Search results

[Bug]: Nix flake fails to run tests, doesn't correctly package engine and server

Status: Open.

#345 In JustVugg/colibri;

· LordMZTE opened on Jul 17, 2026

Kimi K3 (2.8T MoE, 16-of-896 routed experts) — candidate for the streaming architecture?

Status: Open.

#344 In JustVugg/colibri;

· fermionoid opened on Jul 17, 2026

[Feature]: Support Anthropic protocol endpoint

Status: Open.

#343 In JustVugg/colibri;

· yurivict opened on Jul 17, 2026

[Bug]: glm stays at 3000% CPU after the model finished the answer.

Status: Open.

#341 In JustVugg/colibri;

· yurivict opened on Jul 17, 2026

telemetry: hits conflates pin-tier and ecache hits — split them so eviction-policy changes are measurable

Status: Open.

#336 In JustVugg/colibri;

· KingIcyCreamProjects opened on Jul 17, 2026

sampling: dist_build full-qsorts all 151K vocab entries per sampled token — partial top-p select is ~7x cheaper (measured)

Status: Open.

#335 In JustVugg/colibri;

· KingIcyCreamProjects opened on Jul 17, 2026

fmt=4 CUDA attention crash — FIXED (#298); benchmark tracker + follow-ups (PROFILE accounting, cache warmth)

Status: Open.

#334 In JustVugg/colibri;

· woolcoxm opened on Jul 17, 2026

[Feature]: Pre-fetching MoE experts

Status: Open.

#333 In JustVugg/colibri;

· dnnspaul opened on Jul 16, 2026

New int4-g64 container is ~12% more resource-hungry than per-row int4; possible VRAM-crash correlation (unconfirmed)

Status: Open.

#326 In JustVugg/colibri;

· woolcoxm opened on Jul 16, 2026

Always using one cpu core

Status: Open.

#325 In JustVugg/colibri;

· liangstein opened on Jul 16, 2026

convert_olmoe.py: --ebits is parsed but never used — every conversion silently emits int8, blocking the int4-on-OLMoE measurement

Status: Open.

#323 In JustVugg/colibri;

· lornecodes opened on Jul 16, 2026

Make the runtime path Python-free by porting the CLI and API gateway to Go

Status: Open.

#310 In JustVugg/colibri;

· Magnet-js opened on Jul 16, 2026