All pull requests

Pull requests

• vulkan: add dedicated PQ2_0 mat-vec and BC-250 batch tuning

  ggmltestingVulkan
  #281·renovys opened Sep 26, 2026·

  ·

• metal : static-K32 tensor mul_mm for full-tile Q1_0 products (GGML_METAL_Q1_MM_K32=1)

  Apple Metalggmltesting
  #280·jasontitus opened Sep 26, 2026·

  ·

• metal : opt-in PQ2_0 mat-vec for two columns (GGML_METAL_PQ2_MULTICOL=1)

  Apple Metalggmltesting
  #279·jasontitus opened Sep 26, 2026·

  ·

• sycl: dequantize PTQ1_0 and PQ2_0 prompt batches to FP16 (2-3x prompt processing)

  ggmlSYCL
  #278·dwymark opened Sep 26, 2026·

  ·

• metal : PTQ1_0 multi-column mat-vec for 5-8 columns

  Apple Metalggmltesting
  #277·jasontitus opened Sep 26, 2026·

  ·

• qwen35 : in-place recurrent state rows for plain decode (GGML_GDN_ROWS_PLAIN=1)

  model
  #276·jasontitus opened Sep 26, 2026·

  ·

• metal : split graphs into 4 command buffers on iOS

  Apple Metalggml
  #275·jasontitus opened Sep 26, 2026·

  ·

• vulkan: extend PTQ1_0 fast integer-dot path to INTEL_XE1 iGPUs (Arrow Lake-H)

  ggmlVulkan
  #274·rriley-quorum opened Sep 25, 2026·

  ·
  1

• Qwen35 tensor split (created with deepseek V4.1 flash)

  ggml
  #268·Yvi71 opened Sep 25, 2026·

  ·
  1

• arm: NEON vec_dot for PTQ1_0 and PQ2_0

  ggmltesting
  #265·kiljoy001 opened Sep 24, 2026·

  ·
  4

• cuda: bound mmvq rows by the per-slot stride for MUL_MAT_ID

  CUDAggml
  #258·bri-prism opened Sep 24, 2026·

  ·

• ggml-cpu: add NEON, AVX2, AVX and SSSE3 vec_dot for PQ1_0

  ggml
  #255·bri-prism opened Sep 23, 2026·

  ·
  4

• ggml: add PQ1_0, the Q1_0 binary codec at group 64

  Apple MetalconversionCUDAexamplesggmltesting
  #253·bri-prism opened Sep 23, 2026·

  ·
  1

• vulkan: dedicated mul_mat_vec shader for PTQ1_0 (~3x on the dequant path)

  ggmlVulkan
  #252·Preygle opened Sep 23, 2026·

  ·
  3

• ggml-cpu : move repacked weight rows to the NUMA node of the thread that reads them

  ggml
  #251·lenny76 opened Sep 23, 2026·

  ·
  1

• ggml-cpu : AVX-512 VNNI and AVX2 vec_dot for PTQ1_0 on x86 (4x token generation)

  ggml
  #250·lenny76 opened Sep 23, 2026·

  ·
  4

• ggml-cpu : convert each mul_mat src1 row on a single thread (faster token generation)

  ggml
  #249·lenny76 opened Sep 23, 2026·

  ·
  1

• ggml-cpu : split GET_ROWS and CONCAT f32 work when there are few rows

  ggml
  #246·lenny76 opened Sep 23, 2026·

  ·
  1

• common: sanitize malformed UTF-8 to U+FFFD with retry parse

  testing
  #244·renovys opened Sep 23, 2026·

  ·

• Prism b10683 Ganglion Helios port multi slot concurrency speedup from kernel reuse

  CUDAggmltesting
  #243·synchronic1 opened Sep 22, 2026·

  ·
  3

• fix(grammar): an empty schema accepts any value, not only an object

  #234·dongtran16092006 opened Sep 21, 2026·

  ·

• fix(chat): accept </parameter> without the surrounding newlines

  #232·dongtran16092006 opened Sep 21, 2026·

  ·

• vulkan: tune the PQ2_0 mat-vec for AMD BC-250 (gfx1013, RDNA1)

  ggmltestingVulkan
  #229·renovys opened Sep 21, 2026·

  ·
  2

• gguf : guard legacy Q2_0 layout hint against non-divisible rows

  ggmltesting
  #227·AntonioRossi opened Sep 20, 2026·

  ·
  2

• cpu: fall back to generic kernels for tensors without a matching repacker

  ggml
  #226·psylligent opened Sep 20, 2026·

  ·
  5