← TinyGPT · docs · devlog · roadmap · speedup
source: docs/roadmap/datasets.md · view on GitHub ↗

Roadmap — open-source datasets

Verified URLs, sizes, and licenses for everything in the pretrain → SFT → DPO pipeline. Picked for “this fits on a 48 GB Mac” or “we can stream a slice.”

Pretraining corpora

DatasetSizeLicenseNotes
HuggingFaceFW/fineweb-edu1.3 T tokens (default), sample-10BT / 100BT / 350BT subsets, 114 totalODC-BY-1.0Our pick. Educational-quality web text; each doc has a 2.5-5.06 quality score. CC-MAIN snapshots from 2013-2025.
HuggingFaceFW/fineweb15 T tokensODC-BY-1.0Broader, less-filtered parent of FineWeb-edu. Use when fineweb-edu becomes a bottleneck.
togethercomputer/RedPajama-Data-V230 T+ tokensApache 2.0 + variousRedPajama’s open Llama-equivalent. Mostly Common Crawl.
cerebras/SlimPajama-627B627 B tokensvariousDeduplicated subset of RedPajama-V1.
allenai/c4750 GBODC-BYCleaned Common Crawl (T5’s pretraining data).
wikimedia/wikipedia~20 GB EnglishCC-BY-SAClean encyclopedic prose. Pairs well with web corpora.
HuggingFaceTB/cosmopedia~25 B tokensApache 2.0Synthetic educational text generated by Mixtral. Used to train SmolLM.
HuggingFaceTB/smollm-corpus~600 B tokensApache 2.0SmolLM’s pretraining mix — Cosmopedia + FineWeb-edu + Stack.
roneneldan/TinyStories~2 GBCDLA-Permissive-1.0Designed for sub-100M-param models specifically. Vocabulary capped at ~1500 words; the 1M-param coherence threshold.
monology/pile-uncopyrighted~825 GBvariousThe Pile minus copyrighted portions. Pythia’s training data.
bigscience/roots1.6 T tokensvariesBLOOM’s pretraining corpus. Multilingual.
allenai/dolma3 T tokensvariousOLMo’s training data. Web + academic (arXiv + Semantic Scholar) + code (StarCoder) + Reddit + StackExchange + books. More diverse mix than FineWeb (which is pure web). Smaller (3T vs FineWeb’s 15T) but higher per-token diversity.

SFT (supervised fine-tuning)

DatasetSizeLicenseNotes
databricks/databricks-dolly-15k15 K pairsCC-BY-SA-3.0Hand-written, high quality. Our first-run pick.
HuggingFaceH4/no_robots10 K pairsCC-BY-NC-4.0Hand-written, broader topic coverage than Dolly.
allenai/tulu-3-sft-mixture939 K pairsODC-BY-1.0 + mixed18 datasets blended. Includes Persona-MATH, WildChat-GPT4, FLAN-v2, Evol-CodeAlpaca, OpenAssistant-Guanaco, WildGuardMix, etc. The kitchen-sink SFT set.
teknium/OpenHermes-2.5~1 M conversationsmixedDistilled from GPT-4. Higher quality per pair than Tulu.
Open-Orca/SlimOrca~518 KMITSlimmer Orca; distilled instructions.
tatsu-lab/alpaca52 KCC-BY-NC-4.0GPT-3.5-generated; classic. Lower quality per pair.
OpenAssistant/oasst1~10 K conversationsApache 2.0Multi-turn, human-labeled.
HuggingFaceH4/ultrachat_200k200 K conversationsMITLarge-scale multi-turn chat.

Preference data (DPO / KTO / SimPO / ORPO)

DatasetSizeLicenseNotes
HuggingFaceH4/ultrafeedback_binarized187 K paired (61K train_prefs split)MITOur DPO pick. Six splits: train_sft / train_prefs / train_gen × test/train. GPT-4 judgments. TruthfulQA-decontaminated.
argilla/dpo-mix-7k7 K pairedmixedSmaller, cleaner. Good for small DPO runs.
argilla/ultrafeedback-binarized-preferences-cleaned60 K pairedMITUltraFeedback with corrected labels (~hundred fixed).
allenai/llama-3.1-tulu-3-8b-preference-mixture270 K pairedODC-BYTulu-3’s preference mix; uses on-policy DPO data.
anthropic/hh-rlhf160 K pairedMITHuman-labeled “helpful” and “harmless” preferences. Slow to grow but human-grade.
nvidia/HelpSteer340 KCC-BY-4.0Single-label good/bad (use with KTO).
argilla/distilabel-capybara-dpo-7k-binarized7 K pairedmixedSmaller curated set.

Code data

DatasetSizeLicenseNotes
bigcode/the-stack-v23.1 B files (~67 TB raw)variousLargest permissive-license code corpus. Use subsets.
codeparrot/github-code-clean~110 GBvariousPre-filtered Github code.
bigcode/the-stack-dedup~3 TBvariousDeduplicated v1.
nickrosh/Evol-Instruct-Code-80k-v180 KMITCode SFT data via Evol-Instruct.
HuggingFaceH4/CodeAlpaca_20K20 KCC-BY-4.0Hand-cleaned code instructions.

Math + reasoning

DatasetSizeLicenseNotes
meta-math/MetaMathQA395 KMITMath instruction-following.
AI-MO/NuminaMath-CoT860 KApache 2.0Olympiad-style math + Chain-of-Thought.
openai/gsm8k8.5 KMITGrade-school math word problems. The standard math benchmark.
hendrycks/competition_math12.5 KMITMATH dataset; competition-level math.

Eval / benchmark datasets

BenchmarkSourceNotes
MMLUcais/mmlu57-task knowledge benchmark. Standard for “general capability.”
HellaSwagRowan/hellaswagCommon-sense continuation.
TruthfulQAtruthfulqa/truthful_qaTruthfulness eval.
ARCallenai/ai2_arcGrade-school science MC.
GSM8K(see math row above)Math word problems.
MT-BenchHuggingFaceH4/mt_bench_promptsLLM-judge instruction-following.
AlpacaEvaltatsu-lab/alpaca_evalPairwise GPT-4 judgments.
IFEvalgoogle/IFEvalInstruction-following with verifiable constraints. Rule-based eval — no LLM judge.

2025-era datasets

DatasetUseNotes
open-thoughts/OpenThoughts-114kReasoning SFTOpen-source reasoning traces, R1-style
deepseek-ai/DeepSeek-R1-Distill-Qwen-7BDistillation teacher(And siblings: 1.5B, 14B, 32B). If we want to do reasoning distillation, these are the obvious teachers.
agentica-org/DeepScaleR-Preview-DatasetRLVR / mathMath problems with verifiable answers
Magpie-Align/Magpie-Llama-3.3-1MSynthetic SFT1M synthetic SFT pairs from Llama-3.3-70B (Jan 2025).