Fully open release: SGLang integration, H200/B300 kernels and all training details published
francoisfleuret · x · 2026-11-03
In response to a technical question on parallelizability, the team confirms everything is open-sourced: code with SGLang integration and context parallelism, custom kernels for H200 and B300 GPUs, plus full training and inference details. Paper, code, and blog links are provided, with NVIDIA and Nebius as partners.
Related event: Open-Source LSA Sparse Attention Extends Qwen 27B to 10M-Token Context(3 posts)→
More from Infra
- OpenAI's Grid Protectors program lets utilities cut up to 1,000MW from its Georgia data center — rohanpaul_ai · 2026-11-03
- H200 loses money even fully busy: GPU choice beats inference engine for hosting economics — teortaxesTex · 2026-11-03
- Nearly $1 trillion erased from US stock market as tech selloff deepens — IanArawjo · 2026-11-03
- Lambda raising on-demand GPU rates Dec 2026: B300 to $8.99/hr, B200 to $7.49/hr — vasuman · 2026-11-03
- Independent researcher laments: good GPUs keep getting pricier, experiments unaffordable — scaling01 · 2026-11-03
- Compacting Claude Code at 400K instead of 967K cut this dev's cost per request by 43% — Awkward-Let-4628 · 2026-11-03