Search results

Eval bug: Gemma 4 MTP Performance Regression Post-merge

bug-unconfirmed

Status: Open.

#24596 In ggml-org/llama.cpp;

· Schrikvis opened on Jun 13, 2026

Misc. bug: Passing non-unique ids to CUDA ggml_mul_mat_id() implementation results in crash

bugSomething isn't workingSomething isn't working

CUDARelated to the CUDA backendRelated to the CUDA backend

wontfixThis will not be worked onThis will not be worked on

Status: Open.

#24591 In ggml-org/llama.cpp;

· fairydreaming opened on Jun 13, 2026

Eval bug: CDNA2 ROCM Performance Issue: QAT Q4 vs Q8

bug-unconfirmed

Status: Open.

#24589 In ggml-org/llama.cpp;

· 65a opened on Jun 13, 2026

[Bug/Performance] Context checkpoint invalidation loops with dynamic RAG prompts (bge-m3) on SWA models (Qwen)

Status: Open.

#24587 In ggml-org/llama.cpp;

· nshields1512-code opened on Jun 13, 2026

Eval bug: llama-server aborts on cudaErrorLaunchTimeout in MTP speculative draft path during long-context generation

bug-unconfirmed

Status: Open.

#24576 In ggml-org/llama.cpp;

· controldev opened on Jun 13, 2026

Eval bug: unknown projector type: locateanything

bug-unconfirmed

Status: Open.

#24567 In ggml-org/llama.cpp;

· yinbo18120 opened on Jun 13, 2026

Eval bug: Vulkan on Intel ARC - Gemma 4 models (especially with mtp) produce unusable, garbled output

bug-unconfirmed

Status: Open.

#24560 In ggml-org/llama.cpp;

· Tech-no-1 opened on Jun 13, 2026

Misc. bug: alloc_tensor_range does not check realloc result (leak + NULL deref on OOM)

Status: Open.

#24557 In ggml-org/llama.cpp;

· ricku777-bear opened on Jun 13, 2026

Eval bug: EAGLE3 with Qwen3.6 (qwen3_5 hybrid) target — missing t_layer_inp hooks, and llama_decode(ctx_dft) rc=-1 once context exceeds ~700 tokens

Status: Open.

#24541 In ggml-org/llama.cpp;

· Zertruermmerdog opened on Jun 12, 2026

Misc: Optimise docker image size

Status: Open.

#24539 In ggml-org/llama.cpp;

· ihatemakinganaccount opened on Jun 12, 2026

Misc. bug: CUDA flash-attention crashes after server sleep because kernel attributes are not restored

bug-unconfirmed

Status: Open.

#24537 In ggml-org/llama.cpp;

· christophkogler opened on Jun 12, 2026

DiffusionGemma on Apple Silicon appears much slower than expected; local Metal experiment improved usability

Status: Open.

#24529 In ggml-org/llama.cpp;

· dbmcco opened on Jun 12, 2026