35B MoE in 7GB: Mach-1 ships GGUFs and llama.cpp fork for edge devices
pmttyji · reddit · 2026-08-20
SyzygyResearch released GGUF quantizations of Mach-1-Additive-35B plus a custom llama.cpp fork: the 35B MoE fits in roughly 7GB, targeting mobile/edge and low-memory systems, reaching up to 120 t/s on a consumer laptop.
- Two GGUF repos: base and Multimodal variants
- Custom fork: github.com/SyzygyResearch/llama.cpp-mach1
- The team previously teased Laguna S2.1 and Qwen 3.8
The poster also maintains a tracking thread for 1-bit/2-bit/ternary/Bitnet models.
More from Infra
- Help: Building an On-Premise AI Agent for Internal Documents — Remarkable_Mine_1622 · 2026-08-20
- Trail Mate: An offline-first handheld GPS system over LoRa network — tom_doerr · 2026-08-20
- Omarchy and Linux Desktop: Opportunities Fueled by AI — antirez · 2026-08-20
- Microsoft migrates TypeScript repo to Go in major PR merge — wateriscoding · 2026-08-20
- Investor Applies Munger's Three-Basket Rule: Data Centers Are a Clear Yes — RachelVT42 · 2026-08-20
- Benchmark: FP8 Models Run 5x Faster Than GGUF on Low-End Hardware — ROBOTTTTT13 · 2026-08-20