The Stack v3 expands to 114 TB with 224M repos and fixes a dedup bug
anton_lozhkov · x · 2026-07-23
The Stack v3 is out as the largest open code dataset ever released: 114 TB, 224M repositories, 770 languages, and roughly 5T deduplicated, filtered source-code tokens.
Key points:
- The v2 pipeline had a bug: MinHash shingles were tokenized with \W instead of \W+, causing many empty shingles and excessive hash collisions.
- That bug made v2 over-deduplicate heavily.
- v3 fixes the regex, runs MinHash-LSH across all languages together, and validates every candidate pair against its Jaccard estimate.
- The release claims zero false positives.
- It is fully open and excludes restrictively licensed code.
The accompanying chart shows large token growth across major languages, with C++ reaching 802B tokens and the all-language total at about 4.9T.
Related event: Hugging Face Releases The Stack v3: 114TB Largest Open Source Code Dataset(13 posts)→
More from Infra
- Nvidia Is Now Core to Every Major Robotaxi Stack at Commercial Scale — pdamodaran · 2026-09-11
- 12 KV Cache Reduction Techniques Every AI Engineer Should Understand, Explained — blaizedsouza · 2026-09-11
- The shadow GPU capacity market is formalizing, with Meta selling excess compute to outside buyers — DavidLinthicum · 2026-09-11
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11