All issues

Search results

Metal: what promoting 0.34.2 costs, and why the MLX half is probably inert

Status: Open.

#361 In MaxusAI/ollama;

· glennneuber opened on Sep 20, 2026

All platforms: conv_2d im2col demoted to F16 since llama.cpp #23660 — affects every F32-weight vision tower

Status: Open.

#349 In MaxusAI/ollama;

· glennneuber opened on Sep 20, 2026

gfx1151 vision regression root-caused: upstream prop.integrated on HIP — cross-platform verification wanted (and a lead on #17475)

Status: Open.

#313 In MaxusAI/ollama;

· glennneuber opened on Sep 18, 2026

Help wanted (MLX-CUDA): does the gemma4 vision encoder move across the v0.34 fold on a non-Metal backend?

Status: Open.

#312 In MaxusAI/ollama;

· glennneuber opened on Sep 18, 2026

qwen2.5-vl: fp16-accumulate cuBLAS vision GEMMs overflow on specific images — deterministic '?'×31 garbage; one-line env workaround; ≥fp32 accumulation must be the default

Status: Open.

#216 In MaxusAI/ollama;

· glennneuber opened on Aug 26, 2026