Search results

Refactor: mma.cuh shall be refactored for AMD

AMD GPUIssues specific to AMD GPUsIssues specific to AMD GPUs

enhancementNew feature or requestNew feature or request

performanceSpeed related topicsSpeed related topics

Review Complexity : HighGenerally require indepth knowledge of LLMs or GPUsGenerally require indepth knowledge of LLMs or GPUs

Status: Open.

#17856In ggml-org/llama.cpp;

· zhang-hui-yulo opened on Dec 8, 2025

Eval bug: Qwen 3 Next CUDA poor performance

bugSomething isn't workingSomething isn't working

CUDARelated to the CUDA backendRelated to the CUDA backend

performanceSpeed related topicsSpeed related topics

Status: Open.

#17822In ggml-org/llama.cpp;

· fuutott opened on Dec 6, 2025

Efficient inference using llama-mtmd-cli for high resolution images with reduced GPU VRAM usage

enhancementNew feature or requestNew feature or request

needs evalNeeds maintainer evaluation for inclusion viabilityNeeds maintainer evaluation for inclusion viability

performanceSpeed related topicsSpeed related topics

Status: Open.

#17801In ggml-org/llama.cpp;

· deepshnv opened on Dec 5, 2025

Eval bug: Request for Qwen3-Next-80B-A3B Vulkan Inference Optimization

enhancementNew feature or requestNew feature or request

modelModel specificModel specific

performanceSpeed related topicsSpeed related topics

Status: Open.

#17751In ggml-org/llama.cpp;

· engrtipusultan opened on Dec 4, 2025

Misc. bug: [SYCL][Intel GPU] GPT-OSS 20B not loading fully to VRAM with -ngl 99 (A770)

performanceSpeed related topicsSpeed related topics

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

Status: Open.

Feature

#17643In ggml-org/llama.cpp;

· savvadesogle opened on Dec 1, 2025

Misc. bug: Vulkan's performance degradation(TG) on A770 from b7189 and FA problem

bugSomething isn't workingSomething isn't working

need feedbackTesting and feedback with results are neededTesting and feedback with results are needed

performanceSpeed related topicsSpeed related topics

regressionA regression introduced in a new build (something that was previously working correctly)A regression introduced in a new build (something that was previously working correctly)

Status: Open.

Bug

#17628In ggml-org/llama.cpp;

· savvadesogle opened on Nov 30, 2025

Eval bug: unexpected speed fluctuation

bug-unconfirmed

performanceSpeed related topicsSpeed related topics

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

Status: Open.

#17613In ggml-org/llama.cpp;

· eiffel31 opened on Nov 30, 2025

Eval bug: wrong default threads in llama-bench

good first issueGood for newcomersGood for newcomers

low severityUsed to report low severity bugs in llama.cpp (e.g. cosmetic issues, non critical UI glitches)Used to report low severity bugs in llama.cpp (e.g. cosmetic issues, non critical UI glitches)

performanceSpeed related topicsSpeed related topics

Status: Open.

#17611In ggml-org/llama.cpp;

· GermanAizek opened on Nov 30, 2025

[ANALYSE] wrong perf on quantized model with rocm 7.10/7.11

performanceSpeed related topicsSpeed related topics

RoCMIssues related to the RoCM backendIssues related to the RoCM backend

Status: Open.

#17596In ggml-org/llama.cpp;

· Djip007 opened on Nov 29, 2025

Eval bug: failed to restore kv cache on gpt-oss:120b with parallel processing

bugSomething isn't workingSomething isn't working

performanceSpeed related topicsSpeed related topics

server

Status: Open.

Bug

#17527In ggml-org/llama.cpp;

· lostmsu opened on Nov 26, 2025

kv-cache : improve defrag logic

enhancementNew feature or requestNew feature or request

performanceSpeed related topicsSpeed related topics

roadmapPart of a roadmap projectPart of a roadmap project

Status: Open.

#13497In ggml-org/llama.cpp;

· ggerganov opened on May 13, 2025

llama : speed-up grammar sampling

performanceSpeed related topicsSpeed related topics

refactoringRefactoringRefactoring

roadmapPart of a roadmap projectPart of a roadmap project

Status: Open.

#4218In ggml-org/llama.cpp;

· ggerganov opened on Nov 25, 2023