datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
3D_native_rot_gen3D_native_genSWE-Bench-Verified-O1-native-tool-calling-reasoning-high-results
SWE-Bench Verified O1 Dataset
Executive Summary
This repository contains verified reasoning traces from the O1 model evaluating software engineering tasks. Using OpenHands + CodeAct v2.2, we tested O1's bug-fixing capabilities using their native tool calling capabilities on the SWE-Bench Verified dataset, achieving a 45.8% success rate across 500 test instances.
Overview
This dataset was generated using the CodeAct framework, which aims to improve code… See the full description on the dataset page: https://huggingface.co/datasets/AlexCuadron/SWE-Bench-Verified-O1-native-tool-calling-reasoning-high-results.AlGhafa-Arabic-LLM-Benchmark-Native
AlGhafa Arabic LLM Benchmark
New fix: Normalized whitespace characters and ensured consistency across all datasets for improved data quality and compatibility.
Multiple-choice evaluation benchmark for zero- and few-shot evaluation of Arabic LLMs, we adapt the following tasks:
Belebele Ar MSA Bandarkar et al. (2023): 900 entries
Belebele Ar Dialects Bandarkar et al. (2023): 5400 entries
COPA Ar: 89 entries machine-translated from English COPA and verified by native Arabic… See the full description on the dataset page: https://huggingface.co/datasets/OALL/AlGhafa-Arabic-LLM-Benchmark-Native.carla-ue5-native-erp-2048x1024-5m-track-100kglaive-function-calling-v2-openai-native
glaive-function-calling-v2-openai-native
glaiveai/glaive-function-calling-v2 restructured into the native OpenAI / TRL
format: tools is a typed column and tool_calls[].function.arguments is a
real object — not JSON inside a string.
The original is widely used (69k downloads/month) but inactive for ~3 years, and
ships tool calls as <functioncall> text blobs with Python-quoted arguments.
Existing repackagings either keep ShareGPT with tools as a string, or carry
no license at all.… See the full description on the dataset page: https://huggingface.co/datasets/Archangel-system/glaive-function-calling-v2-openai-native.common-native-proc
Dataset Card for [Dataset Name]
Dataset Summary
[More Information Needed]
Supported Tasks and Leaderboards
[More Information Needed]
Languages
[More Information Needed]
Dataset Structure
Data Instances
[More Information Needed]
Data Fields
[More Information Needed]
Data Splits
[More Information Needed]
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data… See the full description on the dataset page: https://huggingface.co/datasets/DTU54DL/common-native-proc.AraMix-Native
AraMix-Native
A native-Arabic-filtered version of
AdaMLLab/AraMix (minhash_deduped),
derived from SultanR/AraMix-Translation-Scores:
machine-translated and garbled-MT documents removed, 162,887,010 rows kept of
178,883,241 (91.06%). All columns preserved.
Filter rules
A document is kept iff all of:
mmbert_translated_score < 0.1, or a classical-text rescue: diacritic
(tashkeel) ratio ≥ 0.02 over Arabic letters and ≥ 3 distinct diacritic
classes (fully/partially… See the full description on the dataset page: https://huggingface.co/datasets/SultanR/AraMix-Native.native-sft
native-sft
A format-alignment remix, not new instruction data. Conversations come from AllenAI Dolci (ODC-By) and NVIDIA Nemotron-Post-Training-Dataset-v1 (CC BY 4.0). Each family config re-renders those chats through a real 2026 instruct template so SFT can keep native special tokens / think / tools markers. Trainers get prompt + completion, so they do not need {% generation %} in jinja.
v1 2026-08-31: ~9609 canonical conversations; 57 unique-hash family configs; 539,326… See the full description on the dataset page: https://huggingface.co/datasets/asaverren/native-sft.babylon-native-v8-noise-op-wisecentral-florida-native-plants
DeepEarth Central Florida Native Plants Dataset v0.2.0
🌿 Dataset Summary
A comprehensive multimodal dataset featuring 33,665 observations of 232 native plant species from Central Florida. This dataset combines citizen science observations with state-of-the-art vision and language embeddings for advancing multimodal self-supervised ecological intelligence research.
Key Features
🌍 Spatiotemporal Coverage: Complete GPS coordinates and timestamps for all… See the full description on the dataset page: https://huggingface.co/datasets/deepearth/central-florida-native-plants.kaggle-native-v8-noise-op-wiseLlamaLens-Arabic-Native
LlamaLens: Specialized Multilingual LLM Dataset
Overview
LlamaLens is a specialized multilingual LLM designed for analyzing news and social media content. It focuses on 18 NLP tasks, leveraging 52 datasets across Arabic, English, and Hindi.
LlamaLens
This repo includes scripts needed to run our full pipeline, including data preprocessing and sampling, instruction dataset creation, model fine-tuning, inference and evaluation.
Features… See the full description on the dataset page: https://huggingface.co/datasets/QCRI/LlamaLens-Arabic-Native.AlGhafa-Arabic-LLM-Benchmark-Native-10percentmyanmar-native-text
Dataset Card for myanmar-native-text
Dataset Summary
ဒီ dataset က myanmar-native-text အတွက် ဖန်တီးထားတာပါ။
Languages
Myanmar (my) / English (en)
Dataset Structure
Data Instances
{ "text": "နမူနာ စာသား", "label": "အညွှန်း" }
Data Fields
text: main content, label: optional.
Data Splits
Split
Files
train
data/native_00000.jsonl, data/native_00001.jsonl, data/native_00002.jsonl… See the full description on the dataset page: https://huggingface.co/datasets/kkomyoeminaung/myanmar-native-text.BFCL-V4-Parallel-Native
BFCL V4 Parallel Native
Native BFCL v4 single-turn parallel function-calling rows for decentralized multi-agent collaboration.
Source data comes from the official Berkeley Function Calling Leaderboard v4 data and possible-answer files.
Fields
id
official_category
task_type
user_prompt
function
ground_truth
Categories
live_parallel
live_parallel_multiple
parallel
parallel_multiple
Counts
train: 352 rows
eval: 88 rows
total: 440… See the full description on the dataset page: https://huggingface.co/datasets/OpenMLRL/BFCL-V4-Parallel-Native.auditbench-graft-vs-native-eval-results
AuditBench — graft vs native organisms, evaluation results
Numeric evaluation results for the AuditBench model-organism grid on two model families:
Qwen3-14B and Llama-3.3-70B-Instruct. The organisms themselves are published separately
(djroytburg/auditbench-qwen3-14b-*, djroytburg/auditbench-llama33-70b-*).
The design
Each cell compares three arms on the same eval, served together:
arm
meaning
bare
the untouched instruct model
native
SDF… See the full description on the dataset page: https://huggingface.co/datasets/djroytburg/auditbench-graft-vs-native-eval-results.hey-native-wakeword
Hey Native — wake-word dataset
Synthetic 16 kHz mono audio for training a "Hey Native" wake-word detector.
data/positive/ — utterances of "Hey Native" (label 1)
data/negative/ — general speech, not the wake word (label 0)
data/hard_negative/ — near-miss confusables, e.g. "hey navy", "hey maybe" (label 0)
metadata.csv — file_name, label, label_name, text, source_model, mos_p808
5,000 positives · 6,000 negatives · 1,050 hard negatives.
voxcpm2-native-generated-audio-user-ref
VoxCPM2 Native Generated Audio (User Ref)
Raw audio files generated from the native VoxCPM2 path in sglang-omni using a user-provided reference clip.
Contents
9 generated .wav files
metadata.json with prompt text, mode, status, size, and latency
Source Reference Audio
Reference clip used for the reference-mode generations:
https://huggingface.co/datasets/adarshxs/voxcpm2-native-test-samples/resolve/main/data/audio.wav
Files
ref_expressive.wav… See the full description on the dataset page: https://huggingface.co/datasets/adarshxs/voxcpm2-native-generated-audio-user-ref.dfm11-toolace-native-tool-use-repaired
dfm11-toolace-native-tool-use-repaired
ToolACE conversations with declared-name parsing and complete parallel result binding.
This is a DFM11 replacement for schneiderkamplab/dfm10-toolace-native-tool-use. All rows pass exhaustive structural validation. See metadata/manifest.json.
vam-cross-target-widowx250-native-corner-frontThis dataset was created using LeRobot.
Dataset Description
30.10 minutes (160 episodes, 160 marked successful) of the VAM-Cross target collection for widowx250 in MuJoCo at 30 Hz using the widowx-texture robot appearance. Observations include RGB, joint state, gripper state, and achieved EE pose; actions include the full commanded EE pose and gripper command. Task assets are derived from the MolmoSpaces THOR 20251117 asset release (MolmoSpaces commit… See the full description on the dataset page: https://huggingface.co/datasets/dreamdifferent/vam-cross-target-widowx250-native-corner-front.entity-native-agent-sessions
Entity-Native vs File-Native Agent Sessions on SWE-bench Verified
Full session logs from a controlled A/B experiment measuring how a coding agent's
retrieval substrate changes its behaviour, cost, and success rate on real
software-engineering tasks.
Both arms run the same model (Claude Sonnet 4.5), on the same tasks, from the
same repository state. The only difference is how the agent is allowed to find code.
Arm
Label
Tools available
A
file-native
Bash, Read, Grep… See the full description on the dataset page: https://huggingface.co/datasets/rs545837/entity-native-agent-sessions.LlamaLens-Hindi-Native
LlamaLens: Specialized Multilingual LLM Dataset
Overview
LlamaLens is a specialized multilingual LLM designed for analyzing news and social media content. It focuses on 18 NLP tasks, leveraging 52 datasets across Arabic, English, and Hindi.
LlamaLens
This repo includes scripts needed to run our full pipeline, including data preprocessing and sampling, instruction dataset creation, model fine-tuning, inference and evaluation.
Features… See the full description on the dataset page: https://huggingface.co/datasets/QCRI/LlamaLens-Hindi-Native.gcd-native-v8-noise-op-wisemedmcqa-openai-native
MedMCQA — OpenAI-native, with a usable test split
MedMCQA is one of the most downloaded medical QA datasets on the Hub. Its test split has been unusable since release: all 6,150 rows carry cop=-1 (no label) and an empty explanation. You cannot score a model on it.
This release rebuilds a labelled, leak-free test split and converts everything to the native messages format, so it loads straight into TRL with no custom parsing.
What was actually wrong
Measured on the… See the full description on the dataset page: https://huggingface.co/datasets/Archangel-system/medmcqa-openai-native.generative-native-ads
Webis Generated Native Ads 2024
Dataset Summary
This dataset was created to train ad blocking systems on the task of identifying advertisements in responses of conversational search engines.
There are two dataset dictionaries available:
responses.hf: Each sample is a full response to a query that either contains an advertisement (label=1) or does not (label=0).
sentence_pairs.hf: Each sample is a pair of two sentences taken from the responses. If one of them… See the full description on the dataset page: https://huggingface.co/datasets/webis/generative-native-ads.tinker-qwen3.5-9b-native-eval-traces
Native Tinker-style reproduction: retained trace archive
This archive contains the retained student evaluation generations, not a complete training trace. It accompanies three public PEFT adapters: SFT step 400, one-step OPD from SFT step 400, and last SFT checkpoint, step 2888.
evaluations/sft/step-*/aime_2024.jsonl holds one-sample AIME 2024 generations for 40 SFT checkpoints (steps 2–24 every two steps, then 100–2800 every 100 steps). evaluations/opd-step1/aime_2024.jsonl… See the full description on the dataset page: https://huggingface.co/datasets/open-athena/tinker-qwen3.5-9b-native-eval-traces.details_chavinlo__alpaca-native
Dataset Card for Evaluation run of chavinlo/alpaca-native
Dataset Summary
Dataset automatically created during the evaluation run of model chavinlo/alpaca-native on the Open LLM Leaderboard.
The dataset is composed of 64 configuration, each one coresponding to one of the evaluated task.
The dataset has been created from 2 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard-old/details_chavinlo__alpaca-native.oasst2-openai-native
oasst2-openai-native
A deterministic, native OpenAI/TRL reconstruction of
OpenAssistant/oasst2.
It turns the original flat parent_id message table into two directly usable
configs without LLM transformation: multilingual SFT conversations and ranked
DPO preference pairs.
At a glance
Config
Train
Test
Unit
sft
12,717
671
alternating conversation ending in assistant
dpo
42,639
2,284
prompt + chosen/rejected assistant pair
The data is multilingual:… See the full description on the dataset page: https://huggingface.co/datasets/Archangel-system/oasst2-openai-native.psp-native-centroids
Praxel/psp-native-centroids
Native-speaker reference artefacts for the PSP (Phoneme Substitution Profile)
benchmark for Indic text-to-speech accent evaluation. Companion to the paper
PSP: An Interpretable Per-Dimension Accent Benchmark for Indic Text-to-Speech
(Menta, 2026).
This dataset is a scoring reference, not a training corpus. It contains
pre-computed acoustic references extracted from publicly-licensed native-speaker
speech corpora, used by the psp-eval package
to score TTS… See the full description on the dataset page: https://huggingface.co/datasets/Praxel/psp-native-centroids.