Search results

Eval bug: --fit-target projection miscalculation with MTP on ROCm

bug-unconfirmed

Status: Open.

#26038 In ggml-org/llama.cpp;

· eaman opened on Jul 23, 2026

compile llama.cpp with cuda-13 for A100

Status: Open.

#26034 In ggml-org/llama.cpp;

· snailwei opened on Jul 23, 2026

Eval bug: Qwen3.6-35B-A3B-Q8_0.gguf multiple clients concurrently produce garbled output b9927 above（b9910 is ok）

bug-unconfirmed

Status: Open.

#26031 In ggml-org/llama.cpp;

· imweijh opened on Jul 23, 2026

Eval bug: GLM-5.2 (glm_moe_dsa) dense-MLA CUDA path produces subtly corrupted output for ANY real transformer layer offloaded to GPU (partial coherent text mixed with garbage)

bug-unconfirmed

Status: Open.

#26027 In ggml-org/llama.cpp;

· maninara opened on Jul 23, 2026

b10091: ~35% generation speed regression on 8GB VRAM — 3 root causes identified (fit+no-mmap incompatibility, GPU not boosting, CPU MoE regression) + 30+ tuning experiments

Status: Open.

#26025 In ggml-org/llama.cpp;

· Sensenkawa opened on Jul 23, 2026

Feature Request: Next generation samplers that should be able to beat top-n sigma and other SOTA samplers

enhancementNew feature or requestNew feature or request

Status: Open.

#26020 In ggml-org/llama.cpp;

· Diablo-D3 opened on Jul 22, 2026

Feature Request: Add ubuntu CUDA 13 release binaries (x64/arm64)

enhancementNew feature or requestNew feature or request

Status: Open.

#26019 In ggml-org/llama.cpp;

· anvme opened on Jul 22, 2026

Eval bug: When I run Gemma 4e4b with MTP, it doesn't work and crashes to cmd

bug-unconfirmed

Status: Open.

#26017 In ggml-org/llama.cpp;

· felipemolinafacundo-del opened on Jul 22, 2026

[SYCL] Low generation throughput on Intel Arc compared to Vulkan backend

enhancementNew feature or requestNew feature or request

Status: Open.

#26010 In ggml-org/llama.cpp;

· flysio opened on Jul 22, 2026

Feature Request: server: add option to concatenate prompts on /completion endpoint

enhancementNew feature or requestNew feature or request

Status: Open.

#26008 In ggml-org/llama.cpp;

· uteki-rain opened on Jul 22, 2026

Eval bug: server -np 4 --kv-unified returns other requests' responses verbatim on integrated HIP GPU (gfx1151) — bisected to c7d87229

Status: Open.

#25992 In ggml-org/llama.cpp;

· walcz-de opened on Jul 22, 2026

Eval bug: gemma4 peg template intermittently unparseable with long multi-line tool-call string arguments

Status: Open.

#25986 In ggml-org/llama.cpp;

· penumbraforge opened on Jul 22, 2026