datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
speedrunbench
SpeedrunBench
LLM agents optimizing speedruns across multiple games. Each run ships as a silent video of the run that
was scored, and (except for Tuxemon) the input tape that produced it: the frame to goal and lower
is always better.
config
game
metric
supertux
SuperTux
frames to goal
astray
Astray
frames to maze exit
tuxemon
Tuxemon
frames to first gym
pokemon_blue
Pokémon Blue
frames to first badge
sml
Super Mario Land
frames to clear 1-1
smb
Super Mario… See the full description on the dataset page: https://huggingface.co/datasets/PatronusAI/speedrunbench.KreyolNER-Corpus
KreyolNER-Corpus
A manually audited, high-precision Named Entity Recognition (NER) dataset for Haitian Creole, structured following strict BIO schema integrity.
So far, we've tagged 2k sentences. The goal is around 7-10k.
Schema
Covers 10 standardized entity types: PER, LOC, ORG, DATE, TIME, DURATION, QUANTITY, MISC, FREQUENCY, and MONEY.
Data Format
Standard JSONL format where each line contains tokenized text and corresponding BIO tags:
{… See the full description on the dataset page: https://huggingface.co/datasets/Speedk4011/KreyolNER-Corpus.need_for_speed_unbound_recordings_01
极品飞车22 raw recordings
This dataset contains raw game recordings managed by Game Data Platform. Access requests require manual approval.
Game ID: game_63484fc85bb6d29bee28f41d71eb1582
Collection: general (泛数据)
Recordings: 15
Layout: recordings/<recording_id>/<raw component>
reckless-driving-speed-thresholds-by-state
Reckless / excessive-speeding mph thresholds that trigger a criminal charge, by state
Canonical, always-current version: https://referencesource.org/reckless-driving-speed-thresholds-by-state/
Machine-readable: https://referencesource.org/reckless-driving-speed-thresholds-by-state/data.json — this mirror is a point-in-time copy.
Last verified: 2026-08-20
Stale after: 2027-08-20 (past this date, prefer the canonical copy —
it re-verifies on a cadence this snapshot does not)… See the full description on the dataset page: https://huggingface.co/datasets/referencesource/reckless-driving-speed-thresholds-by-state.SPEED-Bench-Qualitative-Qwen3.6-35B-A3B-FP8-torchspec
SPEED-Bench Qualitative Qwen3.6 TorchSpec
TorchSpec-compatible chat dataset generated from the 880 fully materialized SPEED-Bench qualitative prompts.
Responses were generated on Doubleword with Qwen/Qwen3.6-35B-A3B-FP8 using /v1/chat/completions and max_tokens=4096.
Files
data/train.jsonl: 880 rows in TorchSpec chat format.
Schema
Each row contains:
{
"id": "<speedbench_question_id>",
"conversations": [
{"role": "user", "content":… See the full description on the dataset page: https://huggingface.co/datasets/jamesdborin/SPEED-Bench-Qualitative-Qwen3.6-35B-A3B-FP8-torchspec.speed_of_lightechidna-round8-speed
echidna-round8-speed
Echidna — round 8 speed/conciseness training (Hedgehog style).
Contents
round8_speed.jsonl (18 rows)
Format
JSON Lines (.jsonl), one example per line.
Provenance
Original content for the Echidna RAG assistant (Michael Anthony Falabella).
ptv3-bericht-lora-de-300
ptv3-bericht-lora-de-300
Synthetic German dataset for fine-tuning LLMs to generate structured psychotherapy reports (PTV-3 / Bericht an den Gutachter) from therapy session transcripts.
Overview
Property
Value
Samples
311 (280 train / 31 val)
Language
German
Format
ChatML JSONL (system / user / assistant)
Teacher model
Qwen2.5-27B (local)
Generation
Two-stage: seed → session transcript → PTV-3 JSON report
Schema
Each sample… See the full description on the dataset page: https://huggingface.co/datasets/speed-brain-ai/ptv3-bericht-lora-de-300.sdf-data-bee_speedLLM4PP_dataset
Code Optimization Format
For the code optimization task, the dataset should consist of a list of examples in a .jsonl format. Each example should be a dictionary with two fields: src_code and tgt_code.
src_code is considered the slow code and tgt_code is considered the fast code.
The code in the dataset should run with only the C++ standard library and OpenMP, however we will try to accommodate external libraries that are easy to install.
An example dataset derived from Leetcode… See the full description on the dataset page: https://huggingface.co/datasets/speedcode/LLM4PP_dataset.jungle_speed
jungle_speed
This dataset was generated using phosphobot.
This dataset contains a series of episodes recorded with a robot and multiple cameras. It can be directly used to train a policy using imitation learning. It's compatible with LeRobot.
To get started in robotics, get your own phospho starter pack..
speed-benchmark
Speed Benchmark
CPU inference speed for all 22 verified dispatchAI models.
Measured with llama-cpp-python (8 threads, 512 context).
🚀 dispatchAI
speed-ranking
Speed Ranking
All 31 working dispatchAI models ranked by CPU inference speed.
🚀 dispatchAI
speed1speed1speed1speed2speed3speed4speed5walking_variable_speed.jsonaa-speedbench-10k