencent-Hunyuan/Hunyuan-A13B-Instruct"
model_type: "vllm"
hardware: "Atlas A2 Series"

serve:
  tensor_parallel_size: 4
  dtype: auto
  max_model_len: 32768
  gpu_memory_utilization: 0.90
  enforce_eager: true
  trust_remote_code: true

tasks:
  - name: "gsm8k"
    metrics:
      - name: "exact_match,strict-match"
        value: 0.37
      - name: "exact_match,flexible-extract"
        value: 0.28

num_fewshot: 5
limit: 1000
apply_chat_template: false
fewshot_as_multiturn: false
batch_size: "auto"