ggml-org/whisper.cpp

Pull Requests Commits

common : pass sample rate to `ffmpeg_decode_audio()`

ggerganov committed 28 days ago

Verified d418d645

ci : remove obsolete self-hosted label

ggerganov committed 28 days ago

Verified 8d7d5445

common : re-implement `ffmpeg-transcode.cpp` + clarify ffmpeg usage (#3846)

ggerganov committed 28 days ago

Verified f39cc712

ggerganov committed 30 days ago

f24588a2

ggml : bump version to 0.13.1 (ggml/1523)

ggerganov committed 30 days ago

92fc3f2a

talk-llama : sync llama.cpp

ggerganov committed 30 days ago

5828fba7

ggerganov committed 30 days ago

cc65eb18

cuda : disables launch_fattn PDL enrollment due to compiler bug (llama/23825)

aendk committed 30 days ago

e90501e1

meta : Add missing `buffer` set in allreduce fallback !COMPUTE clear (llama/23480)

TheBlueMatt committed 30 days ago

f1b687da

hexagon: basic/generic op fusion support and RMS_NORM+MUL fusion (llama/23835)

max-krasnyansky committed 30 days ago

442be178

opencl: move backend info printing into its own function (llama/23702)

lhez committed 30 days ago

94922ce1

ggml: auto apply iGPU flag CUDA/HIP if integrated device (llama/23007)

fl0rianr committed 30 days ago

e1faa7cb

mmvq Optim: add MMVQ_PARAMETERS_TURING(mmvq_parameter_table_id) for … (#23729)

yaohengxu committed 30 days ago

4e8af441

CUDA: route batch>=4 quantized matmul to MMQ on AMD MFMA hardware (llama/23227)

jadenmach2 committed 30 days ago

04795e62

hexagon: minor refresh for HMX FA and MM (llama/23796)

max-krasnyansky committed 30 days ago

1b241b87

vulkan: fast path for walsh-hadamard transform (llama/23687)

jeffbolznv committed 30 days ago

b896e91f

vulkan: fix wrong index variable in inner loop (llama/23665)

winstonma committed 30 days ago

816c3029

vulkan: Fix memory logger unsafe iterator access (llama/23667)

winstonma committed 30 days ago

5db94bac

cuda : fix KQ mask offset integer overflow in fattn MMA kernel (llama/23610)

fairydreaming committed 30 days ago

60e420ff

ggml: fixed Arm SVE usage bug in vec.h, vec.cpp (llama/22841)

martin-klacer-arm committed 30 days ago

8e403258

Hexagon: OP_GATED_DELTA_NET K>1 support (llama/23531)

ymcki committed 30 days ago

d284e1c3

opencl: OP_GATED_DELTA_NET (llama/23312)

ymcki committed 30 days ago

7e843a80

ggml-webgpu: remove legacy constants (llama/23672)

reeselevine committed 30 days ago

8c8f213d

hexagon: add support for Q4_1 in MUL_MAT and MUL_MAT_ID (llama/23647)

max-krasnyansky committed 30 days ago

3bbe9337

ggml-webgpu: Fix how to dispatch WG to some ops (llama/23750)

yomaytk committed 30 days ago

a52bd385

vulkan: Switch MUL_MAT_VEC to 4 K per iteration for F16/32 (llama/22887)

TheBlueMatt committed 30 days ago

8bce478e

vulkan: use GL_NV_cooperative_matrix_decode_vector for faster matmul (llama/23541)

jeffbolznv committed 30 days ago

1b590bbb

vulkan: add REPEAT op support for f16 to f16. (llama/23298)

l8bloom committed 30 days ago

c5cde8c7

CUDA: restrict PDL to CTK >= 12.3 due to MSVC issues (llama/23742)

ORippler committed 30 days ago

98c6722f

vulkan: avoid preferring transfer queue on AMD UMA devices (llama/22455)

winstonma committed 30 days ago

80e87ec4

Older