vLLM 0.25.0 Released
vllm_project · x · 2026-07-12
vLLM releases v0.25.0, featuring 558 commits and 232 contributors, 64 of whom are new.
Major changes include:
- Model Runner V2 becomes the default execution path for all dense models.
- Legacy PagedAttention implementation removed; V1/MRv2 is now the standard path.
- Transformers backend speed increased to near native vLLM levels.
- Added unified Streaming Parser Engine.
- Supports universal speculative decoding (TLI) across heterogeneous vocabularies, with DSpark and DFlash drafters added.
- Added/updated multiple models like Hy3, Unlimited OCR, LLaVA-OneVision-2, MOSS-Transcribe-Diarize, openai/privacy-filter, GLM-5, DeepSeek-V3.2, and MiniMax-M3.
More from Infra
- Anthropic S-1 reveals $800B+ in off-books future compute commitments — GaryMarcus · 2026-10-27
- ~42% of Anthropic's ~11GW contracted compute is NVIDIA GPUs — ns123abc · 2026-10-27
- EdgeRock-4B holds KV cache at 72 MiB across 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- RTX 5080 price jumps from $1658 to $2318 in ten days on Amazon — Frizzy-MacDrizzle · 2026-10-27
- ComfyUI on 12GB VRAM: measured peaks show offloading turns 24s into 11 minutes — Impossible-Mail-4619 · 2026-10-27
- Neon rebuilds its docs search entirely on its own stack: Lakebase Search, Functions, AI Gateway — matei_zaharia · 2026-10-27