{"full_name":"drumih/turbo-fieldfare","owner":"drumih","name":"turbo-fieldfare","shard":"1","url":"https://github.com/drumih/turbo-fieldfare","description":"Gemma 4 26B-A4B inference in ~2 GB of RAM on any M-series MacBook","homepage":"","topics":["gemma","gpgpu","llm","local-ai","macos","metal","on-device-ai","swift","apple-silicon","gemma4","gemma4-26b-a4b","llm-inference","on-device-llm"],"tag_pages":["apple-silicon","gemma","gpgpu","llm","llm-inference","local-ai","macos","metal","on-device-ai","swift"],"language":"Swift","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/8088294?u=4430fa0dd63dba273a77bb87061eca00287a0a54&v=4","created_at":"2026-07-17T15:57:54Z","pushed_at":"2026-11-14T16:38:01Z","metrics":{"stars":7403,"forks":482,"watchers":35,"issues":87,"prs":115,"releases":25,"issues_open":23,"issues_closed":64,"prs_open":34,"prs_closed":30,"prs_merged":51,"discussions":0},"growth":{},"history":{"dates":["2026-11-16"],"stars":[7403],"forks":[482],"watchers":[35],"issues_open":[23],"prs_open":[34]},"boards":{},"same_owner":[],"issues":[{"number":18,"title":"[Feature]: Claude code integration","comments":7,"created_at":"2026-07-29T17:13:40Z"},{"number":44,"title":"[Feature]: implement other MOE family models","comments":5,"created_at":"2026-08-01T13:07:08Z"},{"number":73,"title":"[Bug]: Unable to trigger prompt caching in OpenAI-compatible local server","comments":4,"created_at":"2026-08-03T13:10:57Z"},{"number":84,"title":"[Bug]: tool-call parse fails with GemmaToolCallParserError.malformed after a long generation","comments":3,"created_at":"2026-08-04T12:20:03Z"},{"number":121,"title":"[Bug]: module 'SwiftUIMacros' not found","comments":2,"created_at":"2026-08-11T19:31:09Z"},{"number":54,"title":"[Feature]: Different model for higher ram models","comments":2,"created_at":"2026-08-02T01:23:33Z"},{"number":48,"title":"[Feature]: Link a community Homebrew tap for the CLI tools and Mac app","comments":2,"created_at":"2026-08-01T18:53:32Z"},{"number":47,"title":"[Bug]: the dictation button in the prompt field does not function","comments":2,"created_at":"2026-08-01T18:03:34Z"}],"discussions":[],"latest_release":{"name":"0.10.0","tag":"0.10.0","published_at":"2026-11-14T16:38:02Z","body":"**Speculative expert prefetch**\n\nExpert prefetch can speed up generation on devices with less RAM.\n\n- Enable **Expert prefetch** in the Mac app, then reload the model. The setting is saved.\n- CLI and server users can enable it with `--early-expert-read router`.\n- Improved reader cleanup during cancellation, errors, and model reuse.\n- Updated benchmarks and experiment documentation with M2 and M5 Pro results.\n\nThanks to [@maxslamdunk](https://github.com/maxslamdunk) for the expert-prefetch implementation in [#187](https://github.com/drumih/turbo-fieldfare/pull/187).\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}