MMEB-V3

MMEB-V3 is an omni-modality embedding benchmark for evaluating retrieval and representation models across text, image, video, audio, visual document, and agent-centric tasks. It extends MMEB-V1/V2 and provides a unified local layout for running the evaluation code in VLM2Vec.

Download

export MMEB_V3_ROOT=/path/to/MMEB-V3
hf download VLM2Vec/MMEB-V3 \
  --repo-type dataset \
  --local-dir $MMEB_V3_ROOT

Prepare Data

The uploaded files keep compressed raw assets under _tasks directories. Run the setup script in the VLM2Vec repo to materialize the evaluation-ready -tasks directories:

python experiments/public/data/dataset_setup_v3.py --root $MMEB_V3_ROOT
python experiments/public/data/dataset_setup_v3.py --root $MMEB_V3_ROOT --check-only

Raw archive layout:

MMEB-V3/
  image_tasks/
  audio_tasks/
  video_tasks/
  visdoc_tasks/
  gui_tasks/
  memory_tasks/
  text_tasks/
  tool_tasks/
  omniset.tar.gz

Expected evaluation-ready layout after setup:

MMEB-V3/
  image-tasks/
    MMEB/
    MCMR/
  image-query/
  audio-tasks/
  video-tasks/
    data/
    frames/
      video_cls/
      video_ret/
      video_mret/
      video_qa/
  visdoc-tasks/
    data/
    images/
  text-tasks/
  tool-tasks/
  memory-tasks/
  gui-tasks/
  omniset/
    omniset.jsonl
    catalog.jsonl
    val2014/
    videos/
    audios/
    frames_omni/

Evaluation

For standard MMEB-V3 tasks, pass the dataset root to --data_basedir:

CUDA_VISIBLE_DEVICES=0 python eval.py \
  --pooling mean \
  --normalize true \
  --per_device_eval_batch_size 8 \
  --dataloader_num_workers 1 \
  --model_backbone nvomniembed \
  --model_name /path/to/model \
  --dataset_config experiments/public/eval/image.yaml \
  --encode_output_path exps/vlm2vec/model/image \
  --data_basedir $MMEB_V3_ROOT

For OmniSET:

CUDA_VISIBLE_DEVICES=0 \
MODEL_PATH=/path/to/model \
MODEL_BACKBONE=nvomniembed \
DATA_BASEDIR=$MMEB_V3_ROOT/omniset \
OUTPUT_PATH=exps/vlm2vec/model/omniset \
PER_DEVICE_EVAL_BATCH_SIZE=8 \
bash experiments/public/eval/eval_omniset.sh