All issues

Search results

Independent RTX 3080 (10GB) measurements + engine-matched quality comparison — Ternary Bonsai 2 27B PTQ1_0 vs conventional 2-bit of the same base

Status: Open.

#283 In PrismML-Eng/llama.cpp;

· afi-lambda opened on Sep 26, 2026

Vulkan BC-250 results: packed FP16 PQ2_0 prefill and GDN state-copy fusion

Status: Open.

#282 In PrismML-Eng/llama.cpp;

· renovys opened on Sep 26, 2026

PTQ1_0 Vulkan fast path (#238) misses Xe3 iGPUs reporting minSubgroupSize == 8 (Intel Arc 140T / Panther Lake)

Status: Open.

#270 In PrismML-Eng/llama.cpp;

· rriley-quorum opened on Sep 25, 2026

Misc. bug: missing execute permission for gguf_dspark_to_dflash.py

Status: Open.

#269 In PrismML-Eng/llama.cpp;

· t-hofmann opened on Sep 25, 2026

CUDA: mixed K/V cache types silently run flash attention on CPU (2x slower generation) unless built with GGML_CUDA_FA_ALL_QUANTS

Status: Open.

#267 In PrismML-Eng/llama.cpp;

· GaspardSapin opened on Sep 24, 2026

CUDA: track Q4 KV-cache prefill slowdown and upstream fix validation

Status: Open.

#266 In PrismML-Eng/llama.cpp;

· khosravipasha opened on Sep 24, 2026

Vulkan PTQ1_0: prefill ceiling ~16 tok/s on fp16-less NVIDIA (GTX 1070/Pascal, driver 535.98) — 6x behind RDNA1 on the same build

Status: Open.

#247 In PrismML-Eng/llama.cpp;

· sundog315 opened on Sep 23, 2026

llama-quantize offers PQ2_0/PTQ1_0 but emits no prism.hadamard.*; the result loads silently and generates garbage

Status: Open.

#242 In PrismML-Eng/llama.cpp;

· christopherrobertbrooks-tech opened on Sep 22, 2026

Report: Ternary kernels work on Volta (sm_70) — token-identical with sm_89, ~50 t/s on a V100

Status: Open.

#240 In PrismML-Eng/llama.cpp;

· christopherrobertbrooks-tech opened on Sep 22, 2026

PTQ1_0 Vulkan path falls back to generic LUT decode on int-dot-less devices (gfx1013)

Status: Open.

#231 In PrismML-Eng/llama.cpp;

· renovys opened on Sep 21, 2026

Feature Request: Please add Linux SYCL build for release.

enhancementNew feature or requestNew feature or request

Status: Open.

#228 In PrismML-Eng/llama.cpp;

· Nocturna22 opened on Sep 20, 2026

Eval bug: Bonsai 2 PQ2_0 produces corrupt logits on Windows HIP gfx1151; CPU --no-repack is coherent

Status: Open.

#223 In PrismML-Eng/llama.cpp;

· ShayGus opened on Sep 20, 2026