Cherenkov engine hits 8-22 tok/s Qwen3.8-Flash-Next on a 32GB M4 MacBook Air
alfredr · reddit · 2026-09-11
- Developer released Cherenkov, an open-source inference engine for Apple Silicon that runs Qwen3.8-Flash-Next (Q4-ish mixed precision) at 8-22 tok/s using only 21GB of allocations — claimed record for memory-constrained inference of the model.
- Instead of loading the full model, it keeps a bounded working set of experts in unified memory, uses a one-layer lookahead to predict which experts are needed next, and prefetches them via SSD reads, with optional mixed-precision execution.
More from Infra
- Anthropic's S-1 Reveals Over $600B in Infrastructure Commitments Across Google, Amazon, Microsoft — cedric_chee · 2026-10-26
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26
- Anthropic's disclosed compute commitments: $114.5B Google, $161.2B Broadcom top the list — BenBajarin · 2026-10-26
- SpaceX strikes $40B deal with Apollo Global to lease AI chips and data center hardware — Polymarket · 2026-10-26
- SpaceX AI revenue up 830%, Claude designs 318K antivenom proteins in 64 hours — PeterDiamandis · 2026-10-26