datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
nanochat-climbmix-arithmetic-base10
nanochat ClimbMix + Base-10 Arithmetic
This dataset contains the first 170 shuffled ClimbMix training shards
used by nanochat's speedrun. The deterministic base-10 arithmetic corpus is
mixed into shards 00000..00149; the final
20 train shards are unchanged web-only padding.
The original validation shard (shard_06542.parquet) is also
copied unchanged.
Arithmetic corpus
Family
Examples
a + b = c (all ordered pairs 0..2000, two exposures)
8,008,002
a + b… See the full description on the dataset page: https://huggingface.co/datasets/Yujivus/nanochat-climbmix-arithmetic-base10.TTCW-Based-Review
TTCW Creative Writing Evaluation Dataset
If you use this dataset in your research, please cite our paper — it helps support ongoing academic work. Citation details are at the bottom of this page.
Dataset Description
Summary
A supervised fine-tuning (SFT) dataset for training LLMs to act as creative writing evaluators. Each example contains a creative story and four message-format columns representing different evaluation objectives — from… See the full description on the dataset page: https://huggingface.co/datasets/VibrantVista/TTCW-Based-Review.nanochat-climbmix-arithmetic-base7
nanochat ClimbMix + Arithmetic: base-7 numeral world
This is a deterministic base-7 rendering of
Yujivus/nanochat-climbmix-arithmetic-base10. It preserves
the exact shard names, row order, document order, arithmetic-document placement,
and non-numeric text of the source dataset.
Transformation rule
Every maximal ASCII digit run matching [0-9]+ is interpreted as a base-10
integer and rendered in base 7. Leading zeros are preserved as a prefix; signs,
punctuation… See the full description on the dataset page: https://huggingface.co/datasets/Yujivus/nanochat-climbmix-arithmetic-base7.BenchMAX_Rule-based
Dataset Sources
Paper: BenchMAX: A Comprehensive Multilingual Evaluation Suite for Large Language Models
Link: https://huggingface.co/papers/2502.07346
Repository: https://github.com/CONE-MT/BenchMAX
Dataset Description
BenchMAX_Rule-based is a dataset of BenchMAX, sourcing from IFEval, which is a rule-based benchmark for evaluating the instruction following capabilities in multilingual scenarios.
We extend the original dataset to 16 non-English languages by first… See the full description on the dataset page: https://huggingface.co/datasets/LLaMAX/BenchMAX_Rule-based.maxrl_qwen3_4B_base_polaris_rollouts
MaxRL Qwen3-4B-Base training rollouts (POLARIS math prompts)
Every training rollout from an online RL run, with exact token ids, sampling
log-probs, and raw rewards — usable as a replay buffer to study off-policy RL
for LLM reasoning completely offline.
The run: Qwen3-4B-Base trained with the maxRL advantage estimator
(A = (r - mean)/(mean + eps), group mean over 16 rollouts per prompt;
maxRL paper) and a pure REINFORCE loss
(L = -A * log pi; no importance ratio, no clipping, no… See the full description on the dataset page: https://huggingface.co/datasets/ftajwar/maxrl_qwen3_4B_base_polaris_rollouts.nanochat-climbmix-arithmetic-base6
nanochat ClimbMix + Arithmetic: base-6 numeral world
This is a deterministic base-6 rendering of
Yujivus/nanochat-climbmix-arithmetic-base10. It preserves
the exact shard names, row order, document order, arithmetic-document placement,
and non-numeric text of the source dataset.
Transformation rule
Every maximal ASCII digit run matching [0-9]+ is interpreted as a base-10
integer and rendered in base 6. Leading zeros are preserved as a prefix; signs,
punctuation… See the full description on the dataset page: https://huggingface.co/datasets/Yujivus/nanochat-climbmix-arithmetic-base6.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.BananaMind-Base-Bench-1.1
BananaMind Base Bench 1.1
BananaMind Base Bench 1.1 is an English text-completion benchmark for base causal language models. It contains 350 individually authored examples across seven categories and reports one fixed-scale Overall Elo score.
This is not an instruction-following benchmark. Models receive plain text followed by four possible continuations. The official runner selects the continuation with the highest mean conditional token log-probability. It does not use a chat… See the full description on the dataset page: https://huggingface.co/datasets/BananaMind/BananaMind-Base-Bench-1.1.gemma4-e2b-base-topk128-hf-overlay-v128-seed42
Gemma 4 E2B base top-k-128 HF training overlay
This is the immutable training-engine overlay used to distill traces from Gemma 4 E2B base into
Gemma 4 E4B. It preserves the prompts, responses, and exact response token IDs from
JWei05/gemma4-e2b-base-topk128-traces,
but replaces the source vLLM top-k targets with targets recomputed by the Hugging Face training
engine.
This repository is a reproducibility artifact for the corresponding distillation run. It is not a
new… See the full description on the dataset page: https://huggingface.co/datasets/JWei05/gemma4-e2b-base-topk128-hf-overlay-v128-seed42.nanochat-calendar-arithmetic-base10
nanochat Base-10 Calendar Arithmetic
A deterministic, base-10 arithmetic corpus scoped to three cyclic calendar
units: hour-of-day (mod 24), day-of-week (mod 7), and
month-of-year (mod 12). Companion to
Yujivus/nanochat-climbmix-arithmetic-base10,
built the same way but scoped to real modular calendar units instead of
free-integer add/sub/mul/div/mod.
Every example is a single line — question and answer collapsed into one
equation, no exposed reasoning:
23:00 + 18965h = 04:00… See the full description on the dataset page: https://huggingface.co/datasets/liodon-ai/nanochat-calendar-arithmetic-base10.dclm-baseline-1.0-parquet_urls
Dataset Card for dclm-baseline-1.0-parquet_urls
This dataset provides the URLs and top-level domains associated with training records in mlfoundations/dclm-baseline-1.0-parquet. It is part of a collection of datasets curated to make exploring LLM training datasets more straightforward and accessible.
Dataset Details
Dataset Description
This dataset was created by downloading the source data, extracting URLs and top-level domains, and retaining only… See the full description on the dataset page: https://huggingface.co/datasets/nhagar/dclm-baseline-1.0-parquet_urls.cancer-knowledge-base
Cancer Knowledge Base — the open, verified oncology KB for RAG & LLM evaluation
The only open CC-BY-4.0 oncology knowledge base that combines:
110/110 trials cited with PMID + NCT + PubMed/ClinicalTrials.gov URLs, and 32 prognosis
rows linked to verified SEER 2016–2022 references — no LLM-synthetic dataset has this.
A provable 152-question MCQ benchmark — every answer derives from this KB's own structured
data and carries a citation + golden docs, so it is open-book verifiable… See the full description on the dataset page: https://huggingface.co/datasets/ranjithraj/cancer-knowledge-base.BenchMAX_Model-based
Dataset Sources
Paper: BenchMAX: A Comprehensive Multilingual Evaluation Suite for Large Language Models
Link: https://huggingface.co/papers/2502.07346
Repository: https://github.com/CONE-MT/BenchMAX
Dataset Description
BenchMAX_Model-based is a dataset of BenchMAX, sourcing from m-ArenaHard, which evaluates the instruction following capability via model-based judgment.
We extend the original dataset to include languages that are not supported by m-ArenaHard through… See the full description on the dataset page: https://huggingface.co/datasets/LLaMAX/BenchMAX_Model-based.deepagent
DeepAgent
Hard, Docker-verifiable software-engineering benchmarks from real merged PRs
DeepAgent ships real_pr Harbor hardness packs: live-mined multi-file pull requests, clone@SHA agent images, held-out verifier tests, and Docker dual-truth (solution reward = 1, null reward = 0). Primary product work runs through the deepagent CLI in the GitHub monorepo.
Surface
Ref
Role
HF stable pin
this dataset revision main
Current product on Hub (N=9)
HF automation… See the full description on the dataset page: https://huggingface.co/datasets/BaseIntelligence/deepagent.silicon-based-girlfriend-v2-dataset
矽基女友 v2 · 繁中角色扮演合成語料
繁體中文（臺灣）角色扮演的合成對話語料，2,109 筆、38,460 輪、角色輪合計 1,350 萬字元。
訓練出來的模型見 RX5950XT/silicon-based-girlfriend-v2-GGUF。
⚠️ 全部是模型合成的資料，不是真人對話。 內容包含成人向角色扮演，不適合未成年人。
僅供研究用途。所有角色皆為虛構成年人。
內容
檔案
內容
sharegpt_dataset.json
2,109 筆多輪對話，ShareGPT 格式（id / system / conversations）
grpo_prompts.json
648 題 GRPO 用的提示，與 SFT 語料零重疊
holdout_ids.json
100 筆 holdout ID 清單，這些已從 SFT 訓練集排除，供驗收用
general_probes.json
64 題通用能力探針（5 類），用來檢測微調後有無退化… See the full description on the dataset page: https://huggingface.co/datasets/RX5950XT/silicon-based-girlfriend-v2-dataset.qwen3.5-2b-base-blind-spots
Qwen3.5-2B-Base — Blind Spot Analysis (Text + Vision)
Model Tested
Field
Value
Model
Qwen/Qwen3.5-2B-Base
Parameters
2.27 B (2,274 M per HF metadata)
Architecture
Hybrid Gated-DeltaNet (dense FFN) — 24 LM layers (18 DeltaNet + 6 full-attention), ViT vision encoder
Type
Pre-trained base model (not instruction-tuned)
Context
262 144 tokens
Modalities
Text + Vision (early-fusion multimodal)
Key Contributions
Only multimodal… See the full description on the dataset page: https://huggingface.co/datasets/F555/qwen3.5-2b-base-blind-spots.Competence-Based-Evaluation
Competence-Based Evaluation (Invariance Benchmark)
A benchmark for testing whether language models give the same answer to
semantically equivalent reformulations of a logical-ordering question. Given a
set of pairwise constraints (e.g. Alice is in front of Bob), a model should
answer transitive-closure queries (Is Carol in front of Dave?) consistently
whether the constraints are stated using a relation or its inverse.
Each item exists as a paired (original, equivalent) record… See the full description on the dataset page: https://huggingface.co/datasets/jizej/Competence-Based-Evaluation.T2T-Centromere-Regulatory
T2T Centromere Regulatory
Curated and released by Basepair | Follow updates on X: @BasepairSci.
Dataset Summary
The T2T Centromere Regulatory is the first comprehensive, base-pair resolution mapping of cryptic transcriptional switches and secondary structural elements across the newly sequenced Telomere-to-Telomere (T2T-CHM13 v2.0 / hs1) human centromeres.
For decades, centromeric alpha-satellite DNA (~100–200 Mb across human chromosomes) was considered… See the full description on the dataset page: https://huggingface.co/datasets/Basepair/T2T-Centromere-Regulatory.task966_ruletaker_fact_checking_based_on_given_context
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task966_ruletaker_fact_checking_based_on_given_context
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task966_ruletaker_fact_checking_based_on_given_context.byob-pd-book-corpus
Dataset Card for BYOB LM on Steroids - Public-Domain Book Training Corpus
A multilingual, public-domain corpus of literary, philosophical, and scientific
works, published for character-level language-model pre-training. It ships in two
forms: raw per-author .txt files (one file per author) and a pre-tokenized,
memmap-ready cache (train.bin / val.bin / meta.json) for fast training.
Dataset Sources
Project Gutenberg (https://www.gutenberg.org) - all tiers
Standard… See the full description on the dataset page: https://huggingface.co/datasets/disco-jack-basement/byob-pd-book-corpus.dev-knowledge-base
Dev Knowledge Base (Programming Documentation Dataset)
A large-scale, structured dataset of programming documentation collected from official sources across languages, frameworks, tools, and AI ecosystems.
Do Follow me on Github: https://github.com/nuhmanpk
Overview
This dataset contains cleaned and structured documentation content scraped from official developer docs across multiple domains such as:
Programming languages
Frameworks (frontend, backend)
DevOps &… See the full description on the dataset page: https://huggingface.co/datasets/nuhmanpk/dev-knowledge-base.math-dow-mod-synthetic-v1-base6
Math + Cyclic-Time Synthetic Dataset
Synthetic dataset for training a small (~10M-100M param), task-specialized
LLM on arithmetic (addition, multiplication), cyclic time arithmetic
(days-of-week, months, 12-hour and 24-hour clock), and mod-k remainder
probes — generalization-focused rather than memorization, following on
from the T2/T5/T10 modular-circuit discussion. Days-of-week, months, and
hours are all instances of the same underlying cyclic/modular-addition
structure… See the full description on the dataset page: https://huggingface.co/datasets/liodon-ai/math-dow-mod-synthetic-v1-base6.nanoJEPA-base
nanoJEPA EN/ZH Ultra-FineWeb Dataset
This is a small pretraining dataset package for nanoJEPA. It is built by
streaming openbmb/Ultra-FineWeb split en and/or zh.
Files
train.jsonl: {"text": "...", "source": "...", "dataset": "...", "language": "en|zh", "score": 0.0}
valid.jsonl: same schema as train.jsonl
test.jsonl: same schema as train.jsonl
Generation Command
uv run python data/build_hf_dataset.py \
--out-dir dataset/nanojepa-small \
--languages en,zh… See the full description on the dataset page: https://huggingface.co/datasets/TerenceLau/nanoJEPA-base.math-dow-mod-synthetic-v1-base7
Math + Cyclic-Time Synthetic Dataset
Synthetic dataset for training a small (~10M-100M param), task-specialized
LLM on arithmetic (addition, multiplication), cyclic time arithmetic
(days-of-week, months, 12-hour and 24-hour clock), and mod-k remainder
probes — generalization-focused rather than memorization, following on
from the T2/T5/T10 modular-circuit discussion. Days-of-week, months, and
hours are all instances of the same underlying cyclic/modular-addition
structure… See the full description on the dataset page: https://huggingface.co/datasets/liodon-ai/math-dow-mod-synthetic-v1-base7.GAR_baseDataset
GAR Base Dataset
This repository contains the base dataset used in the paper GAR: Generative Adversarial Reinforcement Learning for Formal Theorem Proving.
GAR (Generative Adversarial Reinforcement learning) is a comprehensive RL training framework that jointly trains a problem composer and a solver in an adversarial loop. This dataset serves as the starting point for the implicit curriculum learning mechanism, which aligns task difficulty with the prover's evolving capability to… See the full description on the dataset page: https://huggingface.co/datasets/RickyDeSkywalker/GAR_baseDataset.pkm-agent-baseline-v2
PKM Agent Baseline — 500 + 50 Scenarios + Six-Grader Artifacts (v2)
Two deterministically generated, Korean-language benchmarks for evaluating multi-tool Personal Knowledge Management (PKM) agents over Notion, Gmail, and Google Calendar, plus the Six-Grader Ensemble scoring artifacts (100-scenario reference subset + per-scenario six-metric scores for vanilla and LoRA models).
Released alongside the preprint:
Vault-Grounded 4B Agent: A Hybrid Reasoning–Fact Architecture for Local… See the full description on the dataset page: https://huggingface.co/datasets/newtype-2038/pkm-agent-baseline-v2.wikipedia-embeddings-cs-e5-baseThis dataset contains the Czech subset of the wikimedia/wikipedia dataset. Each page is divided into paragraphs, stored as a list in the chunks column. For every paragraph, embeddings are created using the intfloat/multilingual-e5-base model.
Usage
Load the dataset:
from datasets import load_dataset
ds = load_dataset("karmiq/wikipedia-embeddings-cs-e5-base", split="train")
ds[1]
{
'id': '1',
'url': 'https://cs.wikipedia.org/wiki/Astronomie',
'title': 'Astronomie'… See the full description on the dataset page: https://huggingface.co/datasets/karmiq/wikipedia-embeddings-cs-e5-base.MSD_manual_topics_user_base
MSD_manual_topics_user_base
This dataset has been built with the website https://www.msdmanuals.com/ provided by Merck & Co for the greater audience.
The MSD manual is an essential source of knowledge for many topics related to symptoms, diseases, health and other related topics. The manual makes an extra effort to make it available both for professionals and patients by having two distinct version.
The content, while being labelled the same, differs by the type of user in order to… See the full description on the dataset page: https://huggingface.co/datasets/nuvocare/MSD_manual_topics_user_base.ashaar-with-enhanced-descriptions-baseform-final-sft-lte20-min500-splits
Ashaar Enhanced Description SFT Stratified Splits
Source dataset:
Shaer-AI/ashaar-with-enhanced-descriptions-baseform-final-sft-lte20-min500
Target dataset:
Shaer-AI/ashaar-with-enhanced-descriptions-baseform-final-sft-lte20-min500-splits
This dataset publishes deterministic train / eval / test splits with a 94 / 3 / 3 policy.
Split policy
Primary stratification key:
base_meter
form
length_bucket
Length buckets:
1-3
4-6
7-10
11-20
Small groups fall back… See the full description on the dataset page: https://huggingface.co/datasets/Shaer-AI/ashaar-with-enhanced-descriptions-baseform-final-sft-lte20-min500-splits.GAR_baseDataset_NuminaMath
GAR-Official
This is the official repository for the paper GAR: Generative Adversarial Reinforcement Learning for Formal Theorem Proving.
GitHub Repository: RickySkywalker/GAR-Official
Trained Models:
GAR_Goedel-Prover-V2
GAR_DeepSeek-Prover-V2
Base Datasets:
Original base dataset
Base dataset under Numina-Math
Introduction
We introduce GAR: Generative Adversarial Reinforcement Learning, an RL training method that intends to solve inefficiency and suboptimal… See the full description on the dataset page: https://huggingface.co/datasets/RickyDeSkywalker/GAR_baseDataset_NuminaMath.