Search results

[BUG] When loading a model, other idle models were not unloaded at the correct time.

Status: Open.

#1867 In jundot/omlx;

· funnygeeker opened on Jun 13, 2026

0.3.8 and 0.4.4: cache hit rate decreased from 95% to 51%.

Status: Open.

#1866 In jundot/omlx;

· TLin-Y opened on Jun 13, 2026

Failing to install OMLX via Homebrew on MacOS 27.0 Beta

Status: Open.

#1865 In jundot/omlx;

· RetepV opened on Jun 13, 2026

thinking_budget: reasoning spills past the close at low budgets (intermittent, correlated with non-English / Cyrillic prompts)

Status: Open.

#1864 In jundot/omlx;

· efortin opened on Jun 13, 2026

Minimax 3 version model type error

Status: Open.

#1862 In jundot/omlx;

· bbongtree1004 opened on Jun 13, 2026

force_sampling pins max_tokens too — per-request max_tokens ignored; intended?

Status: Open.

#1857 In jundot/omlx;

· tlnini-afk opened on Jun 13, 2026

fail to quantize deepseek v4 by v0.4.4rc1

Status: Open.

#1856 In jundot/omlx;

· taozhiyuai opened on Jun 13, 2026

Just in time model eviction

Status: Open.

#1855 In jundot/omlx;

· harish0201 opened on Jun 13, 2026

Model switch in chat settings behaves unexpectedly

Status: Open.

#1848 In jundot/omlx;

· KEO-RSP opened on Jun 12, 2026

Export and import the configration funtion

Status: Open.

#1847 In jundot/omlx;

· abooaboo opened on Jun 12, 2026

Gemma 4 26B (A4B MoE): long agentic contexts trigger parenthesized call:name(...) tool calls; gemma4 parser silently strips them (empty content, no recovery path)

Status: Open.

#1846 In jundot/omlx;

· hugo-valcourt opened on Jun 12, 2026

Proposal: hybrid TurboQuant eligibility (sliding-window models get no TQ at all today) + last-resort mid-prefill KV conversion — removes the cold-prefill memory wall (184K → model limit on Gemma 4 26B / 36 GB)

Status: Open.

#1843 In jundot/omlx;

· hojin12312 opened on Jun 12, 2026