FreeLLMMeta FAIR launches e-means, a k‑means drop‑in that converges 10× faster
e-means reaches Lloyd's 25‑epoch loss in 1.4‑2.6 epochs, offering up to 45× GPU speedup and MIT licensing via FAISS 1.16.0.
Meta FAIR researchers introduced e-means, a mini‑batch k‑means variant that updates centroids with exponential moving averages.
The algorithm matches the loss of Lloyd's k‑means after 25 epochs in just 1.4 to 2.6 epochs, according to self‑reported tests on 24 dataset configurations.
Benchmarks show 10‑45× GPU wall‑clock speedup over Flash‑KMeans and 8‑15× CPU speedup versus FAISS k‑means, with a 2.5‑epoch run on 10 billion 1,024‑dim embeddings split into 16.7 M clusters.
The code is released under the MIT license, making it immediately available for vector‑indexing, quantization and data‑curation pipelines.
Ce que cela signifie pour vous
For developers, e-means reduces clustering time dramatically, enabling faster indexing of massive embedding collections on existing hardware. Free‑software users gain a high‑performance, MIT‑licensed alternative to proprietary k‑means implementations, supporting both CPU and GPU workflows.
Sources
- Meta FAIR Releases e-means: An Algorithm Beating k-means in 10x Fewer Epochs, MarkTechPost, 10 Nov 2026
- [2611.09501] e-means: outperforming k-means (in one-tenth the epochs), arxiv.org (official), 10 Nov 2026
Rédigé automatiquement à partir des sources ci‑dessus, puis vérifié par notre code : chaque nombre et chaque nom doivent y figurer, et aucune phrase ne doit copier une source. Une erreur ? Signalez‑la.
Quelle IA pouvez‑vous utiliser gratuitement aujourd’hui ?
Tous les modèles IA gratuits, classés par votes humains anonymes, avec la mémoire et la vitesse dont ils ont besoin sur votre propre GPU.