Sentence Transformers 5.6.1 fixes silent flash-attention embedding regressions
tomaarsen · x · 2026-07-23
Sentence Transformers v5.6.1 is out as a patch release.
- It fixes a bug where flashattention2 silently degraded embeddings for XLM-R and RoBERTa models.
- The issue affects users encoding with Transformers v5 and flash attention enabled.
- The maintainer says anyone using that setup should upgrade.
Related event: Sentence Transformers 5.6.1 Fixes Flash Attention Embedding Regression(3 posts)→
More from Infra
- a16z backs Etched’s bet that AI inference will define the next computing era — a16z · 2026-07-23
- Google’s capex is on track to nearly triple in two years, sparking payback questions — SumitGup · 2026-07-23
- JADEPUFFER ransomware is targeting AI pipelines without using a zero-day — TechNadu · 2026-07-23
- Google TPU revenue may be around $1B to $1.5B, with outside customers to watch — BenBajarin · 2026-07-23
- Most of Google’s bought GPUs are still in warehouses, not datacenters — SumitGup · 2026-07-23
- Databricks and Microsoft extend their AI partnership through the 2030s — jefrankle · 2026-07-23