UMAP 0.6 released: 3M arXiv papers in under 5 minutes

On November 11, UMAP author Leland McInnes released UMAP 0.6, a version that upgrades the dimensionality-reduction tool in both speed and quality. On 3 million arXiv papers (titles + abstracts embedded into 4096-dimensional vectors by F2LLM-v2 8B), the new version completes dimensionality reduction in under 5 minutes, a major improvement for large-scale use.

Confirmed

• Neighbor search is roughly 4x faster thanks to better use of PyNNDescent
• A new initialization method makes initialization faster
• A new optimizer produces reproducible results with a fixed seed, and runs about 12x faster on all cores
• Despite the speed gains, the new optimizer resolves data structure at finer scales, with neighborhood accuracy notably better than UMAP 0.5—improving both fidelity and detail
• An official tutorial, "Embedding Very Large Datasets," was released alongside, covering the key steps at million-scale: nearest-neighbor search, embedding initialization, layout optimizer tuning, and more

Why it matters

UMAP is a widely used dimensionality-reduction and visualization tool, and speed plus initialization overhead on large data has long been its main bottleneck. Version 0.6 makes embedding million- or even tens-of-millions-scale datasets feasible in reasonable time and reproducible, while actually improving accuracy—a real win for users handling massive high-dimensional embeddings such as LLM-generated vectors.

2026-11-11 ~ 2026-11-11 · 13 related posts

• [source] UMAP 0.6 Released: 3M arXiv Papers Embedded in Under 5 Minutes — leland_mcinnes · 2026-11-11
• UMAP 0.6's new optimizers resolve finer structure and beat 0.5's neighborhood accuracy while running faster — leland_mcinnes · 2026-11-11
• UMAP 0.6 recursive init: UMAP initializing itself, 32s vs 4min for spectral — leland_mcinnes · 2026-11-11
• UMAP 0.6 recursive init converges to within 3% of final layout in far fewer epochs — leland_mcinnes · 2026-11-11
• UMAP 0.6 fidelity gains at scale driven by efficient hard-negative mining — leland_mcinnes · 2026-11-11
• UMAP 0.6 optimizers lift attraction-repulsion restrictions, boost kNN recall — leland_mcinnes · 2026-11-11
• UMAP 0.6 docs: how to embed million-scale datasets, tuned via 3M arXiv papers — leland_mcinnes · 2026-11-11
• [source] UMAP 0.6 now installable via PyPI and conda-forge — leland_mcinnes · 2026-11-11
• UMAP 0.6 is live on GitHub and installable with a single pip command — leland_mcinnes · 2026-11-11

4 near-duplicate retellings: leland_mcinnes · leland_mcinnes · leland_mcinnes · leland_mcinnes