Search results

metal: reuse K/V in flash-attn vec for spec-decode

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

testingEverything test relatedEverything test related

Status: Open (in progress).

ggml-org/llama.cppnumber 23114

#23114 In ggml-org/llama.cpp;

· forforever73 opened on May 15, 2026

Feat/qlora training

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

examples

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Nvidia GPUIssues specific to Nvidia GPUsIssues specific to Nvidia GPUs

pythonpython script changespython script changes

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

Status: Draft (not ready).

ggml-org/llama.cppnumber 22705

#22705 In ggml-org/llama.cpp;

· srossitto79 opened on May 5, 2026

Feat/backward mul mat

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Nvidia GPUIssues specific to Nvidia GPUsIssues specific to Nvidia GPUs

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

Status: Draft (not ready).

ggml-org/llama.cppnumber 22704

#22704 In ggml-org/llama.cpp;

· srossitto79 opened on May 5, 2026

llama + spec: MTP Support

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

examples

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

modelModel specificModel specific

Nvidia GPUIssues specific to Nvidia GPUsIssues specific to Nvidia GPUs

pythonpython script changespython script changes

server

testingEverything test relatedEverything test related

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

Status: Open (in progress).

ggml-org/llama.cppnumber 22673

#22673 In ggml-org/llama.cpp;

· am17an opened on May 4, 2026

metal : pair rsets_add with rsets_rm on buffer free (fix #22593)

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Status: Open (in progress).

ggml-org/llama.cppnumber 22595

#22595 In ggml-org/llama.cpp;

· joelteply opened on May 1, 2026

ggml-metal: implement async 2D tensor copy functions

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Status: Draft (not ready).

ggml-org/llama.cppnumber 22515

#22515 In ggml-org/llama.cpp;

· ggerganov opened on Apr 29, 2026

cmake: use cmake's install(EXPORT) to track ggml target properties

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Nvidia GPUIssues specific to Nvidia GPUsIssues specific to Nvidia GPUs

OpenCLIssues specific to the OpenCL backendIssues specific to the OpenCL backend

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

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

Status: Open (in progress).

ggml-org/llama.cppnumber 22482

#22482 In ggml-org/llama.cpp;

· SchaichAlonso opened on Apr 28, 2026

ggml-metal: add Metal kernel for ggml_roll

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Status: Open (in progress).

ggml-org/llama.cppnumber 21782

#21782 In ggml-org/llama.cpp;

· stephencox-ict opened on Apr 11, 2026

metal : add CONV_2D_DW (depthwise convolution) support

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

testingEverything test relatedEverything test related

Status: Open (in progress).

ggml-org/llama.cppnumber 21565

#21565 In ggml-org/llama.cpp;

· Sou-ly opened on Apr 7, 2026

metal : add GATED_LINEAR_ATTN op

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

documentationImprovements or additions to documentationImprovements or additions to documentation

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

testingEverything test relatedEverything test related

Status: Open (in progress).

ggml-org/llama.cppnumber 21452

#21452 In ggml-org/llama.cpp;

· TheTom opened on Apr 5, 2026

metal: speed up Qwen3-VL image encoding on large images by ~11%

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Status: Open (in progress).

ggml-org/llama.cppnumber 21443

#21443 In ggml-org/llama.cpp;

· Avidanborisov opened on Apr 4, 2026

Cross-backend profiler

AMD ZenDNNIssues related to the AMD ZenDNN backendIssues related to the AMD ZenDNN backend

Apple Metalhttps://en.wikipedia.org/wiki/Metal_(API)https://en.wikipedia.org/wiki/Metal_(API)

Ascend NPUissues specific to Ascend NPUsissues specific to Ascend NPUs

documentationImprovements or additions to documentationImprovements or additions to documentation

examples

ggmlchanges relating to the ggml tensor library for machine learningchanges relating to the ggml tensor library for machine learning

Hexagon

IBM zDNNissues specific to IBM zDNN Acceleratorissues specific to IBM zDNN Accelerator

Nvidia GPUIssues specific to Nvidia GPUsIssues specific to Nvidia GPUs

OpenCLIssues specific to the OpenCL backendIssues specific to the OpenCL backend

OpenVINO

pythonpython script changespython script changes

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

testingEverything test relatedEverything test related

VulkanIssues specific to the Vulkan backendIssues specific to the Vulkan backend

WebGPU

Status: Draft (not ready).

ggml-org/llama.cppnumber 21160

#21160 In ggml-org/llama.cpp;

· pwilkin opened on Mar 29, 2026