ggerganov/whisper.cpp

Pull Requests Commits

CUDA: fuse shared experts into MMVQ (llama/29184)

am17an committed 8 days ago

Verified cf5d9e30

ggml-cuda : fix cpy transposed path corrupting non-contiguous dst (llama/27663)

devYRPauli committed 8 days ago

Verified d7682b58

ggml-quants : avoid invalid rounding in qkx3 scale search (llama/29817)

devYRPauli committed 8 days ago

Verified 6b704713

ggml-cpu : fix soft_max_back wrong output when dst aliases src1 (llama/27096)

devYRPauli committed 8 days ago

Verified 0b35d1f3

metal : add tensor API flash attention kernel for F16 KV (llama/29570)

ethanhq committed 8 days ago

Verified d0dc4363

opencl: use sigmoid f16 for bf16 (llama/29787)

lhez committed 8 days ago

Verified cca8f72b

SYCL: Q8_0 DMMV ESIMD and MMVQ wide load (llama/29186)

cwriter committed 8 days ago

Verified cd1bee52

vulkan: disable large matmul tile on Samsung GPUs with 32KB shared memory (llama/28531)

jwsong98 committed 8 days ago

Verified 95a05ed4

sycl: large register file for D=512 FA vec kernels (llama/29062)

Titaniumtown committed 8 days ago

Verified c4051a33

sycl : do not use slow oneDNN reference matmul and fattn (llama/28985)

lslusarczyk committed 8 days ago

Verified ae92605f

qwen4exp : optimize mask constructions (llama/29824)

ggerganov committed 8 days ago

Verified 9c968f7e

ggml : add `alloc_buffer_n` to buffer type interface (llama/23671)

ggerganov committed 8 days ago

Verified 4435763c

vulkan: add logging to pipeline compile issues (llama/29794)

0cc4m committed 8 days ago

Verified 8291ab84

hexagon: install rebuilt HTP skels (llama/29828)

kurquhar committed 8 days ago

Verified eaadab3d

hexagon: add q2_k and q3_k quant type support (llama/29717)

jhen0409 committed 8 days ago

Verified 0295ef68

CUDA: fix 2 broken Volta FA cases (llama/29803)

JohannesGaessler committed 8 days ago

Verified 396f68d3

llama: refer to segment documentation [no ci] (llama/29074)

JohannesGaessler committed 8 days ago

Verified 3a3598cc

hexagon: shared strided DMA copy for CPY and CONCAT, any-dim CONCAT via DMA (llama/29685)

njsyw1997 committed 8 days ago

Verified 15229ac9

cuda : route sm70 to the Turing MMVQ nwarps table (llama/29753)

tkittich committed 8 days ago

Verified aa538005

metal : release temporary private transfer buffers (llama/29777)

mvanlamz committed 8 days ago

Verified 81ca4f81

webgpu: add bfloat16 support for MUL_MAT/MUL_MAT_ID/GET_ROWS- #29358 (llama/29358)

yomaytk committed 8 days ago

Verified 88948bdb

CUDA: Handle compute type for NVFP4 on cublass path (llama/29173)

ynankani committed 8 days ago

Verified 10872be7

CUDA: Make CCCL configurable + pin it to 3.4.3 for CI jobs (llama/29792)

ORippler committed 8 days ago

Verified 3b68f901

meta: clear inactive AllReduce shards with FILL, not SCALE (llama/29793)

ggerganov committed 8 days ago

Verified 56500f4f

HIP: avoid treating CDNA as dgx spark for gqa_ratio 20 in fattn_mma dqk 576 (llama/29572)

IMbackK committed 8 days ago

Verified 817294a9

hex-workqueue: fix race condition in seqn getting out of sync with idx_read/write (llama/29785)

max-krasnyansky committed 8 days ago

Verified b4a085a0

BLAS : Document AOCL-BLAS build and label the device AOCL-BLAS (llama/29640)

pradeeptrgit committed 8 days ago

Verified 0ecc317e

opencl: mark vec subgroup bcast as supproted for Adreno E17 compiler (llama/29698)

lhez committed 8 days ago

Verified 67ad86d0

metal : use bf16 math for mxfp4 mul-mat (llama/29770)

ggerganov committed 8 days ago

Verified 61ea5027

webgpu: fix SSM_SCAN binding aliasing (llama/29750)

yomaytk committed 8 days ago

Verified d371e376

Newer Older