Issues

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

  PrismML-Eng/llama.cpp#283 · Filter by author afi-lambda opened on Sep 26, 2026
  2

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

  PrismML-Eng/llama.cpp#282 · Filter by author renovys opened on Sep 26, 2026
  2

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

  PrismML-Eng/llama.cpp#270 · Filter by author rriley-quorum opened on Sep 25, 2026
  2

• Misc. bug: missing execute permission for gguf_dspark_to_dflash.py

  PrismML-Eng/llama.cpp#269 · Filter by author 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

  PrismML-Eng/llama.cpp#267 · Filter by author GaspardSapin opened on Sep 24, 2026

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

  PrismML-Eng/llama.cpp#266 · Filter by author 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

  PrismML-Eng/llama.cpp#247 · Filter by author 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

  PrismML-Eng/llama.cpp#242 · Filter by author christopherrobertbrooks-tech opened on Sep 22, 2026
  1

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

  PrismML-Eng/llama.cpp#240 · Filter by author christopherrobertbrooks-tech opened on Sep 22, 2026
  1

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

  PrismML-Eng/llama.cpp#231 · Filter by author renovys opened on Sep 21, 2026

• Feature Request: Please add Linux SYCL build for release.

  enhancement
  PrismML-Eng/llama.cpp#228 · Filter by author Nocturna22 opened on Sep 20, 2026
  2

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

  PrismML-Eng/llama.cpp#223 · Filter by author ShayGus opened on Sep 20, 2026
  3

• Eval bug: Binary release CUDA 13 version have always 'Segmentaion Fault'

  PrismML-Eng/llama.cpp#222 · Filter by author qkboo opened on Sep 20, 2026
  1

• Bonsai 2 27B (PQ2_0): --split-mode tensor/row unusable — 4 crash sites root-caused, safe fallback defeats tensor-split's VRAM purpose

  PrismML-Eng/llama.cpp#213 · Filter by author leduclinh7141 opened on Sep 19, 2026
  1

• Bug: Support consumer RDNA2 (gfx1030) under ROCm/HIP backend for Ternary-Bonsai-27B

  PrismML-Eng/llama.cpp#264 · Filter by author bumblebeejoe opened on Sep 19, 2026
  5

• --spec-type ngram-* silently no-ops on Ternary-Bonsai-2 (hybrid GDN): flag accepted, never activates, zero logging

  PrismML-Eng/llama.cpp#203 · Filter by author lyb618888-bojian opened on Sep 19, 2026
  5

• PTQ1_0 generation is ~9x slower on Vulkan than CUDA on the same GPU; Q4_K_M on the same card is 0.80x

  PrismML-Eng/llama.cpp#201 · Filter by author blanham opened on Sep 19, 2026
  1

• ptq1_0 Vulkan matvec kernel ~6x slower than tq2_0 at n=1 (195 vs 1080 GFLOPS, RX 6600)

  PrismML-Eng/llama.cpp#259 · Filter by author gatrivi opened on Sep 18, 2026
  1

• RTX 5090 (sm_120) prebuilt linux tarballs: erratic/catastrophic decode on driver 570 hosts; build-config + data inside

  PrismML-Eng/llama.cpp#199 · Filter by author rafiksalama opened on Sep 18, 2026
  2

• Vulkan: PTQ1_0 kernel hangs GPU (device lost, xe driver job timeout) on Intel Arc/Battlemage at longer prompts

  PrismML-Eng/llama.cpp#192 · Filter by author flatlinebb opened on Sep 18, 2026
  4

• Perf bug: q5_0 KV cache ~8x slower than f16/q8_0/q4_0 on long-context decode (CUDA, Blackwell)

  PrismML-Eng/llama.cpp#191 · Filter by author hot-YUser opened on Sep 18, 2026
  1

• Vulkan: PTQ1_0 decode is 6-119x slower than necessary (serial trit recurrence) + no MUL+FWHT fusion

  PrismML-Eng/llama.cpp#186 · Filter by author MrFadiAi opened on Sep 18, 2026
  2

• PTQ1_0 Vulkan kernel (4d583a9f3, committed as untested) — confirmed correct on gfx1152, but slow

  PrismML-Eng/llama.cpp#185 · Filter by author c2p-cmd opened on Sep 18, 2026
  4

• Feature Request: Support for CUDA 13.4

  enhancement
  PrismML-Eng/llama.cpp#184 · Filter by author igorlakic opened on Sep 18, 2026

• Segfault in CPU repack loading Ternary-Bonsai-2-27B-PQ2_0.gguf (output.weight)

  PrismML-Eng/llama.cpp#180 · Filter by author Mattwmaster58 opened on Sep 18, 2026
  11