@badlogicgames We have checked exceptions at home
export type Result<T, E extends Error = Error> =
| { ok: true; value: T; error?: never }
| { ok: false; error: E; value?: never };
@steipete I used just this /goal: Use a fresh-context subagent to review the current code. Then fix issues found. Document deliberate considerations and tradeoffs in docs/ARCHITECTURE.md, track progress with backlog CLI only, and repeat until a fresh reviewer returns ALL POSITIVE.
@jundotkim@Prince_Canuma@bstnxbt Thank you for diligently and quickly approving my contributions :)
I am playing around with ds4 by @antirez a lot, would it make sense to include it like mlx like a inference core for deepseek?
About M5 support, if you are at a company that is big enough to send me an M5 Max 128GB Mac, consider it because I can't support hardware I don't own. I have stable access to my M3 Max (mine), to an M3 Ultra + DGX Spark (stable access via Catania Hacklab). Is your company interested in local inference in high end Macs? Send me the hardware :D
@mitsuhiko@antirez oh nice. is this runnig 2 bit quant? tried it when deepseek-flash -q2 was brand new with oMLX - had unusable tool calling behavior - is this potentially fixed when using this new ds4 runtime?
@badlogicgames i use https://t.co/8d5YcOEEbr - works well for nvidia, oMLX, LM studio and others that support the v1/models endpoint - but it only enumerates, no detailed specs sadly yet.
@thsottiaux I can't move sessions to different projects/folders.. Annoying for those important foundational threads that spin off a new idea/project. Also multi machine sync.