OmniEvalKit/omnievalkit-dataset

OmniEvalKit Evaluation Datasets

Evaluation datasets for OmniEvalKit, a comprehensive evaluation framework for omni-modal (audio + video + image + text) models.

Overview

• Total subsets: 65
• Total samples: 315,264
• Total size: 620.3 GB (Parquet with embedded audio/image/video)
• Subsets with embedded video: 15
• Subsets requiring external video download: 2

Usage

from datasets import load_dataset

ds = load_dataset("OmniEvalKit/omnievalkit-dataset", "aishell1_test")
for sample in ds["test"]:
    print(sample)

Available Subsets

Audio (44 subsets, 267,616 samples, 186.5GB)

Subset	Display Name	Samples	Size	Subcategory	Video
aishell1_test	AISHELL-1 Test	7,176	1.1GB	asr	No
aishell2_test	AISHELL-2 Test	5,000	432MB	asr	No
audio_trivia_qa	Audio Trivia QA	1,024	322MB	qa	No
audio_web_questions	Audio Web Questions	2,032	222MB	qa	No
audiocaps_test	AudioCaps Test	3,985	139.2GB	caption	No
clothocaption_test	ClothoCaption Test	1,045	1.8GB	caption	No
commonvoice_en	CommonVoice English v15	16,386	664MB	asr	No
commonvoice_fr	CommonVoice French v15	16,132	653MB	asr	No
commonvoice_yue	CommonVoice Cantonese v15	5,593	217MB	asr	No
commonvoice_zh	CommonVoice Chinese v15	10,625	470MB	asr	No
covost2_en_zh	CoVoST2 EN-ZH	15,530	672MB	ast	No
covost2_zh_en	CoVoST2 ZH-EN	4,897	225MB	ast	No
fleurs_en	FLEURS 英文	647	383MB	asr	No
fleurs_zh	FLEURS 中文	945	664MB	asr	No
gigaspeech_test	GigaSpeech Test	19,870	4MB	asr	No
kespeech_test	KeSpeech Test	19,723	3.2GB	asr	No
librispeech_dev_clean	LibriSpeech Dev Clean	2,703	326MB	asr	No
librispeech_dev_other	LibriSpeech Dev Other	2,864	304MB	asr	No
librispeech_test_clean	LibriSpeech Test Clean	2,620	334MB	asr	No
librispeech_test_other	LibriSpeech Test Other	2,939	318MB	asr	No
meld	MELD	2,610	1.4GB	cls	Yes
mmar_bench	MMAR Bench	1,000	3.3GB	qa	No
mmau_test_mini	MMAU Test Mini	1,000	1.4GB	qa	No
mmsu_bench	MMSU Bench	4,996	1.3GB	qa	No
peoples_speech_test	People's Speech Test	34,898	3.8GB	asr	No
tedlium3_test	TED-LIUM v3 Test	3,737	790MB	asr	No
vocalsound	VocalSound	3,591	1.0GB	cls	No
voice_cmmlu	Voice CMMLU	11,582	13.3GB	qa	No
voicebench_advbench	VoiceBench AdvBench	520	0MB	qa	No
voicebench_alpacaeval	VoiceBench AlpacaEval	199	0MB	qa	No
voicebench_alpacaeval_full	VoiceBench AlpacaEval Full	636	0MB	qa	No
voicebench_bbh	VoiceBench BBH	1,000	0MB	qa	No
voicebench_commoneval	VoiceBench CommonEval	200	0MB	qa	No
voicebench_ifeval	VoiceBench IFEval	345	0MB	qa	No
voicebench_mmsu	VoiceBench MMSU	3,074	0MB	qa	No
voicebench_openbookqa	VoiceBench OpenBookQA	455	0MB	qa	No
voicebench_sdqa	VoiceBench SD-QA	6,083	0MB	qa	No
voicebench_wildvoice	VoiceBench WildVoice	1,000	0MB	qa	No
voxpopuli_en	VoxPopuli English	1,842	899MB	asr	No
wavcaps_audioset_sl	WavCaps AudioSet_SL	11,676	3.7GB	caption	No
wavcaps_freesound	WavCaps FreeSound	1,060	2.7GB	caption	No
wavcaps_soundbible	WavCaps SoundBible	1,232	530MB	caption	No
wenetspeech_test_meeting	WenetSpeech Test Meeting	8,370	377MB	asr	No
wenetspeech_test_net	WenetSpeech Test Net	24,774	682MB	asr	No

Omni (20 subsets, 46,648 samples, 432.1GB)

Subset	Display Name	Samples	Size	Subcategory	Video
av_odyssey	AV-Odyssey	4,555	31.2GB	qa	No
avmeme_full	AVMeme-Exam Full	1,032	777MB	meme_understanding	Yes
avmeme_main	AVMeme-Exam Main	846	643MB	meme_understanding	Yes
avut_benchmark_gemini	AVUT-Benchmark Gemini	9,874	49.4GB	qa	Yes
avut_benchmark_human	AVUT-Benchmark Human	1,734	12.3GB	qa	Yes
daily_omni	Daily-Omni	1,197	966MB	qa	Yes
futureomni	FutureOmni	1,034	50.7GB	qa	Yes
jointavbench	JointAVBench	2,853	37.4GB	qa	Yes
livesports3k_cc	LiveSports-3K CC	1,702	63.0GB	caption	Yes
omnibench	OmniBench	1,142	1.2GB	qa	No
ovavel	OV-AVEL Test	5,818	11.2GB	event_localization	Yes
ovobench	OVO-Bench	1,468	64.1GB	qa	Yes
streamingbench_real	StreamingBench-Real (Offline)	2,499	55.6GB	qa	Yes
streamingbench_sqa	StreamingBench-SQA (Offline)	250	6.5GB	qa	Yes
unobench	UNO-Bench	3,730	7.8GB	qa	No
unobench_mc	UNO-Bench-MC	1,000	3.6GB	qa	No
video_holmes	Video-Holmes	1,837	5.8GB	qa	Yes
videomme	Video-MME	5	26MB	qa	Yes
videomme_short	Video-MME (Short)	900	3.8GB	qa	Yes
worldsense	WorldSense	3,172	26.0GB	qa	Yes

Unknown (1 subsets, 1,000 samples, 1.7GB)

Subset	Display Name	Samples	Size	Subcategory	Video
streamingbench_omni	streamingbench_omni	1,000	1.7GB	unknown	No

Data Format

Each subset is stored as Parquet file(s):

• Metadata fields: All original JSONL fields (question, answer, paths, etc.)
• audio: HuggingFace Audio type — struct<bytes: binary, path: string>, playable in Dataset Viewer
• image: HuggingFace Image type — struct<bytes: binary, path: string>, viewable in Dataset Viewer
• video: Video binary — struct<bytes: binary, path: string>, embedded for applicable subsets
• audio_bytes_dict: Multiple audio files as base64-encoded JSON string
• image_bytes_dict: Multiple image files as base64-encoded JSON string

License

Apache 2.0. Individual datasets may have their own licenses.