@github.awesome: Swiftlet runs 35B and 80B Qwen mixture-of-experts models on Apple devices by keeping the dense core in memory and streaming routed experts from SSD. QPack files turn each expert fetch into one read, while caching and Metal kernels handle inference. The maintainer reports the 35B model using about 2.5 gigabytes of RAM on an iPhone 17 at roughly one token per second.#github #opensource

GitHub Awesome

Open In TikTok:

Region: US

Tuesday 04 August 2026 13:34:18 GMT