whisper.cpp / ggml.c

Commit History

whisper : reduce memory usage during inference (#431)
3aa9e6c
unverified

ggerganov commited on

whisper : PPC64 big-endian support (#398)
239569b
unverified

fitzsim commited on

bench : add memcpy and ggml_mul_mat benchmarks
a660ed9
unverified

ggerganov commited on

ggml : remove obsolete zeroing + comment fixes (#390)
9c35c0d
unverified

ggerganov commited on

ggml : correct behaviour of ggml_vec_sum_f32 (#390)
ffffc6e
unverified

Abitofevrything commited on

ggml : improve vec_dot_f16 unrolling in flash_attn_f16
6e57274
unverified

ggerganov commited on

ggml : fix bug in new soft max computation
c59ce76
unverified

ggerganov commited on

ggml : when using BLAS start only 1 CPU thread
6c4692f
unverified

ggerganov commited on

ggml : fix running tasks with variable number of threads
2078d85
unverified

ggerganov commited on

ggml : unroll ggml_vec_dot_f16 in ggml_compute_forward_flash_attn_f16
f07fecd
unverified

ggerganov commited on

whisper : revert accidental MB change
db991e1
unverified

ggerganov commited on

ggml : speed-up soft max via Accelerate + unroll
fdaf59a
unverified

ggerganov commited on

ggml : use vDSP_sve and vDSP_maxv from Accelerate
ed14a8b
unverified

ggerganov commited on

ggml : make gcc happy (minor)
496acd2
unverified

ggerganov commited on

ggml : add SSE3 and fp16 conversion lookup table (#368)
2c3f7d4
unverified

Abitofevrything ggerganov commited on

whisper : document POWER VSX support
4dbf7ee

Thomas Fitzsimmons commited on

ggml : reorganize POWER9 ppc64le SIMD code
e0a5614

Thomas Fitzsimmons commited on

ggml : change f16 load and store macro arguments
4a68b87

Thomas Fitzsimmons commited on

ggml : add void to argument-less functions
f06f912
unverified

ggerganov commited on

ggml : define MIN / MAX only if not defined (minor)
2117da6
unverified

ggerganov commited on

ggml : improve f16 acceleration for POWER9 ppc64le
f92a260

Thomas Fitzsimmons commited on

ggml : barrier refactor + static functions
7b501c1
unverified

ggerganov commited on

ggml : simplify the SIMD code (#324)
6fe850c
unverified

ggerganov commited on

ggml : use vaddvq_f32 for slightly more efficient reduce
550fbf8
unverified

ggerganov commited on

ggml : add f16 acceleration for POWER9 ppc64le
0d5a830

Thomas Fitzsimmons commited on

ggml : make consts static (#317)
786be65
unverified

Andy Maloney commited on

minor : small code cleanups (#302)
142f526
unverified

Andy Maloney ggerganov commited on

Check for both __ARM_NEON and __ARM_FEATURE_FMA so that the project can be compiled for armv7a.
1fff54f

Kevin Brothaler commited on

Add AVX,AVX2 support for ggml_vec_scale_f32
666b50a

katsu560 commited on

ggml : implement ggml_compute_forward_dup_f16() special cases
b3b8141
unverified

ggerganov commited on

ggml : make more compatible with c99 (#262)
52bc68d
unverified

ggerganov commited on

ggml : fix indentation
09b6ef7
unverified

ggerganov commited on

ggml : make compatible with c99 (#262)
d9c1974
unverified

ggerganov commited on

Remove C++20 requirement (#257)
6702756
unverified

Roland Rabien commited on

talk : talk with AI in the terminal
7aad96d

ggerganov commited on

ggml : add alternative cblas_sgemm call
2f68de6
unverified

ggerganov commited on

fix compilation on haiku
5d3218a

Al Hoang commited on

ggml : fix typo in previous commit
d7b9296
unverified

ggerganov commited on

ggml : use macros to inline FP16 <-> FP32 conversions
23e5614
unverified

ggerganov commited on

ggml : add F16C CPU flag check
2661c19
unverified

ggerganov commited on

add fp16/fp32 convert intrinsics
03a5fa2

katsu560 commited on

ggml : remove inline specifier from fp16 <-> fp32 converters
cdd3359
unverified

ggerganov commited on

ggml : fix the fix
8b90727
unverified

ggerganov commited on

ggml : fix cross-compile Linux -> Window with mingw (#168)
29fe0ee
unverified

ggerganov commited on

ggml: change inline ggml_fp16_to_fp32, ggml_fp16_t ggml_fp32_to_fp16
b2f844a

katsu560 commited on

add AVX support
00ac035

katsu560 commited on

ggml : fix Windows build
42bbdcd

ggerganov commited on

Fix the Windows pthread_create shim
0c17e51

boolemancer commited on

ggml : multi-thread the ggml_add operator
c36d8ed
unverified

ggerganov commited on

ggml : fix the check for NEON support (#7)
1f7e8fa
unverified

ggerganov commited on