27B model tops 100 tok/s on a MacBook via M5-tuned quantization and speculative decoding
teortaxesTex · x · 2026-09-09
The norpadon team co-designed quantization and speculative decoding around Apple's M5 neural accelerators, getting a 27B model to run at over 100 tokens/s on a MacBook — for now on M5+ chips only. The latest version also shows significantly improved performance, especially on prefill.
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- SpaceX locks in $40B GPU financing deal with Apollo, the largest ever committed — TechCrunch AI · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27