model_name: "LLM-Research/Llama-3.2-3B-Instruct"
model_type: "vllm"
hardware: "Atlas A2 Series"

serve:
  tensor_parallel_size: 1
  dtype: auto
  max_model_len: 4096
  gpu_memory_utilization: 0.8
  trust_remote_code: false

tasks:
- name: "gsm8k"
  metrics:
  - name: "exact_match,strict-match"
    value: 0.71
  - name: "exact_match,flexible-extract"
    value: 0.76

num_fewshot: 5
apply_chat_template: true
fewshot_as_multiturn: true
batch_size: "auto"