🚀 Project Page | 🤗 HF Dataset | 📜 arXiv | 🏆 Leaderboard | 📝 blog | 🤗 HF Paper | 𝕏 Twitter
MixEval-X encompasses eight input-output modality combinations and can be further extended. Its data points reflect real-world task distributions. The last grid presents the scores of frontier organizations’ flagship models on MixEval-X, normalized to a 0-100 scale, with MMG tasks using win rates instead of Elo. Section C of the paper presents example data samples and model responses.
[2024-10-20] MixEval-X is released! Checkout the Paper and Leaderboard to learn more about this real-world any-to-any benchmark!🌟
MixEval-X is the first any-to-any, real-world benchmark featuring diverse input-output modalities, real-world task distributions, consistent high standards across modalities, and dynamism. It achieves up to 0.98 correlation with arena-like multi-modal evaluations while being way more efficient.
See the project page and paper for more details.
This repo contains the grading code for MixEval-X. Once you have prepared your model outputs according to the required format, you will be able to get the final scores in just a few steps.
The MixEval-X data can be downloaded from the huggingface.
Feel free to use your own grading code, as long as it's fair.
(Step 1) Clone repo and setup the environment:
git clone https://github.com/Psycoy/MixEval-X.git
cd MixEval-X
conda create -n MixEval-X python=3.11 --yes
conda activate MixEval-X
bash setup.sh
# setup done
(Step 2) Setup the OpenAI API key for model parser. Create .env file under root dir (MixEval-X/) and add the below line to it:
MODEL_PARSER_API=<your openai api key>
The values in Leaderboard use
GPT-3.5-Turbo-0125as the default model parser for MMU tasks,gpt-4o-2024-08-06as the default model parser for agent tasks.
(Step 3) Prepare the model outputs as specified here on your own side, and use the below command to compute the results. That's all!
Image2Text
# Normal Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark image2text \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
# Hard Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark image2text_hard \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
Video2Text
# Normal Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark video2text \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
# Hard Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark video2text_hard \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
Audio2Text
# Normal Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark audio2text \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
# Hard Version
python -m mixeval_x.compute_metrics_mmu \
--benchmark audio2text_hard \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
Text2Action
python -m mixeval_x.compute_metrics_mmg_agent \
--benchmark text2action \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--judge_model "gpt-4o-2024-08-06" \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
Image2Action
python -m mixeval_x.compute_metrics_mmg_agent \
--benchmark image2action \
--judge_model "gpt-4o-2024-08-06" \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--image2action_image_dir THE_PATH_TO_IMAGE2ACTION_INPUT_IMAGES \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
Text2Image
python -m mixeval_x.compute_metrics_mmg_agent \
--benchmark text2image \
--judge_model "gpt-4o-2024-08-06" \
--model_response_dir THE_PATH_TO_MODEL_OUTPUT_FOLDER \
--models_to_eval \
gemini_1_5_pro \
gemini_1_5_flash
The MMG results (Text2Image, Text2Video, and Text2Audio) in Leaderboard were graded with Amazon Mechanical Turk workers. Text2Image also supports model parsing (See Section 4.2 of the paper). However, the Text2Video and Text2Audio lack capable model judges, and thus their grading are not implemented. You should hire human evaluators to grade these two subsets.
🥇 It extends all the benefits of MixEval to multi-modal evaluations, including comprehensive and less biased query distribution; fair grading (except open-ended tasks); dynamism; accurate model ranking; fast, cost-effective, reproducible execution; and challenging nature.
🥇 It establishes unified, high standards across modalities and communities. For single-modality models, it ensures its evaluation keeps up with the state-of-the-art standards; for multi-modality models, it ensures consistent, high-standard evaluations across modalities, preventing any from becoming a bottleneck.
🥇 Beyond model evaluation, MixEval-X benchmarks different organizations (as shown in the first Figure) with balanced dimensions (modalities), unlocking a new level of evaluation.
Feel free to hit the ⭐star button or 🦾contribute! We review new issues and PRs regularly and will acknowledge your contributions!
If you found this repository useful, please consider 📑citing:
@article{ni2024mixevalx,
title={MixEval-X: Any-to-Any Evaluations from Real-World Data Mixtures},
author={Ni, Jinjie and Song, Yifan and Ghosal, Deepanway and Li, Bo and Zhang, David Junhao and Yue, Xiang and Xue, Fuzhao and Zheng, Zian and Zhang, Kaichen and Shah, Mahir and Jain, Kabir and You, Yang and Shieh, Michael},
journal={arXiv preprint arXiv:2410.13754},
year={2024}
}
@article{ni2024mixeval,
title={MixEval: Deriving Wisdom of the Crowd from LLM Benchmark Mixtures},
author={Ni, Jinjie and Xue, Fuzhao and Yue, Xiang and Deng, Yuntian and Shah, Mahir and Jain, Kabir and Neubig, Graham and You, Yang},
journal={arXiv preprint arXiv:2406.06565},
year={2024}
}