Search results

Switching between global profiles with different parameters, does not affect the form fields

Status: Open.

#1816 In jundot/omlx;

· merken opened on Jun 11, 2026

[BUG] omlx serve hangs at startup banner when spawned by launchd / brew-services (works in a shell/tmux); stale server process blocks restart

Status: Open.

#1814 In jundot/omlx;

· igor opened on Jun 11, 2026

Community Intelligence Benchmark

Status: Open.

#1808 In jundot/omlx;

· rink-lab opened on Jun 11, 2026

Support diffusiongemma

Status: Open.

#1805 In jundot/omlx;

· black-snow opened on Jun 10, 2026

Allow qwen3_omni_moe models to load.

Status: Open.

#1802 In jundot/omlx;

· monroewilliams opened on Jun 10, 2026

Qwen3.6 VLM build: any second concurrent request crashes engine loop (cache_offset scalar/array mismatch in mlx_vlm)

Status: Open.

#1800 In jundot/omlx;

· gcp0703 opened on Jun 10, 2026

Regression risk from #1764: MLX fused full (prefill) SDPA only supports head_dim 64/80/128 — estimator now under-estimates head_dim=256 prefill transients, disabling the 413 guard

Status: Open.

#1797 In jundot/omlx;

· hojin12312 opened on Jun 10, 2026

Memory guard counts serialized hot-cache CPU bytes as GPU pressure — prefill safety cap shrinks 1:1 as the hot cache fills

Status: Open.

#1796 In jundot/omlx;

· hojin12312 opened on Jun 10, 2026

Hitting the context wall with Gemma 4

Status: Open.

#1794 In jundot/omlx;

· vapnen opened on Jun 10, 2026

TurboQuant: chunked-prefill path skips KV convert, storing fp16 blocks into TQ prefix chains → permanent 0% cache hit ("'array' object has no attribute 'norms'")

Status: Open.

#1793 In jundot/omlx;

· hojin12312 opened on Jun 10, 2026

QWEN3.5 small models loop and print "!"

Status: Open.

#1790 In jundot/omlx;

· undici77 opened on Jun 10, 2026

gemma4 performance has significantly deteriorated

Status: Open.

#1789 In jundot/omlx;

· Skjz opened on Jun 10, 2026