All issues

Search results

Eval bug: gemma4 thinking starts to emit progressivly long trailing garbage

bug-unconfirmed

Status: Open.

#28827 In ggml-org/llama.cpp;

· xxxajk opened on Sep 13, 2026

Eval bug: segmentation fault with -sm tensor and DeepSeek V4 Flash

bug-unconfirmed

Status: Open.

#28826 In ggml-org/llama.cpp;

· Volunteer-1 opened on Sep 13, 2026

Misc. bug: server: (router) --api-key is not forwarded to child instances while --api-key-file is

bug-unconfirmed

Status: Open.

#28820 In ggml-org/llama.cpp;

· Penguin-KSU opened on Sep 12, 2026

Compile bug: Linker error for bin/ggml-rpc-server, LLVM 23 and HIP

bug-unconfirmed

Status: Open.

#28814 In ggml-org/llama.cpp;

· mattmcadoo opened on Sep 12, 2026

Vulkan: enabling all supported device features (incl. protectedMemory) kills the process on Huawei Maleoon GPU

bug-unconfirmed

Status: Open.

#28812 In ggml-org/llama.cpp;

· ihumster opened on Sep 12, 2026

Eval bug: Vulkan n_ubatch 512 hangs the GPU ring on gfx1010 / Navi 10 (384 is fine)

Status: Open.

#28808 In ggml-org/llama.cpp;

· sizeak opened on Sep 12, 2026

Eval bug: Vulkan CONCAT (SSM conv input) hangs the GPU ring on gfx1010 / Navi 10

Status: Open.

#28807 In ggml-org/llama.cpp;

· sizeak opened on Sep 12, 2026

Eval bug: qwen4exp (Qwen3.8-Flash-Next) on Metal: decode emits 1 token then EOS at long context — silent empty output, stochastic at threshold, threshold moves with model quant / KV quant / n_ctx

Status: Open.

#28805 In ggml-org/llama.cpp;

· dongzhang777-source opened on Sep 12, 2026

Feature Request: Support for fraunhofer-iis/elmod-2.7b-it

enhancementNew feature or requestNew feature or request

Status: Open.

#28804 In ggml-org/llama.cpp;

· spotbot2k opened on Sep 12, 2026

Unexpected Subjectivity Emergence in Low-Precision Quantization

bug-unconfirmed

Status: Open.

#28801 In ggml-org/llama.cpp;

· zhichiamd opened on Sep 12, 2026

Eval bug: SIGSEGV in ggml_compute_forward_flash_attn_ext_tiled (CPU) on long prompts — reproduces across multiple models/commits, workaround is --flash-attn off

Status: Open.

#28798 In ggml-org/llama.cpp;

· fluffyxenia opened on Sep 12, 2026

Eval bug: MTP (--spec-type draft-mtp) causes ~57x prefill slowdown on a self-compiled Windows build (MSVC + CUDA 12.8)

Status: Open.

#28790 In ggml-org/llama.cpp;

· lifeidle opened on Sep 11, 2026