datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
voxbox
VoxBox
This dataset is a curated collection of bilingual speech corpora annotated clean transcriptions and rich metadata incluing age, gender, and emotion.
Dataset Structure
.
├── audios/
│ └── aishell-3/ # Audio files (organised by sub-corpus)
│ └── ...
└── metadata/
├── aishell-3.jsonl
├── casia.jsonl
├── commonvoice_cn.jsonl
├── ...
└── wenetspeech4tts.jsonl # JSONL metadata files
Each JSONL file corresponds to a… See the full description on the dataset page: https://huggingface.co/datasets/SparkAudio/voxbox.safedocs-1M-muse-spark-1.3-judged
SafeDocs: Muse Spark 1.3 judge annotations
Incrementally published, one complete shard per commit. All original source columns,
images, complete Paddle JSON, rows and row order are preserved. No language or quality
filtering. New columns: judge_verdict (PERFECT/ERROR), judge_reason, judge_status,
and judge_error. Operational failures retain the original page with a null verdict
and reason, status failed, and a diagnostic in judge_error; they are not OCR ERRORs.
Direct Meta API… See the full description on the dataset page: https://huggingface.co/datasets/albertklorer/safedocs-1M-muse-spark-1.3-judged.Sparkle
Sparkle: Realizing Lively Instruction-Guided Video Background Replacement via Decoupled Guidance
Ziyun Zeng, Yiqi Lin, Guoqiang Liang, and Mike Zheng Shou
📦 Dataset
Sparkle is a large-scale video background replacement dataset comprising ~140K high-quality source–edited video pairs. It is fully open-sourced at 🤗stdKonjac/Sparkle. For full methodology and dataset details, please refer to our paper.
The dataset is organized into five themes along different… See the full description on the dataset page: https://huggingface.co/datasets/stdKonjac/Sparkle.Spark-234K
Spark-234K: Skeleton-Guided Scientific Reasoning from Large-Scale Literature
🎉 Accepted to EMNLP 2026 Findings!
Spark-234K is a scientific reasoning dataset containing 234K question-answer pairs synthesized from frontier scientific literature. Instead of directly generating QA pairs from full papers, SPARK first distills each paper into a compact reasoning skeleton—preserving its central claim, supporting evidence, quantitative relations, assumptions, and boundary… See the full description on the dataset page: https://huggingface.co/datasets/OpenDataArena/Spark-234K.thomas-2018-spark-wt
SPARK (wild-type accumulator phenotype): Human-curated and standardized MICs
These data were collated by the authors of:
Joe Thomas, Marc Navre, Aileen Rubio, and Allan Coukell
Shared Platform for Antibiotic Research and Knowledge: A Collaborative Tool to SPARK Antibiotic Discovery
ACS Infectious Diseases 2018 4 (11), 1536-1539
DOI: 10.1021/acsinfecdis.8b00193
We cleaned the original SPARK dataset to subset the most relevant columns, remove empty values,
give succint column… See the full description on the dataset page: https://huggingface.co/datasets/scbirlab/thomas-2018-spark-wt.SageLM-CosyVoice2-SparkTTSSageLM-Spark-TTSTimeWarp-GPT5-TracesSparkle-Bench
Sparkle: Realizing Lively Instruction-Guided Video Background Replacement via Decoupled Guidance
Ziyun Zeng, Yiqi Lin, Guoqiang Liang, and Mike Zheng Shou
📦 Dataset
Sparkle is a large-scale video background replacement dataset comprising ~140K high-quality source–edited video pairs. It is fully open-sourced at 🤗stdKonjac/Sparkle. For full methodology and dataset details, please refer to our paper.
The dataset is organized into five themes along different… See the full description on the dataset page: https://huggingface.co/datasets/stdKonjac/Sparkle-Bench.AutomotiveUI-Bench-4K
AutomotiveUI-Bench-4K
Dataset Overview: 998 images and 4,208 annotations focusing on interaction with in-vehicle infotainment (IVI) systems.
Key Features:
Serves as a validation benchmark for automotive UI.
Scope: Covers 15 automotive brands/OEMs, model years 2018-2025.
Image Source: Primarily photographs of IVI displays (due to screenshot limitations in most vehicles), with some direct screenshots (e.g., Android Auto).
Annotation Classes:
Test Action: Bounding box + imperative… See the full description on the dataset page: https://huggingface.co/datasets/sparks-solutions/AutomotiveUI-Bench-4K.sparkproof-miningamc2023spark2-5-tiny-cpu-repro-v1
Spark2.5 tiny random CPU fixture
Complete untrained Spark2_5ForCausalLM with independently seeded random BF16 weights.
This is a reproducibility fixture, not a useful language model, distilled model,
quality benchmark, or production registry measurement. No upstream weights, training
data, paid GPU or cloud rentals were used.
Architecture, code and license
Source: XHToken/Spark-X2.5-4B at 5e10fcc0286756aebf7c41dc52c1e42d95c70281.
The complete text causal model… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/spark2-5-tiny-cpu-repro-v1.dgx-spark-benchmarks
DGX Spark LLM Arena benchmarks
Reproducible LLM inference benchmarks on an NVIDIA DGX Spark (GB10, 128 GB unified memory). The suite defines eleven tests: six closed-loop (llama-benchy) and five open-loop (vllm bench serve). Results cover all eleven: the ten throughput tests under results, and the rate sweep under rateSweep. Raw results remain inspectable, but only complete runs without a failed sanity check count toward rankings and aggregate throughput. Open-loop tests must… See the full description on the dataset page: https://huggingface.co/datasets/Djangodevreng/dgx-spark-benchmarks.safedocs-1M-muse-spark-1.3-first3
SafeDocs first three shards: Muse Spark 1.3
Source: albertklorer/safedocs-1M, revision 87faff9053aa50c745f1359bef3592219ccb8c8b.
PaddleOCR-VL 1.6 teacher labels are compared to original pages using the existing side-by-side renderer and binary Muse Spark 1.3 contributor judge. Quality verdicts are only PERFECT or ERROR, with no quality reason. Operational failures have no verdict. These are model labels, not human ground truth. Native Paddle block list order is preserved. A… See the full description on the dataset page: https://huggingface.co/datasets/albertklorer/safedocs-1M-muse-spark-1.3-first3.NewsLensSync
Dataset Card for NewsLensSync
Dataset Description
This dataset, named NewsLensSync, contains a curated collection of news articles, sourced from trusted domains such as BBC, Reuters, AP News, NPR, PBS, The Guardian, WSJ, NY Times, and ProPublica. Each article includes both the original content and a synthetic "falsified" version of the article description, generated using a transformer-based negation model. The dataset is designed for research in misinformation… See the full description on the dataset page: https://huggingface.co/datasets/sparklessszzz/NewsLensSync.spark-tts-twi-preppedInstaArt-HumanAI
Instagram AI Art vs Human Art: Engagement & Comment Dataset
Dataset Summary
This dataset was created and contributed by Akshaya, Cynthia, Grace, and Soham as part of a project at UC San Diego.
This dataset supports research into how audiences engage with AI-generated art versus human-made art on Instagram, with a specific focus on comment sentiment, reaction types, and engagement patterns. It consists of 40 matched pairs of Instagram posts - one human art post and one… See the full description on the dataset page: https://huggingface.co/datasets/sparklessszzz/InstaArt-HumanAI.spark2-5-tiny-fidelity-root-v1
spark2-5 random CPU fixture root
A root fidelity dataset in hidden form, produced by engines/tools/hf_capture.py from malaiwah/spark2-5-tiny-random-bf16.
The cut
the final hidden state handed to lm_head -- after the text model's final norm and immediately before the head matmul -- captured as the head module's input via torch.nn.Module.register_forward_pre_hook; replay applies the head ONLY (no final norm at replay time: the capture already sits after it). Same… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/spark2-5-tiny-fidelity-root-v1.nova-industry-benchmark
Nova Industry Benchmark
22 curated questions across six categories, used to evaluate NovaLiveSystem.
This dataset contains questions only. Model outputs live in
SparkSupernova/nova-industry-benchmark-results,
joined on id.
Why the split
Run outputs used to be stored here as additional splits. Each model version wrote a
different set of columns, so adding the v5 run left this dataset with two splits whose
schemas disagreed, and load_dataset raised… See the full description on the dataset page: https://huggingface.co/datasets/SparkSupernova/nova-industry-benchmark.Everything_about_dogs
Dataset Card for Everything About Dogs Dataset
This dataset is built from the book Everything About Dogs
Everything about dogs is a book by AL G Eberhart.
The book contains topics on diseases, how to feed, environment to provide, how to take care of pups and everything about dogs.
Motivation
In Memory of Caspu
Dogs can't speak but they can express their pain. However this expression is often misunderstood or ignored.
My I aim is to build a retrieval… See the full description on the dataset page: https://huggingface.co/datasets/SparkleDark/Everything_about_dogs.Lab2spark-math-audit-20260911
Spark-X2.5: solving and auditing misleading worked solutions
Status: experiment running; not a completed competition entry yet.
Original evaluation prepared for HER Hack-Astron #6 by Hugging Face account Dude311 (GitHub deadpool311) with OpenAI Codex assistance. Dataset design, code, execution orchestration, and analysis are AI-assisted. Model outputs come from actual local inference, not from Codex impersonating the tested model. No human review of the model's reasoning traces… See the full description on the dataset page: https://huggingface.co/datasets/Dude311/spark-math-audit-20260911.dgx-spark-benchmarks
DGX Spark LLM Benchmarks
First comprehensive benchmark suite for NVIDIA DGX Spark (GB10 Blackwell).
Hardware
GPU: NVIDIA GB10 Blackwell (1 PFLOP FP4)
Memory: 128GB unified LPDDR5x (273 GB/s)
CPU: 20-core ARM (10x Cortex-X925 + 10x Cortex-A725)
Storage: 4TB NVMe
Framework: Ollama 0.18.3
CUDA: 13.0 | Driver: 580.142
Benchmark Results
Run 1 — General Inference (11 models)
Model
Size
Prompt tok/s
Gen tok/s
Load Time
Llama 3.1 8B
4.9 GB… See the full description on the dataset page: https://huggingface.co/datasets/G3nadh/dgx-spark-benchmarks.SPARK
⚡ SPARK (multi-vision Sensor Perception And Reasoning benchmarK)
🌐 github | 🤗 Dataset | 📃 Paper
Dataset Details
SPARK can reduce the fundamental multi-vision sensor information gap between images and multi-vision sensors. We generated 6,248 vision-language test samples automatically to investigate multi-vision sensory perception and multi-vision sensory reasoning on physical sensor knowledge proficiency across different formats, covering different types of… See the full description on the dataset page: https://huggingface.co/datasets/topyun/SPARK.sparkle_preview
🤔 About SPARKLE
The work examines how auxiliary information (hints) influences model behavior under RL, studying four types of hints:
Partial Step Scaffolding,
High-level Plans,
External Knowledge, and
Chains of Subproblems.
The SPARKLE dataset provides the structured annotations—plans, knowledge snippets, and decomposed subproblems—that enable this analysis.
These annotations allow researchers to probe how models respond to different forms of auxiliary information and how… See the full description on the dataset page: https://huggingface.co/datasets/sparkle-reasoning/sparkle_preview.dgx-spark-moe-benchmarks
Four MoE models on a DGX Spark: speed, tool-calling, and what actually breaks
Full measurement campaign on NVIDIA DGX Spark (GB10, 128 GB unified, ~273 GB/s),
vLLM 0.23.1rc1.dev301+g04c2a8dea, arm64/sm121. Every number here is measured on this
hardware, with the raw evidence included.
The headline: on synthetic tool-calling benchmarks all four models score 91-95 %. In a
real coding agent, three of them score 0-1 out of 14 and one scores 11 out of 14.
If you pick a model from the… See the full description on the dataset page: https://huggingface.co/datasets/pocharlies/dgx-spark-moe-benchmarks.craft-arithmetic24-spark-budget-study
CraftArithmetic24 and Spark CPU budget study
Twenty-four original synthetic arithmetic questions across six related problem families. Each family contains a base problem, paraphrase, changed-number problem and irrelevant-detail variant. See dataset.jsonl and dataset_metadata.json for prompts, exact labels, version and hash.
The original questions and labels are released under CC0-1.0. The original evaluation code is released under MIT. Model-generated outputs are included as… See the full description on the dataset page: https://huggingface.co/datasets/eruantion87/craft-arithmetic24-spark-budget-study.kupe-spark-150m-conversationsSparkMe-SyntheticUsers
SparkMe-SyntheticUsers
Synthetic user profiles for evaluating AI interview systems, released alongside the SparkMe. Each profile represents a simulated workforce participant with demographic metadata, a shuffled list of persona facts, and structured ground-truth interview notes across 10 topics covering the impact of AI in the workplace.
Dataset Description
The 200 profiles were generated from WorkBank worker seed data using SparkMe's user agent pipeline. Each user has:… See the full description on the dataset page: https://huggingface.co/datasets/SALT-NLP/SparkMe-SyntheticUsers.