QVAC-18300: sync with upstream ggml-org/whisper.cpp master (v1.8.4.3 prep)#12

Merged

GustavoA1604 merged 246 commits into

tetherto:masterfrom

mario-rei:feat/upstream-sync-v1.8.4.3

May 20, 2026

Commits on Mar 21, 2026

ruby : fix dangling pointers, memory leak, and SEGV on parallel transcription (#3715)
KitaitiMakoto
authored

Commits on Mar 29, 2026

sycl : fix for untransposed GDA recurrent state (llama/20583)

CISC
authored and
ggerganov
committed
CUDA: GDN hide memory latency (llama/20537)

am17an
authored and
ggerganov
committed
vulkan: fix flash attention dot product precision (llama/20589)

0cc4m
authored and
ggerganov
committed
kleidiai: add data type check to get_tensor_traits (llama/20639)

martin-klacer-arm
and
ggerganov
committed
ehance UPSCALE to support all UT cases (llama/20637)

arthw
authored and
ggerganov
committed
vulkan: allow graphics queue only through env var (llama/20599)

0cc4m
authored and
ggerganov
committed
kleidiai : fix MUL_MAT support for batched (3D) inputs (llama/20620)

jabr
authored and
ggerganov
committed
vulkan: async and event fixes (llama/20518)

0cc4m
authored and
ggerganov
committed
ggml-cpu: fix RVV checks in quants and repacking (llama/20682)

taimur-10x
authored and
ggerganov
committed
ggml-blas: set mkl threads from thread context (llama/20602)

kannon92
authored and
ggerganov
committed
vulkan: disable mmvq on Intel Windows driver (llama/20672)

0cc4m
authored and
ggerganov
committed
hexagon: add neg, exp, sigmoid, softplus ops, cont, repeat ops (llama/20701)

authored and
ggerganov
committed
HIP : ignore return of hipMemAdvise [no ci] (llama/20696)

IMbackK
authored and
ggerganov
committed
ggml-cpu/x86: fix unused changemask warning in repack (llama/20692)

mrshaw01
authored and
ggerganov
committed
Move to no timeout for WaitAny in graph submission to avoid deadlocks in some cases on llvm-pipe backends (llama/20618)

reeselevine
authored and
ggerganov
committed
CANN: support flash attention for head dim not multiple of 16, fix ALiBi slope offset (llama/20031)

noemotiovon
authored and
ggerganov
committed
ggml-webgpu: Add supports for `DIAG` and `TRI` (llama/20664)

yomaytk
authored and
ggerganov
committed
ggml-webgpu: Update the `RMS_NORM` preprocessor and add `L2_NORM` (llama/20665)

yomaytk
authored and
ggerganov
committed
CANN: handle in-place ROPE on non-contiguous f32 tensors (llama/20274)

noemotiovon
authored and
ggerganov
committed
cmake : fix build warning when kleidiai is enabled (llama/20457)

chaxu01
authored and
ggerganov
committed
vulkan: dequantize iq4_xs 4 at a time (llama/20657)

netrunnereve
authored and
ggerganov
committed
ggml webgpu: ops support for qwen3.5 (SET, TRI_SOLVE, SSM_CONV, GATED_DELTA_NET) + GET_ROWS optimization (llama/20687)

reeselevine
authored and
ggerganov
committed
ci : add hip quality check (llama/20430)

authored and
ggerganov
committed
hexagon: add Matrix Extensions (HMX) for Hexagon NPU backend (llama/20693)

authored and
ggerganov
committed
hip: Avoid compiler bug in RDNA code generation during debug builds on Windows (llama/20655)

Exile333
authored and
ggerganov
committed
ggml: guard KleidiAI DOWNLOAD_EXTRACT_TIMESTAMP for cmake < 3.24 (llama/20767)

sundaram123krishnan
authored and
ggerganov
committed
CANN: add BF16 support for core operators (llama/20152)

hipudding
authored and
ggerganov
committed
vulkan: change gated_delta_net to shard a column across a subgroup (llama/20662)

jeffbolznv
authored and
ggerganov
committed
ggml-cpu: add always_inline to tinyBLAS_PPC accumulator saves (llama/20791)

shalinib-ibm
authored and
ggerganov
committed
Add shader count for Intel Arc Pro B60 (llama/20818)

TheBlueMatt
authored and
ggerganov
committed
fix(rpc): prevent division by zero in deserialize_tensor (llama/20712)

y198nt
authored and
ggerganov
committed
Increase number of output elements per-thread block if the K-dimension is small (llama/20635)

gaugarg-nv
authored and
ggerganov
committed
ggml-cuda: native bf16 flash attention for vec kernel (llama/20525)

authored and
ggerganov
committed
support bf16 and quantized type (llama/20803)

arthw
authored and
ggerganov
committed
CUDA: fix BF16 FA compilation (llama/20865)

JohannesGaessler
authored and
ggerganov
committed
opencl: add flattened Q4_K mv and general Q4_K mm (llama/20773)

shaofeiqi
authored and
ggerganov
committed
fix(openvino): explicit memset in buffer_context allocation (llama/20857)

committed
CANN: add RoPE cache preload before ACL graph capture (llama/20747)

noemotiovon
authored and
ggerganov
committed
metal: add CONV_3D (llama/19927)

Ra5hidIslam
and
ggerganov
committed
rpc : RCE patch (llama/20908)

las7
authored and
ggerganov
committed
opencl: add q6_K gemm and gemv kernels for Adreno (llama/20089)

lhez
authored and
ggerganov
committed
hexagon: general DMA and Binary Op fixes for large strides (llama/20918)

max-krasnyansky
authored and
ggerganov
committed
metal : add FA instantiations for HSK=512, HSV=512 (llama/20902)
ggerganov
committed
metal : add FLOOR, CEIL, ROUND, TRUNC unary ops (llama/20930)

authored and
ggerganov
committed
sycl : fix wrong variable check by assert (llama/20903)

arthw
authored and
ggerganov
committed
llama: fix llama-model-saver (llama/20503)

authored and
ggerganov
committed
mtmd: Add DeepSeekOCR Support (llama/17400)

authored and
ggerganov
committed
CUDA & CPU: support F32 kernel type for `CONV_TRANSPOSE_2D` (llama/17094)

authored and
ggerganov
committed
ggml-cuda: Add NVFP4 dp4a kernel (llama/20644)

authored and
ggerganov
committed
fix(ggml): correct RISC-V ISA string canonical ordering for RVV in CMake (llama/20888)

ihb2032
authored and
ggerganov
committed
opencl: allow large buffer for adreno (llama/20997)

lhez
authored and
ggerganov
committed
hip: use fnuz fp8 for conversion on CDNA3 (llama/21040)

IMbackK
authored and
ggerganov
committed
metal : Fix dimension constraint violation in matmul2d descriptor (llama/21048)

lathrys-at
authored and
ggerganov
committed
rpc : proper handling of data pointers to CPU buffers (llama/21030)

rgerganov
authored and
ggerganov
committed
hexagon: support for IQ4_NL and MXFP4 (llama/21018)

njsyw1997
authored and
ggerganov
committed
vulkan: add noncontiguous GLU support (llama/21081)

0cc4m
authored and
ggerganov
committed
sync : ggml
ggerganov
committed

Commits on Apr 17, 2026

whisper : add stateless VAD detect + explicit state reset for streaming (#3677)

danielbodart
and
claude
authored

Commits on Apr 20, 2026

bench : sync submit-results URL to ggml-org (#3769)
jinweihan-ai
authored

Commits on Apr 30, 2026

ggml : bump version to 0.9.9 (ggml/1449)
ggerganov
committed
hexagon: dma optimizations (mostly fixing regressions) (llama/21137)

max-krasnyansky
authored and
ggerganov
committed
Optimize MOE GEMV kernel for BS > 1. (llama/20905)

authored and
ggerganov
committed
rpc : fix misleading error log (llama/21184)

rgerganov
authored and
ggerganov
committed
CUDA : Fix CUB's argsort when nrows % block_size == 0 CCCL < 3.1 (llama/21181)

ORippler
authored and
ggerganov
committed
opencl: add q4_K gemm and gemv kernels for Adreno (llama/20919)

authored and
ggerganov
committed
sycl : enhance fattn perf (llama/21185)

arthw
authored and
ggerganov
committed
CANN: fix multi-thread set_tensor race conditions (llama/20151)

hipudding
authored and
ggerganov
committed
ggml-webgpu: port all AOT operators to JIT (llama/20728)

abhijitramesh
authored and
ggerganov
committed
ggml webgpu: quantized buffers to u32 + wider browser/device support (llama/21046)

reeselevine
authored and
ggerganov
committed
CUDA: Add Flash Attention Support for Head Dimension 512 (llama/20998)

authored and
ggerganov
committed
ggml-cpu: fix fallback for RVV kernels without zvfh (llama/21157)

taimur-10x
authored and
ggerganov
committed
ggml : fix RWKV ops thread assignment (llama/21226)
ggerganov
committed
CUDA/HIP: Fix kernel slection for mmvq mmid kernel to align host selection with device launch bounds (llama/21238)

IMbackK
authored and
ggerganov
committed
ggml : bump version to 0.9.10 (ggml/1454)
ggerganov
committed
ggml-cuda: Add generic NVFP4 MMQ kernel (llama/21074)

authored and
ggerganov
committed
sycl : support nvfp4 type in mul_mat (llama/21227)

arthw
authored and
ggerganov
committed
hexagon: improve RMS_NORM and DIV accuracy (llama/21251)

authored and
ggerganov
committed
CUDA: fix FA kernel selection logic (llama/21271)

JohannesGaessler
authored and
ggerganov
committed
opencl: fix leak in Adreno q8_0 path (llama/21212)

lhez
authored and
ggerganov
committed
hexagon : add cumsum op support (llama/21246)

authored and
ggerganov
committed
ggml : bump version to 0.9.11 (ggml/1456)
ggerganov
committed
sycl : fix llama_kv_cache hang when kv_cache is huge: 5GB (llama/21283)

arthw
authored and
ggerganov
committed
ggml-webgpu: add vectorized flash attention (llama/20709)

authored and
ggerganov
committed
rpc : reuse compute graph buffers (llama/21299)

rgerganov
authored and
ggerganov
committed
ggml-zendnn : add MUL_MAT_ID op support for MoE models (llama/21315)

authored and
ggerganov
committed
ggml-webgpu: move from parameter buffer pool to single buffer with offsets (llama/21278)

reeselevine
authored and
ggerganov
committed
hexagon: slight optimization for argosrt output init (llama/21463)

YardenTal44
authored and
ggerganov
committed
sycl : handle other FA case (llama/21377)

arthw
authored and
ggerganov
committed
Write an optimized flash_attn_stream_k_fixup kernel (llama/21159)

gaugarg-nv
authored and
ggerganov
committed
ggml: add Q1_0 1-bit quantization support (CPU) (llama/21273)

authored and
ggerganov
committed
ggml-webgpu: Add the support of `MUL_MAT_ID` (llama/21147)

authored and
ggerganov
committed
Add Q8_0 reorder optimization (~3x tg speedup on Intel Arc) (llama/21527)

PMZFX
authored and
ggerganov
committed
ggml-cuda : fix CDNA2 compute capability constant for gfx90a (MI210) (llama/21519)

aviallon
authored and
ggerganov
committed
vulkan: add FA dequant for q4_1, q5_0, q5_1, iq4_nl (llama/21029)

mkoker
authored and
ggerganov
committed
ggml: Vulkan build, Linux -- output error string for errno on fork failure (#20868) (llama/20904)

tomoverlund
authored and
ggerganov
committed
ggml : deprecate GGML_OP_ADD1 (llama/21363)
ggerganov
committed
CUDA: check for buffer overlap before fusing (llama/21566)

am17an
authored and
ggerganov
committed
ggml-webgpu: parameterize submission size and add iOS specific limits (llama/21533)

reeselevine
authored and
ggerganov
committed
ggml-cuda: ds_read_b128 for q4_0 and q4_1 mmq kernels (llama/21168)

authored and
ggerganov
committed
CUDA: make cuda graphs props check faster (llama/21472)

am17an
authored and
ggerganov
committed
metal: Q1_0 backend (llama/21528)

khosravipasha
and
ggerganov
committed
webgpu : Query for adapter support when registering WebGPU backend (llama/21579)

reeselevine
authored and
ggerganov
committed
fix: free ctx_copy in ggml_opt_free to plug per-training-session leak (llama/21592)

authored and
ggerganov
committed
CUDA: also store `node->src->data` ptrs for equality check (llama/21635)

am17an
authored and
ggerganov
committed
vulkan: unify type macros to use Vx instead of _VECx (llama/21605)

0cc4m
authored and
ggerganov
committed
sycl : add flash-attn support for head size 512 (llama/21654)

qnixsynapse
authored and
ggerganov
committed
metal : add missing mm-id specializations for q1_0 (llama/21662)
ggerganov
committed
ggml : check return value of CUB calls used in argsort and top-k (they all return cudaError_t) (llama/21676)

authored and
ggerganov
committed
ggml: backend-agnostic tensor parallelism (experimental) (llama/19378)

committed
HIP: add CDNA4 (gfx950) architecture support for MI350X/MI355X (llama/21570)

andyluo7
authored and
ggerganov
committed
CUDA: fuse muls (llama/21665)

am17an
authored and
ggerganov
committed
vulkan: Support Q1_0 (llama/21539)

jeffbolznv
authored and
ggerganov
committed
ggml-webgpu: address quantization precision and backend lifecycle managment (llama/21521)

authored and
ggerganov
committed
ggml-webgpu: support non-square subgroup matrix configs for Intel GPUs (llama/21669)

SharmaRithik
authored and
ggerganov
committed
hexagon: improved Op queuing, buffer and cache management (llama/21705)

authored and
ggerganov
committed
CUDA: also store node->src ne/nb for graph equality (llama/21736)

am17an
authored and
ggerganov
committed
ggml : fix a few instances of missing GGML_TYPE_Q1_0 cases (llama/21716)

CISC
authored and
ggerganov
committed
opencl: add basic support for q5_k (llama/21593)

shaofeiqi
authored and
ggerganov
committed
CUDA: skip compilation of superfluous FA kernels (llama/21768)

JohannesGaessler
authored and
ggerganov
committed
mtmd: add Gemma 4 audio conformer encoder support (llama/21421)

stephencox-ict
authored and
ggerganov
committed
sycl: disable Q1_0 in backend and cleanup unused variables (llama/21807)

qnixsynapse
authored and
ggerganov
committed
Remove extra conditional check on debug mode. (llama/21798)

yomaytk
authored and
ggerganov
committed
CUDA: Limit DeviceSegmentedSort to immediate mode (llama/21718)

ORippler
authored and
ggerganov
committed
vulkan: Flash Attention DP4A shader for quantized KV cache (llama/20797)

0cc4m
authored and
ggerganov
committed
vulkan: Support GGML_TYPE_NVFP4 (llama/21455)

jeffbolznv
authored and
ggerganov
committed
ggml-webgpu: Update register tiling matmul to use f32 accumulation (llama/21644)

reeselevine
authored and
ggerganov
committed
cmake: fix CMP0194 warning on Windows with MSVC (llama/21630)

texasich
authored and
ggerganov
committed
ggml : fix ARM NEON nvfp4 dot product on non-dotprod targets (llama/21559)

richarddd
authored and
ggerganov
committed
metal : add XIELU unary op (llama/20802)

seyoungjeong
authored and
ggerganov
committed
ci : re-enable mac workflows (llama/21894)
ggerganov
committed
vulkan: Programmatically add RoundingModeRTE to all shaders when the device supports it (llama/21572)

jeffbolznv
authored and
ggerganov
committed
metal : fix FA support logic (llama/21898)
ggerganov
committed
ggml : remove ggml-ext.h (llama/21869)

ngxson
and
ggerganov
committed
hexagon: optimization for HMX mat_mul (llama/21554)

authored and
ggerganov
committed
docs: more extensive RoPE documentation [no ci] (llama/21953)

ngxson
authored and
ggerganov
committed
rpc : add native RDMA transport for RPC backend (RoCEv2) (llama/20590)

dvv101111
authored and
ggerganov
committed
CUDA: manage NCCL communicators in context (llama/21891)

JohannesGaessler
authored and
ggerganov
committed
CUDA: require explicit opt-in for P2P access (llama/21910)

JohannesGaessler
authored and
ggerganov
committed
ggml-webgpu: Fix dequantization helpers to not pass in pointers (llama/21872)

reeselevine
authored and
ggerganov
committed
cuda: Q1_0 initial backend (llama/21629)

authored and
ggerganov
committed
vulkan: optimize im2col (llama/21713)

0cc4m
authored and
ggerganov
committed
Fix Q8_0 reorder: garbage on 2nd prompt + crash on full VRAM (llama/21638)

authored and
ggerganov
committed
ggml-webgpu: compute pass batching and removing profiling overhead (llama/21873)

reeselevine
authored and
ggerganov
committed
ggml : implemented simd_gemm kernel for riscv vector extension (llama/20627)

authored and
ggerganov
committed
ggml-cpu: add 128-bit RVV implementation for Quantization Vector Dot (llama/20633)

authored and
ggerganov
committed
metal: Implement ROLL op (llama/21946)

kushagharahi
authored and
ggerganov
committed
ggml: add graph_reused (llama/21764)

am17an
authored and
ggerganov
committed
opencl: add q5_K gemm and gemv kernels for Adreno (llama/21595)

shaofeiqi
authored and
ggerganov
committed
hexagon: optimize HMX matmul operations (llama/21071)

chraac
authored and
ggerganov
committed
opencl: refactor q8_0 set_tensor and mul_mat host side dispatch for Adreno (llama/21938)

lhez
authored and
ggerganov
committed
CUDA: use LRU based eviction for cuda graphs (llama/21611)

am17an
authored and
ggerganov
committed
ggml-webgpu: fix compiler warnings and refactor FlashAttention encoding (llama/21052)

reeselevine
authored and
ggerganov
committed
ggml-backend-meta: add multi-segment read support in get_tensor (llama/22063)

ssam18
authored and
ggerganov
committed
rpc : refactor the RPC transport (llama/21998)

rgerganov
authored and
ggerganov
committed
cmake: remove CMP0194 policy to restore MSVC builds (llama/21934)

texasich
authored and
ggerganov
committed
ggml : reduce CPU overhead in meta backend (llama/22041)

authored and
ggerganov
committed
HIP: Remove unesscary NCCL_CHECK (llama/21914)

IMbackK
authored and
ggerganov
committed
CUDA: refactor mma data loading for AMD (llama/22051)

JohannesGaessler
authored and
ggerganov
committed
Fix reorder MMVQ assert on unaligned vocab sizes (llama/22035)

PMZFX
authored and
ggerganov
committed
ggml-webgpu: updated matrix-vector multiplication (llama/21738)

authored and
ggerganov
committed
ggml-cpu: Optimized x86 and generic cpu q1_0 dot (follow up) (llama/21636)

pl752
authored and
ggerganov
committed
TP: fix 0-sized tensor slices, AllReduce fallback (llama/21808)

JohannesGaessler
authored and
ggerganov
committed
Tensor-parallel: Fix delayed AllReduce on Gemma-4 MoE (llama/22129)

gaugarg-nv
authored and
ggerganov
committed
ggml-cuda: flush legacy pool on OOM and retry (llama/22155)

leonardHONG
authored and
ggerganov
committed
ggml : bump version to 0.10.0 (ggml/1463)
ggerganov
committed
vulkan: Support F16 OP_FILL (llama/22177)

jeffbolznv
authored and
ggerganov
committed
metal : workaround macOS GPU interactivity watchdog (llama/22216)
ggerganov
committed
openvino: driver setup, CI split, thread safety, and NPU optimizations (llama/21944)

authored and
ggerganov
committed
hexagon: fix missing v79 entry in libggml-htp.inf (llama/22194)

mengshengwu
authored and
ggerganov
committed
Hexagon: DAIG op (llama/22195)

shreyajn
authored and
ggerganov
committed
ggml-webgpu: reset CPU/GPU profiling time when freeing context (llama/22050)

yomaytk
authored and
ggerganov
committed
hexagon: add support for FILL op (llama/22198)

authored and
ggerganov
committed
ggml-webgpu(shader): support conv2d kernels. (llama/21964)

authored and
ggerganov
committed
sycl: Improve mul_mat_id memory efficiency and add BF16 fast path (llama/22119)

qnixsynapse
authored and
ggerganov
committed
ggml-webgpu: Add fused RMS_NORM + MUL (llama/21983)

yomaytk
authored and
ggerganov
committed
Implement async tensor api and event api (llama/22099)

nikhilJain17
authored and
ggerganov
committed
HIP: flip GGML_HIP_GRAPHS to default on (llama/22254)

IMbackK
authored and
ggerganov
committed
CUDA: fuse relu + sqr (llama/22249)

anavp-nvidia
authored and
ggerganov
committed
ggml-webgpu: add support for im2col (llama/22259)

Constannnnnt
authored and
ggerganov
committed
sycl : fused MoE mul_mat_vec_q for TG (llama/21920)

authored and
ggerganov
committed
ggml-base: use MATH_LIBRARY variable instead of hardcoded 'm' (llama/22239)
ggerganov
committed
metal : fix event synchronization (llama/22260)
ggerganov
committed
hexagon: add support for basic and extended Op profiling (llama/22269)

authored and
ggerganov
committed
fix(shader): handle the buffer aliasing for rms fuse (llama/22266)

Constannnnnt
authored and
ggerganov
committed
hexagon: add SOLVE_TRI op (llama/21974)

authored and
ggerganov
committed
ggml : minor coding style (llama/22308)
ggerganov
committed
metal : print GPU description (llama/22318)
ggerganov
committed
hexagon: use DIRID 13 in libggml-htp.inf for modern InfVerif (llama/22306)

mengshengwu
authored and
ggerganov
committed
ggml-webgpu: enable FLASH_ATTN_EXT on browser without subgroup matrix (llama/22199)

ArberSephirotheca
authored and
ggerganov
committed
Hexagon: Bump HMX Frequency to Max Corner (llama/22334)

trivikram-reddy1
authored and
ggerganov
committed
ggml-webgpu: support for SSM_SCAN and disable set_rows error checking (llama/22327)

reeselevine
authored and
ggerganov
committed
Optimize Q4_0 mul_mat for Arc770, add scripts (llama/22291)

arthw
authored and
ggerganov
committed
metal : optimize Metal Tensor API usage for GGML_OP_MUL_MAT (llama/20962)

Developer-Ecosystem-Engineering
and
ggerganov
committed
CUDA: reduce MMQ stream-k overhead (llama/22298)

JohannesGaessler
authored and
ggerganov
committed
hexagon: guard HMX clock request for v75+ platforms (llama/22377)

trivikram-reddy1
authored and
ggerganov
committed
opencl: add iq4_nl support (llama/22272)

lhez
authored and
ggerganov
committed
ggml-cpu: optimize avx2 q6_k (llama/22345)

netrunnereve
authored and
ggerganov
committed
ggml-cpu : re-enable fast gelu_quick_f16 (llama/22339)

CISC
authored and
ggerganov
committed
CUDA: better coalesce data-access for contiguous concat (llama/22330)

ORippler
authored and
ggerganov
committed
Fix recurrent state serialization for partial reads and writes (llama/22362)

gaugarg-nv
authored and
ggerganov
committed
add performance-portable tuning for register-tile and subgroup matmul (llama/22241)

SharmaRithik
authored and
ggerganov
committed
ggml : use 64 bytes aligned tile buffers (llama/21058)

angt
authored and
ggerganov
committed
fix: rpc-server cache may not work in Windows environments (llama/22394)

authored and
ggerganov
committed
add fast mat-vec kernels for i-quants (llama/22344)

SharmaRithik
authored and
ggerganov
committed
ggml-webgpu: add Q1_0 support (llama/22374)

SharmaRithik
authored and
ggerganov
committed
CANN: add new ops, optimize existing ops (llama/21204)

hipudding
authored and
ggerganov
committed
ggml : revert to -lm linking instead of find_library (llama/22355)

angt
authored and
ggerganov
committed
ggml : skip already registered backends and devices (llama/22296)

angt
authored and
ggerganov
committed
ggml: improve SPIR-V headers detection with __has_include (llama/21918)

EmilAskerov
authored and
ggerganov
committed
vulkan: add barrier after writetimestamp (llama/21865)

jeffbolznv
authored and
ggerganov
committed
ggml-webgpu: fix buffer aliasing for ssm_scan and refactor aliasing logic (llama/22456)

reeselevine
authored and
ggerganov
committed
vulkan: Coalesce Q4_K/Q5_K scale loads (llama/21751)

TheBlueMatt
authored and
ggerganov
committed
ggml-cuda: add flash-attn support for DKQ=320/DV=256 with ncols2=32 (… (#22286)

authored and
ggerganov
committed
ggml-cuda: Repost of 21896: Blackwell native NVFP4 support (llama/22196)

michaelw9999
authored and
ggerganov
committed
TP: fix delayed AllReduce + zero-sized slices (llama/22489)

JohannesGaessler
authored and
ggerganov
committed
ggml : add sve tuned code for gemm_q8_0_4x8_q8_0() kernel (llama/21916)

authored and
ggerganov
committed
ggml-webgpu: Fix bug in FlashAttention support check (llama/22492)

reeselevine
authored and
ggerganov
committed
ggml-cpu: cmake: append xsmtvdotii march for SpacemiT IME (llama/22317)

authored and
ggerganov
committed
ggml-cuda: refactor fusion code (llama/22468)

am17an
authored and
ggerganov
committed
ggml : bump version to 0.10.1 (ggml/1469)
ggerganov
committed
sync : ggml
ggerganov
committed

Commits on May 1, 2026

examples : update to Q1_0
ggerganov
committed
ggml-cpu : disable tiled matmul on AIX to fix page boundary segfault (llama/22293)

authored and
ggerganov
committed
CUDA: fuse SSM_CONV + ADD(bias) + SILU (llama/22478)

anavp-nvidia
authored and
ggerganov
committed
hexagon: make vmem and buffer-size configurable (llama/22487)

authored and
ggerganov
committed
add fast matmul iquants (llama/22504)

SharmaRithik
authored and
ggerganov
committed
CUDA: fix tile FA kernel on Pascal (llama/22541)

JohannesGaessler
authored and
ggerganov
committed
vulkan: add get/set tensor 2d functions (llama/22514)

authored and
ggerganov
committed
ggml-webgpu: Improve performance of mat-vec and mat-mat for MUL_MAT_ID (llama/22464)

yomaytk
authored and
ggerganov
committed
ggml-webgpu: add the upscale shader (llama/22419)

Constannnnnt
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
ggml : try fix win32 build (#0)
ggerganov
committed

Commits on May 2, 2026

Commits on May 4, 2026