DeepSeek open-sources DeepInfer inference framework, sparks 'open weights' debate
alex_verem · x · 2026-10-21
DeepSeek announced DeepInfer, its fully open-source in-house inference framework that powers DeepSeek-V4.1-Flash at scale with high throughput and low latency, shared to help the community cut compute costs and energy use.
@alexverem pushes back on the 'open source' framing: V4.1-Flash is free to download and use commercially, but training code and data are unavailable and the 700B+ parameter model isn't runnable locally — the accurate term, he argues, is 'open weights.'
More from Infra
- Nextdc Implodes 45% as Angry Residents Cut Power to Melbourne Data Center — AIFlow_ML · 2026-10-21
- NVIDIA teases big Europe push in event intro, photos show — testingcatalog · 2026-10-21
- Jensen Huang takes stage at NVIDIA GTC Berlin, keynote livestreams now — nvidia · 2026-10-21
- PyTorch Foundation quietly adds nine members, Anyscale joins as Platinum — YvesMulkers · 2026-10-21
- Building a $300 PC to run Qwen 27B locally with dual RTX 3060 12GB — advice wanted — Loose_Doubt367 · 2026-10-21
- CoreWeave owes $42B, half of all debt held by the 7 biggest GPU rental firms — FinanceYF5 · 2026-10-21