datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
da-code-evaluation-resultstransformers-gh-memory
huggingface/transformers issues and pull requests, as a funes memory
Every issue and pull request of huggingface/transformers
with activity since 2022-01-01 — opening bodies, comments, reviews, inline review comments and PR
diffs — chunked, embedded and written to a Lance table by
funes, so the tracker can be searched by meaning and read
back thread by thread. Kept fresh every few minutes by the
funes-github Space.
Use it
Set funes up for your agent the usual way… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/transformers-gh-memory.funes-nvidia-Open-SWE-Traces
Funes recall store — NVIDIA Open-SWE-Traces (resolved)
A funes recall store built by indexing the
resolved==1 trajectories of
nvidia/Open-SWE-Traces
(65244 sessions, across both harnesses — SWE-agent and OpenHands — and both models,
Minimax-M2.5 and Qwen3.5-122B).
What this is
This is not a raw trace dataset — it is a pre-built funes index: the source
trajectories chunked into content blocks and embedded, stored as a
Lance table (chunks.lance).
Source… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/funes-nvidia-Open-SWE-Traces.dacomp-da-zh-eval
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
✍️ Citation
If you find our work helpful, please cite as
@misc{lei2025dacompbenchmarkingdataagents,
title={DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle},
author={Fangyu Lei and Jinxiang Meng and Yiming Huang and Junjie Zhao and Yitong Zhang and Jianwen Luo and Xin Zou and Ruiyi Yang and Wenbo Shi and Yan Gao and Shizhu He and Zuo Wang and Qian Liu and… See the full description on the dataset page: https://huggingface.co/datasets/DAComp/dacomp-da-zh-eval.funes-handoff-recall-benchmark
handover-vs-recall
A long investigation bloats an agent session until each new turn costs more to carry the context than to
do the work. Switching to a fresh session avoids that — but the findings have to travel somehow, and the
ways of moving them differ in cost. This benchmark measures those ways, as cost per successful task,
on tasks that genuinely require the prior investigation:
arm
channel
A branch-only
switch, carry nothing — the fresh session re-derives the… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/funes-handoff-recall-benchmark.modified_libero_rlds
Modified LIBERO RLDS Datasets
This repository contains the four modified LIBERO datasets
used in the OpenVLA fine-tuning experiments, stored in RLDS data format. See Appendix E in the
OpenVLA paper for details about the fine-tuning experiments and
specific dataset modifications, and see the OpenVLA GitHub README
for instructions on how to run OpenVLA in LIBERO environments.
Citation
BibTeX:
@article{kim24openvla,
title={OpenVLA: An Open-Source… See the full description on the dataset page: https://huggingface.co/datasets/dachengzisks/modified_libero_rlds.dacomp-da-eval
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
✍️ Citation
If you find our work helpful, please cite as
@misc{lei2025dacompbenchmarkingdataagents,
title={DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle},
author={Fangyu Lei and Jinxiang Meng and Yiming Huang and Junjie Zhao and Yitong Zhang and Jianwen Luo and Xin Zou and Ruiyi Yang and Wenbo Shi and Yan Gao and Shizhu He and Zuo Wang and Qian Liu and… See the full description on the dataset page: https://huggingface.co/datasets/DAComp/dacomp-da-eval.dacl10k
Dataset Card for dacl10k
dacl10k stands for damage classification 10k images and is a multi-label semantic segmentation dataset for 19 classes (13 damages and 6 objects) present on bridges.
The dacl10k dataset includes images collected during concrete bridge inspections acquired from databases at authorities and engineering offices, thus, it represents real-world scenarios. Concrete bridges represent the most common building type, besides steel, steel composite, and wooden bridges.… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/dacl10k.funes-xiaowu0162-longmemeval-cleaned-s
Funes recall store — LongMemEval_s cleaned corpus
A funes recall store built by indexing the
longmemeval_s_cleaned.json haystack of
xiaowu0162/longmemeval-cleaned
(LongMemEval, arXiv:2410.10813) — every unique
chat session across all 500 questions' haystacks, in one corpus-wide store.
What this is
This is not a raw trace dataset — it is a pre-built funes index: the source
sessions chunked into content blocks and embedded, stored as a
Lance table (chunks.lance).… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/funes-xiaowu0162-longmemeval-cleaned-s.commonvoice22-sidon-dacvae
CommonVoice 22 (Sidon-enhanced) converted to DAC VAE latents
Source
sarulab-speech/commonvoice22_sidon
Format
Each tar shard (~2GB) contains samples with three files per sample:
{sample_key}.audio.flac # Original audio (FLAC, original sample rate)
{sample_key}.dacvae.npy # DAC VAE latent [T_latent, 128] numpy float32
{sample_key}.metadata.json # All metadata + duration_seconds + chars_per_second
DAC VAE Latent Format
Model:… See the full description on the dataset page: https://huggingface.co/datasets/TTS-AGI/commonvoice22-sidon-dacvae.dach_bike_graph
DACH Bike + Rail Routing Graph
A prebuilt, ready-to-route cycling + railway graph covering Germany, Austria, and Switzerland (DACH), stored as lat/lon-tiled GeoParquet. Built from OpenStreetMap by the Bike Route Optimizer for flat-preferring, surface-aware bike routing that can also hop on a train uphill.
Everything routing needs is baked in — node elevations and full 3D edge geometry — so an application downloads this once and routes offline, with no Overpass and no elevation… See the full description on the dataset page: https://huggingface.co/datasets/MichaelMedek/dach_bike_graph.dacomp-da-zh
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
Paper | Project Page | Code
This repository contains DAComp, a benchmark of 210 tasks that mirrors complex real-world enterprise data intelligence workflows. It includes:
Data Engineering (DE) tasks: Require repository-level engineering on industrial schemas, including designing and building multi-stage SQL pipelines from scratch and evolving existing systems under evolving requirements.
Data Analysis (DA)… See the full description on the dataset page: https://huggingface.co/datasets/DAComp/dacomp-da-zh.librispeech_asr-audiodec_dac_16k
Dataset Card for "librispeech_asr-audiodec_dac_16k"
More Information needed
transformers-coding-session-pi-traces
dacorvo/transformers-coding-session-pi-traces
pi coding-agent session traces produced by
agentcap runs. Each run
contributes one folder under data/<run_id>/; inside, one file per
session in pi's native export format.
The on-the-wire HTTP captures for these same runs live in
dacorvo/transformers-coding-session-captures.
Both belong to the
transformers-coding-session Collection
— join on run_id to align captures with traces.
ViMed-PET-part1
Dataset description for three years: 2017, 2018, 2019
This dataset contains data from three years (2017, 2018, 2019). Each year has several month folders, which are named as THANG {month}.
Each year folder is compressed into zip files (chunks), each with an average size of approximately 2.5 GB.
Please unzip the .zip files to fully extract all data folders.
Folder structure after extraction
Each folder named THANG {month} of a year is divided into 3 subfolders… See the full description on the dataset page: https://huggingface.co/datasets/dacthai2807/ViMed-PET-part1.hf-hub-session-pi-traces
dacorvo/hf-hub-session-pi-traces
pi coding-agent session traces produced by
agentcap runs. Each run
contributes one folder under data/<run_id>/; inside, one file per
session in pi's native export format.
The on-the-wire HTTP captures for these same runs live in
dacorvo/hf-hub-session-captures.
Both belong to the
hf-hub-session Collection
— join on run_id to align captures with traces.
Datadacomp-de-gold
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
✍️ Citation
If you find our work helpful, please cite as
@misc{lei2025dacompbenchmarkingdataagents,
title={DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle},
author={Fangyu Lei and Jinxiang Meng and Yiming Huang and Junjie Zhao and Yitong Zhang and Jianwen Luo and Xin Zou and Ruiyi Yang and Wenbo Shi and Yan Gao and Shizhu He and Zuo Wang and Qian Liu and… See the full description on the dataset page: https://huggingface.co/datasets/DAComp/dacomp-de-gold.dacomp-da
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
✍️ Citation
If you find our work helpful, please cite as
@misc{lei2025dacompbenchmarkingdataagents,
title={DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle},
author={Fangyu Lei and Jinxiang Meng and Yiming Huang and Junjie Zhao and Yitong Zhang and Jianwen Luo and Xin Zou and Ruiyi Yang and Wenbo Shi and Yan Gao and Shizhu He and Zuo Wang and Qian Liu and… See the full description on the dataset page: https://huggingface.co/datasets/DAComp/dacomp-da.dacl10k
Dataset Card for dacl10k
dacl10k stands for damage classification 10k images and is a multi-label semantic segmentation dataset for 19 classes (13 damages and 6 objects) present on bridges.
The dacl10k dataset includes images collected during concrete bridge inspections acquired from databases at authorities and engineering offices, thus, it represents real-world scenarios. Concrete bridges represent the most common building type, besides steel, steel composite, and wooden… See the full description on the dataset page: https://huggingface.co/datasets/jhoncrad/dacl10k.ViMed-PET-part3
Dataset description for year 2023
This dataset contains data from three months: October, November, and December, stored in the following folders respectively:
THANG 10
THANG 11
THANG 12
The data is compressed into zip files (chunks), each with an average size of approximately 2.5 GB.
Please unzip the .zip files to fully extract the data folders.
Folder structure after extraction
Each folder named THANG {month} is divided into 3 subfolders, corresponding to 2… See the full description on the dataset page: https://huggingface.co/datasets/dacthai2k/ViMed-PET-part3.librispeech_asr-audiodec_dac_24kspeech-dac-tokens-3cb
Speech DAC Tokens (3 Codebooks)
Pre-tokenized speech dataset using the Descript Audio Codec (DAC). Each audio clip has been encoded into discrete codebook tokens from DAC's first 3 residual vector quantization codebooks, paired with its text transcription.
Dataset Summary
Stat
Value
Total samples
241,451
Total audio
~780 hours
Language
English
Codebooks
3 (of DAC's 9)
Codebook size
1,024 entries each
DAC model
44kHz
Tokens per second
~258 (86 frames… See the full description on the dataset page: https://huggingface.co/datasets/treadon/speech-dac-tokens-3cb.dacomp-da-zh-eval
DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle
✍️ Citation
If you find our work helpful, please cite as
@misc{lei2025dacompbenchmarkingdataagents,
title={DAComp: Benchmarking Data Agents across the Full Data Intelligence Lifecycle},
author={Fangyu Lei and Jinxiang Meng and Yiming Huang and Junjie Zhao and Yitong Zhang and Jianwen Luo and Xin Zou and Ruiyi Yang and Wenbo Shi and Yan Gao and Shizhu He and Zuo Wang and Qian Liu and… See the full description on the dataset page: https://huggingface.co/datasets/jjjsadhfgj/dacomp-da-zh-eval.longbench_synthetic_v4_1
LongBench Synthetic V4.1
Dataset statistics
v4.1 = lbs_v4 train+val (verbatim) + a test split absorbing every config from dac-research/extra_evals_v1 not already in v4. Five ZeroScrolls configs whose upstream corpora collide with lbs_v3.1 train/test (gov_report, qmsum, qasper, narrative_qa, musique) were dropped outright. All test rows have rubrics backfilled via the longbench_generate_rubric_for_imported.jinja template (same path as lbs_v3.1 / lbs_v4 Stage 1b).… See the full description on the dataset page: https://huggingface.co/datasets/dac-research/longbench_synthetic_v4_1.dacvae-tts-tr-w512-clean
DACVAE-TTS Turkish run C (width 512, clean data)
Generated audio of every evaluated checkpoint of the training run tr-w512-clean (Turkish zero-shot voice-cloning TTS,
dacvae-tts, frozen Meta DACVAE latents, 48 kHz). This repository holds
model outputs and metrics, not training data. Training data: Vyvo/tr-dataset-12 (Turkish podcast segments).
Run C: configs/nano_tr_w512.yaml (66.5M parameters: width 512, 8 heads, batch expansion 4, frame budget 6000), trained from scratch on… See the full description on the dataset page: https://huggingface.co/datasets/VoiceHub/dacvae-tts-tr-w512-clean.transformers-coding-session-captures
dacorvo/transformers-coding-session-captures
HTTP captures of agent ↔ model interactions — one parquet row per
/v1/chat/completions call. Produced by
agentcap.
Native session traces for the same runs live in companion datasets
named transformers-coding-session-<agent>-traces. They're all grouped under the
transformers-coding-session Collection
alongside this dataset. Join on run_id.
Loading
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/transformers-coding-session-captures.dacvae-tts-tr-nano-b-ke4
DACVAE-TTS Turkish run B (batch expansion 4)
Generated audio of every evaluated checkpoint of the training run tr-nano-b-ke4 (Turkish zero-shot voice-cloning TTS,
dacvae-tts, frozen Meta DACVAE latents, 48 kHz). This repository holds
model outputs and metrics, not training data. Training data: Vyvo/tr-dataset-12 (Turkish podcast segments).
Run B: configs/nano_tr_ke4.yaml (same 51.4M model as run A, context-sharing batch expansion 4, frame budget 7000), same data (~77 h), one RTX… See the full description on the dataset page: https://huggingface.co/datasets/VoiceHub/dacvae-tts-tr-nano-b-ke4.dacvae-tts-tr-nano-a
DACVAE-TTS Turkish run A (nano recipe, full data)
Generated audio of every evaluated checkpoint of the training run tr-nano-a (Turkish zero-shot voice-cloning TTS,
dacvae-tts, frozen Meta DACVAE latents, 48 kHz). This repository holds
model outputs and metrics, not training data. Training data: Vyvo/tr-dataset-12 (Turkish podcast segments).
Run A: configs/nano_tr.yaml (51.4M parameters), 17 shards of Vyvo/tr-dataset-12 with quality >= 55 (~77 h train), one RTX 4090, frame budget… See the full description on the dataset page: https://huggingface.co/datasets/VoiceHub/dacvae-tts-tr-nano-a.mls-enhanced-dacvae
Multilingual LibriSpeech converted to DAC VAE latents
Source
facebook/multilingual_librispeech
Format
Each tar shard (~2GB) contains samples with three files per sample:
{sample_key}.audio.flac # Original audio (FLAC, original sample rate)
{sample_key}.dacvae.npy # DAC VAE latent [T_latent, 128] numpy float32
{sample_key}.metadata.json # All metadata + duration_seconds + chars_per_second
DAC VAE Latent Format
Model:… See the full description on the dataset page: https://huggingface.co/datasets/TTS-AGI/mls-enhanced-dacvae.