All pull requests

Pull requests

• ci : force python 3.12 on ubuntu 26.04

  devops
  #31222·CISC opened Oct 26, 2026Member·

  ·

• cpu: support for dynamic reshape/splitting for large single-row ops

  ggml
  #31218·max-krasnyansky opened Oct 26, 2026Member·

  ·

• ggml-rpc: fix integer overflow in ggml_nbytes() leading to OOB read/w…

  ggml
  #31217·omarrrfds opened Oct 26, 2026·2 tasks done·

  ·
  1

• tests : fix multi-seq split replay and partial-state cases for trained models

  testing
  #31216·ggerganov opened Oct 26, 2026Member·

  ·

• ui: List org-less models first in the model selector

  server/ui
  #31215·allozaur opened Oct 26, 2026Contributor·

  ·
  1

  1

• ui: Fix draft for new message disappearing after using "Add System Message" action

  merge readyserver/ui
  #31214·allozaur opened Oct 26, 2026Contributor·

  ·
  1

  1

• ui: Highlight reasoning toggle on hover and align it with model selector

  merge readyserver/ui
  #31212·allozaur opened Oct 26, 2026Contributor·

  ·
  1

• common: use physical cores for thread count when -1 is specified

  ggml
  #31209·YuvrajSinghBhadoria2 opened Oct 26, 2026·

  ·
  2

• Remove check for host waits in fattn-mkl from graph checks

  ggmlSYCL
  #31208·malsbat opened Oct 26, 2026Contributor·

  ·

• ARM NEON: optimize Q4_K quantization and dequantization

  ggml
  #31207·YuvrajSinghBhadoria2 opened Oct 26, 2026·

  ·
  1

• rpc : catch std::length_error in recv_msg

  ggml
  #31204·sohaibharraoui opened Oct 26, 2026Contributor·

  ·

• hexagon : use DMA for REPEAT to avoid stale data after buffer reuse

  ggmlHexagontesting
  #31203·hattori7243 opened Oct 26, 2026·1 task done·

  ·

• Speed up IQ4 quant creation with pre-computed halfway points

  ggml
  #31201·bartowski1182 opened Oct 26, 2026Contributor·

  ·
  2

• llama : remove obsolete n_pad from memory instances

  #31200·ggerganov opened Oct 26, 2026Member·

  ·

• mtmd: add Tencent Hy-SimulMT architecture support

  conversiondocumentationmtmd
  #31198·lihy715 opened Oct 26, 2026·

  ·
  1

• vulkan: store CM1 attention P row-major and reuse its fragments

  ggmlVulkan
  #31197·ravel7524 opened Oct 26, 2026Contributor·

  ·

• vulkan: reuse Q8_0 weights in MMVQ for 9-16 columns

  ggmltestingVulkan
  #31196·ravel7524 opened Oct 26, 2026Contributor·

  ·

• Fix #29854 (token format matching issue)

  testing
  #31193·BlueIrisSky opened Oct 26, 2026·

  ·

• model: add MiniMax-M3.1

  conversionmodel
  #31192·danielhanchen opened Oct 26, 2026Contributor·

  ·
  1

• rpc : reject duplicate nodes in graph_compute

  ggml
  #31190·sohaibharraoui opened Oct 26, 2026Contributor·

  ·
  1

• CUDA: fuse MoE gate/up matmuls and GLU in MMQ

  CUDAggmltesting
  #31188·anujj opened Oct 26, 2026Contributor·

  ·

• ui: use shared confirmation dialog for conversation deletion

  merge readyserver/ui
  #31187·allozaur opened Oct 26, 2026Contributor·

  ·
  1

• feat: external PLE file loading

  conversiondocumentationtesting
  #31185·pwilkin opened Oct 26, 2026Member·

  ·
  16

• GGML-CUDA: Optimized build time of project ggml-cuda

  CUDAggml
  #31182·kmorennv opened Oct 26, 2026Contributor·

  ·
  2

• docs : fix DSpark confidence threshold flag name (--spec-draft-p-min)

  documentation
  #31178·jajamabhijith opened Oct 26, 2026·

  ·
  1

  1