Pull Requests ggml-org/ggml

CUDA/HIP: honor GGML_PREC_F32 in the flash-attention tile kernel

#1536 opened 2026-06-10 14:45 by RapidMark

sycl: fix soft_max_f32 max reduction

#1534 opened 2026-06-10 09:43 by someoneinjd

vulkan: add e4m3 (fp8) matmul for AMD RDNA4 — portable dequant default + opt-in native fp8

#1528 opened 2026-06-05 03:35 by RapidMark

metal : graceful failure instead of GGML_ABORT

#1527 opened 2026-06-03 10:49 by fiorelorenzo

perf(hip): enable -funsafe-math-optimizations for the ROCm backend

#1526 opened 2026-06-02 21:11 by RapidMark

ggml-cpu : fix soft_max_back wrong output when dst aliases src1 (y)

#1521 opened 2026-05-28 19:35 by devYRPauli

ggml : assert b->ne[3] == 1 in ggml_conv_transpose_2d_p0 (#1448)

#1520 opened 2026-05-28 19:17 by devYRPauli

ggml-cpu : fix rms_norm_back wrong output under in-place aliasing (#1491)

#1519 opened 2026-05-28 19:05 by devYRPauli

tests : add OLMoE-sized K-quant shapes to test_mul_mat_id (ref #1506)

#1518 opened 2026-05-28 18:31 by devYRPauli

Fix build and test running for Windows MSVC

#1512 opened 2026-05-26 22:16 by MaxwellGengYF

docs(gguf): extend Encoding slot to support percentage-mixed recipes

#1489 opened 2026-05-13 09:43 by mishig25

ggml-cpu/riscv: gate cpu-riscv64 backend on Zv* sub-extensions

#1475 opened 2026-05-02 23:39 by mikey

opencl: add Adreno xmem attention fast path

#1461 opened 2026-04-19 12:19 by happyyzy

Update GGUF description to include full name

#1457 opened 2026-04-02 21:20 by mmontag

fix: gate AVX/AVX-512/AMX on OS XSAVE support via XGETBV

#1435 opened 2026-03-11 09:01 by Mattbusel

ggml-cpu: BMI2 is only available on amd64

#1392 opened 2025-11-26 18:15 by candrews

metal: implement cross-entropy and count-equal ops for MNIST

#1390 opened 2025-11-22 07:02 by xi-guo-0

metal: accelerated conv2d

#1384 opened 2025-11-11 18:38 by bghira

ggml-cpu : Add GGML_CPU_FFAST_MATH for sine autovectorization

#1243 opened 2025-05-28 06:26 by danielzgtg

Temporary fix for rms norm backward on CPU.

#1197 opened 2025-04-22 21:15 by Alkamist

Update example readme/shell output for gpt-2-backend

#1159 opened 2025-03-26 20:59 by rememberlenny

Add an option not to abort on cuda OOM

#1110 opened 2025-02-11 20:43 by WilliamTambellini

ggml : add `ggml_fft` and `ggml_ifft` operator

#1105 opened 2025-02-06 07:36 by ilhamsyahids

ggml-cpu: use lookuptable for ggml op and parallelized some of the memcpy memset etc. calls before ggml_barriers

#1101 opened 2025-02-04 05:20 by Kamayuq

CUDA: update compilation flags for improved performance

#1099 opened 2025-02-03 18:02 by royshil

ggml-cpu: Add bounds checking in `make_block_q4_0x4` function

#1095 opened 2025-02-02 18:50 by ilhamsyahids

GGUF: ggml backend support for writing tensor data

#1033 opened 2024-11-30 10:34 by JohannesGaessler

Add conv_transpose_1d_gemm

#940 opened 2024-08-28 22:30 by smeso

feat: implement backward computation for more operators

#921 opened 2024-08-12 20:11 by Ronsor

feat: adds new OP ggml_unfold_1d

#867 opened 2024-06-21 20:49 by balisujohn