Search results

Misc. bug: SYCL backend not working on Intel Arc B580 (Windows 10), only Vulkan works

bug-unconfirmed

SYCLhttps://en.wikipedia.org/wiki/SYCL - GPU programming languagehttps://en.wikipedia.org/wiki/SYCL - GPU programming language

windowsIssues specific to WindowsIssues specific to Windows

Status: Open.

#20169 In ggml-org/llama.cpp;

· k3bc999 opened on Mar 6, 2026

Misc. bug: Parsed model names in model selector can be ambiguous

bug-unconfirmed

Status: Open.

#20165 In ggml-org/llama.cpp;

· rtpt-erikgeiser opened on Mar 6, 2026

Eval bug: Tool calling may repeatedly fail under long context when a tool has multiple optional parameters (Qwen3.5-35B, Qwen3-Coder-Next)

bug-unconfirmed

Status: Open.

#20164 In ggml-org/llama.cpp;

· morirain opened on Mar 6, 2026

Misc. bug: GBNF's <think> not working

bug-unconfirmed

Status: Open.

#20163 In ggml-org/llama.cpp;

· fiesh opened on Mar 6, 2026

Eval bug: Codex, 'type' of tool must be 'function'

bug-unconfirmed

Status: Open.

#20156 In ggml-org/llama.cpp;

· chigkim opened on Mar 6, 2026

Eval bug: forcing full prompt re-processing due to lack of cache data in Qwen3.5 27B

bug-unconfirmed

Status: Open.

#20153 In ggml-org/llama.cpp;

· lostmsu opened on Mar 6, 2026

Eval bug: CUDA crash on tensor copy

bugSomething isn't workingSomething isn't working

CUDARelated to the CUDA backendRelated to the CUDA backend

Status: Open.

#20146 In ggml-org/llama.cpp;

· pwilkin opened on Mar 5, 2026

Eval bug: Hybrid Mamba/Attention model (Qwen3.5-27B) — SSM state checkpoint causes persistent performance degradation on VRAM-constrained GPUs

bug-unconfirmed

Status: Open.

#20145 In ggml-org/llama.cpp;

· hp1337 opened on Mar 5, 2026

Eval bug: Thread sanitizer reports data race for llama-completion

bug-unconfirmed

Status: Open.

#20144 In ggml-org/llama.cpp;

· JohannesGaessler opened on Mar 5, 2026

Misc. bug: ggml_metal_synchronize error: Mac M4 Pro Tahoe 26.3 Crash due to kIOGPUCommandBufferCallbackErrorInnocentVictim

bug-unconfirmed

Status: Open.

#20141 In ggml-org/llama.cpp;

· thehybridtechnician opened on Mar 5, 2026

Eval bug: Router mode <code>--models-max</code> not enforced under concurrent requests (TOCTOU race in unload_lru)

bug-unconfirmed

Status: Open.

#20137 In ggml-org/llama.cpp;

· BenRacicot opened on Mar 5, 2026

Misc. bug: It's taking a long time for the download or model to load.

bug-unconfirmed

Status: Open.

#20136 In ggml-org/llama.cpp;

· Red4Hack opened on Mar 5, 2026