datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
kinopoisk
Dataset Summary
Kinopoisk movie reviews dataset (TOP250 & BOTTOM100 rank lists).
In total it contains 36,591 reviews from July 2004 to November 2012.
With following distribution along the 3-point sentiment scale:
Good: 27,264;
Bad: 4,751;
Neutral: 4,576.
Data Fields
Each sample contains the following fields:
part: rank list top250 or bottom100;
movie_name;
review_id;
author: review author;
date: date of a review;
title: review title;
grade3: sentiment score Good, Bad… See the full description on the dataset page: https://huggingface.co/datasets/blinoff/kinopoisk.kin-cyber-dpo-v2
KIN Cybersecurity DPO v2 Preference Dataset
Empirically mined and zero-leak sanitized preference dataset for training cybersecurity and agentic code repair models.
Dataset Summary
Total DPO Pairs: 1,635 (Updated 2026-09-07)
Baseline v4 pairs: 1,495
Hermetic expansion (v5): +140 pairs (AST-invariant vulnerability repair, CWE-79 XSS guards, CWE-89 SQLi, CWE-22 Path Traversal, and supply chain integrity)
Format: Direct Preference Optimization (DPO) schema: {prompt… See the full description on the dataset page: https://huggingface.co/datasets/nyxspecter4/kin-cyber-dpo-v2.reasoning-base-20k
Dataset Card for Reasoning Base 20k
Dataset Details
Dataset Description
This dataset is designed to train a reasoning model. That can think through complex problems before providing a response, similar to how a human would. The dataset includes a wide range of problems from various domains (science, coding, math, etc.), each with a detailed chain of thought (COT) and the correct answer. The goal is to enable the model to learn and refine its reasoning process… See the full description on the dataset page: https://huggingface.co/datasets/KingNish/reasoning-base-20k.kinopoisk-sentiment-classificationAIME-COD
Dataset Summary
AIME-COD is a synthetic dataset created to solve problems from the American Invitational Mathematics Examination (AIME) using chain of draft reasoning, as proposed in the Chain of Draft: Thinking Faster by Writing Less paper.
The dataset was generated using Curator and synthetic reasoning produced by Gemini 2.0 Flash. Problems are sourced from the gneubig/aime-1983-2024 dataset.
Dataset Details
Purpose and Scope
The AIME-COD… See the full description on the dataset page: https://huggingface.co/datasets/KingNish/AIME-COD.lm-eval-results-PotatoB-Kinship-Exp-2-private
Dataset Card for Evaluation run of PotatoB/Kinship-Exp-2
Dataset automatically created during the evaluation run of model PotatoB/Kinship-Exp-2
The dataset is composed of 62 configuration(s), each one corresponding 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 always pointing to the latest results.
An additional… See the full description on the dataset page: https://huggingface.co/datasets/nyu-dice-lab/lm-eval-results-PotatoB-Kinship-Exp-2-private.read-along-ai-agent-traces
Read-Along AI - Agent Traces
This dataset contains the raw agent traces and conversation logs from the development of Read-Along AI, a submission for the Hugging Face Build Small Hackathon.
Dataset Description
These .jsonl files represent the unedited, behind-the-scenes "agent traces" of the AI coding assistant orchestrating the build of this project.
Sharing these traces fulfills the requirements for the "Sharing is Caring" bonus badge, providing the community… See the full description on the dataset page: https://huggingface.co/datasets/kingkw1/read-along-ai-agent-traces.KINA
Dataset Summary
Homepage | Paper | Hugging Face | GitHub
KINA (Knowledge Index of Noah's Ark) is a multidisciplinary knowledge benchmark for evaluating whether large language models can solve high-density, source-grounded, graduate-level questions across a broad map of human disciplines.
The dataset contains 899 ten-option pseudo-multiple-choice questions covering 261 fine-grained subfields, 70 fields, and 12 top-level disciplines.
KINA targets three problems in… See the full description on the dataset page: https://huggingface.co/datasets/2077AIDataFoundation/KINA.ScholarScope-data
ScholarScope Data
Research-session logs from ScholarScope, an AI grant and scholarship research agent. Each
row is one opportunity surfaced for a user's funding profile, ranked and extracted by the
ScholarScope pipeline from live web sources.
Dataset Summary
Rows: 97 opportunity records
Format: JSON Lines (scholarscope_sessions.jsonl); a Parquet mirror is auto-generated on
the refs/convert/parquet
branch and powers the dataset viewer.
Language: English
License:… See the full description on the dataset page: https://huggingface.co/datasets/kingabzpro/ScholarScope-data.made-repro-traces
Agent traces
Agent sessions published from a Trackio Logbook.
kingdom-return-path-bench
KINGDOM Return Path Bench v0
Return Path Bench is a small multiple-choice benchmark for inspecting how
feedback travels through a learning system. It keeps three evaluation lanes
separate because they establish different kinds of evidence:
Model behaviour records what an answer-selection policy does. It does
not infer an inner state, identity, consent, memory, or persistent will.
System/pipeline reasoning probes whether a model can identify
aggregation, evaluator-independence… See the full description on the dataset page: https://huggingface.co/datasets/Yu-and-Ai/kingdom-return-path-bench.gemma-4-e4b-kinetics_54K
Gemma-4 Kinetics 54K Video Caption Data
What: 54,618 cleaned Kinetics-600 video-caption records (75 action labels) in multimodal chat JSON, for video-VLM supervised fine-tuning.
Splits: train 43,696 / validation 5,461 / test 5,461 (80/10/10, stratified per label, seed 42, zero video overlap across splits).
Two prompt variants: annotations/splits-MQ/ (recommended) randomly combines 3 system × 5 user prompts per record to prevent prompt overfitting and format collapse;… See the full description on the dataset page: https://huggingface.co/datasets/bear7011/gemma-4-e4b-kinetics_54K.KingNish__Reasoning-0.5b-details
Dataset Card for Evaluation run of KingNish/Reasoning-0.5b
Dataset automatically created during the evaluation run of model KingNish/Reasoning-0.5b
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 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 always pointing to the latest results.
An additional… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/KingNish__Reasoning-0.5b-details.SRUM_6k_CompBench_Train
SRUM: Fine-Grained Self-Rewarding for Unified Multimodal Models - CompBench Training Dataset
This repository contains the training dataset for SRUM (Self-Rewarding for Unified Multimodal Models), a post-training framework that leverages a model's own understanding module to provide corrective signals and improve its generation module. This dataset is specifically designed for enhancing performance on compositionality benchmarks like T2I-CompBench.
Paper: SRUM: Fine-Grained… See the full description on the dataset page: https://huggingface.co/datasets/Wayne-King/SRUM_6k_CompBench_Train.ProjectAirSim-UAV-Kinematic-Trajectories
ProjectAirSim UAV Kinematic Trajectories
This dataset contains UAV trajectory episodes collected from ProjectAirSim. Each row is one episode. The states field is a variable-length sequence sampled at approximately 10.00 Hz.
State vector:
[t, x, y, z, qw, qx, qy, qz, vx, vy, vz, wx, wy, wz, ax, ay, az]
Fields:
episode_id: integer episode index.
trajectory_type: trajectory family used to generate waypoints.
start_position: NED start position [x, y, z] in meters.
goal_position:… See the full description on the dataset page: https://huggingface.co/datasets/qincheng037/ProjectAirSim-UAV-Kinematic-Trajectories.flame-kindling-v1
flame-kindling-v1
A small, opinionated SFT dataset for finetuning a 3B-class instruct model into a character designer that emits a strict JSON schema from a free-text seed. Built to replace a general RP model (Mistral-Nemo-12B Mahou finetune) being shoehorned into JSON output for flammen.ai's Create-a-Flame pipeline.
400 (seed → DesignedFlame) pairs distilled from Claude Sonnet 4.5 with tool-forcing, validated against a strict pydantic schema, deduplicated by name and… See the full description on the dataset page: https://huggingface.co/datasets/flammenai/flame-kindling-v1.Dr.Sparse-RL-train-562
Dr.Sparse SpGEMM training pool (562 matrices)
The complete RL / selector training pool of Dr.Sparse (branch v2): 562 SuiteSparse matrices in the
harness .bin layout (int32 rows, cols, nnz; int32 row_ptr; int32 col_ind; float32 values; float32 x),
laid out as level1_small/ (91), level2_medium/ (273), level3_large/ (198); the huge tier is deliberately left out of training and evaluation.
Every matrix has a cuSPARSE SpGEMM reference (C = AA, or AA^T when rectangular) on an H200;… See the full description on the dataset page: https://huggingface.co/datasets/KinGeorge/Dr.Sparse-RL-train-562.NinjaMasker-PII-RedactionEstate-Rawkin_med_2Mss-pocfrench-customer-review-sentiment-free-2k
French Customer Review Sentiment Free 2K
French Customer Review Sentiment Free 2K is a free 2,000-record sample extracted from the full commercial dataset French Customer Review Sentiment (100k synthetic reviews) provided by Kinoux.
Each entry is a synthetic French customer review labeled with a 3-class sentiment:
positive
neutral
negative
The data is 100% synthetic (no personal data, no real platform exports) and was generated and curated specifically for training and evaluating… See the full description on the dataset page: https://huggingface.co/datasets/Kinoux/french-customer-review-sentiment-free-2k.KingNish__qwen-1b-continued-v2-details
Dataset Card for Evaluation run of KingNish/qwen-1b-continued-v2
Dataset automatically created during the evaluation run of model KingNish/qwen-1b-continued-v2
The dataset is composed of 38 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 1 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 always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/open-llm-leaderboard/KingNish__qwen-1b-continued-v2-details.oczy-r20-calibration-dev-2557aa4african-kingdoms-history
African Kingdoms — Historical Dataset
Structured historical data on major African kingdoms and their military, economic, and strategic innovations. All public domain sources.
Kingdoms Covered
Zulu Kingdom (Shaka)
Ethiopian Empire (Adwa victory)
Mali Empire (Sundiata, Mansa Musa)
Kingdom of Aksum
Ghana Empire
Swahili Coast City-States
Mfecane (military revolution, southern Africa)
All content sourced from public domain primary sources and documented historical… See the full description on the dataset page: https://huggingface.co/datasets/gmahia/african-kingdoms-history.gemma-4-e4b-kinetics_330K
This datset compose of 295,612 training and 32,845 validation Kinetics-600 video-caption pairs across 479 action labels.
Please unzip the file first
peft_test_safemaze2d_easy_native256_cot_chunk_kinf_20260707_perseg
maze2d_easy_native256_cot_chunk_kinf_20260707_perseg
Maze2d (native 256px, JPEG q95; navigation with stop-required success, easy→hard split) — action-conditioned visual world-model SFT data (CoT self-rollout) for the
BAGEL-7B-MoT feedback-interval study.
Format: gzipped JSONL shards under training/, 1 row = 1 packed episode. CoT rows: per-segment
layout — <think> per-step imagined frame (MSE target) </think> + committed action chunk, with a
loss-0 "Action executed." + real… See the full description on the dataset page: https://huggingface.co/datasets/ultrastar111/maze2d_easy_native256_cot_chunk_kinf_20260707_perseg.french-spam-ham-detection-free-2k
French Spam/Ham Detection Free 2K
French Spam/Ham Detection Free 2K is a free 2,000-record sample extracted from the full commercial dataset French Spam/Ham Detection (56,400 synthetic messages) provided by Kinoux.
Each entry is a synthetic French message labeled with a binary classification:
spam
ham
The data is 100% synthetic (no personal data, no scraped emails, no real platform exports) and was generated specifically for training and evaluating French-native spam detection… See the full description on the dataset page: https://huggingface.co/datasets/Kinoux/french-spam-ham-detection-free-2k.Dans-Kinomaxx-VanillaBackrooms