All issues

Search results

Vulkan MoE cache: 512 KiB min-expert default silently excludes small-expert MoE models (Qwen3.6-35B-A3B IQ2_XS, 296 KiB/expert)

Status: Open.

#365 In TheTom/llama-cpp-turboquant;

· Scintsz opened on Sep 9, 2026

SYCL: TurboQuant flash-attn decode (vec) kernels fail JIT on Arc Pro B60 (G21, 0xe211) with head_dim 256 models - pp OK, tg aborts

Status: Open.

#361 In TheTom/llama-cpp-turboquant;

· fantaskiss opened on Sep 7, 2026

Windows MSVC LNK1179 duplicate COMDAT in ggml-cuda.cu.obj

Status: Open.

#358 In TheTom/llama-cpp-turboquant;

· Hansonste opened on Sep 6, 2026

Eval bug: First decode after prefill runs 2-45x slower: HIP CUDA-graph capture warmup (Qwen3.8-Flash-Next / qwen4exp QSA, gfx1201)

Status: Open.

#356 In TheTom/llama-cpp-turboquant;

· Draco-Lunaris opened on Sep 5, 2026

CUDA: full test-backend-ops sweep segfaults in ggml_backend_cuda_graph_compute at ROLL on GB10 (passes with graphs disabled)

bugSomething isn't workingSomething isn't working

Status: Open.

#347 In TheTom/llama-cpp-turboquant;

· TheTom opened on Sep 3, 2026

Compile bug: Windows release archives link the runner's OpenSSL dynamically and never ship it (still open since #109) — one-line fix: -DLLAMA_BUILD_BORINGSSL=ON

Status: Open.

#346 In TheTom/llama-cpp-turboquant;

· breakerh opened on Sep 3, 2026

--gdn-replay makes qwen35 generate degenerate looping text

Status: Open.

#345 In TheTom/llama-cpp-turboquant;

· tmark00 opened on Sep 3, 2026

Proposal: TurboAnchorKV cold prompt KV experiment

Status: Open.

#332 In TheTom/llama-cpp-turboquant;

· TheTom opened on Aug 31, 2026

Vulkan: Polaris wave64 incompatible with hard-coded subgroup size 32

Status: Open.

#330 In TheTom/llama-cpp-turboquant;

· gaigher opened on Aug 31, 2026

Eval bug: llama-server OOM Over Time

Status: Open.

#325 In TheTom/llama-cpp-turboquant;

· shizhx opened on Aug 28, 2026

Eval bug: llama_model_load: error loading model: missing tensor 'blk.64.ssm_conv1d.weight'

Status: Open.

#323 In TheTom/llama-cpp-turboquant;

· JohnZhuYX opened on Aug 26, 2026

Misc. bug: failed to initialize router models: option 'reasoning-effort' not recognized

Status: Open.

#317 In TheTom/llama-cpp-turboquant;

· Jackal830 opened on Aug 25, 2026