model_name: "OpenGVLab/InternVL3_5-8B-hf"
model_type: "vllm-vlm"
hardware: "Atlas A2 Series"

serve:
  tensor_parallel_size: 1
  dtype: auto
  max_model_len: 40960
  gpu_memory_utilization: 0.8
  trust_remote_code: true

tasks:
  - name: "mmmu_val"
    metrics:
    - name: "acc,none"
      value: 0.58

num_fewshot: 0
apply_chat_template: false
fewshot_as_multiturn: false
batch_size: "auto"