eval_backend: VLMEvalKit
work_dir: outputs
eval_config:
  model:
    - type: internvl2-8b   # model id of the model
      name: CustomAPIModel # Don't change, must be CustomAPIModel for deploy evaluation
      api_base: http://localhost:8801/v1/chat/completions # deployed model api
      key: EMPTY
      temperature: 0.0
      img_size: 224
      img_detail: low
      video_llm: false
      system_prompt: "xxx"
  data:
    # - MVBench_MP4
    # - MLVU
    # - Video-MME
    # - MMBench-Video
    # - custom_vqa
    # - MMStar
    # - COCO_VAL
    # - AI2D_TEST
    # - POPE
    # - RealWorldQA
    # - SEEDBench2_Plus
    - MME
  mode: all
  limit: 2
  reuse: true
  nproc: 1
  judge: exact_matching
  # judge model server config
  # OPENAI_API_KEY: EMPTY
  # OPENAI_API_BASE: http://localhost:11434/v1/chat/completions # judge model api
  # LOCAL_LLM: llama3.1:latest                              # judge model type