f3ee4a9673
whisper : reduce memory usage during inference ( #431 )
...
* ggml : add "scratch" buffer support
* ggml : support for scratch ring-buffer
* ggml : bug fix in ggml_repeat()
* ggml : error on scratch buffer overflow
* whisper : use scratch buffers during inference (base model only)
* whisper : update memory usage for all models
* whisper : fix encoder memory usage
* whisper : use whisper_context functions instead of macros
* whisper : fix FF + remove it from README
* ggml : reuse ggml_new_i32
* ggml : refactor the scratch buffer storage
* whisper : reorder scratch buffers in the decoder
* main : add option to disable temp fallback
* Update README.md
2 years ago
ae16c21e9c
whisper : PPC64 big-endian support ( #398 )
...
* ggml : set cache line size to 128 on POWER9
* whisper : add PPC64 big endian support
2 years ago
1290fc6457
bench : add memcpy and ggml_mul_mat benchmarks
2 years ago
4ef3398e8f
ggml : remove obsolete zeroing + comment fixes ( #390 )
2 years ago
8d7b29cedd
ggml : correct behaviour of ggml_vec_sum_f32 ( #390 )
2 years ago
52a3e0c92a
ggml : improve vec_dot_f16 unrolling in flash_attn_f16
2 years ago
f30b5d322c
ggml : fix bug in new soft max computation
2 years ago
d347a59a5f
ggml : when using BLAS start only 1 CPU thread
2 years ago
6394c906af
ggml : fix running tasks with variable number of threads
2 years ago
74ffa14e1d
ggml : unroll ggml_vec_dot_f16 in ggml_compute_forward_flash_attn_f16
2 years ago
65fdcbbbbb
whisper : revert accidental MB change
2 years ago
d61d55cd4b
ggml : speed-up soft max via Accelerate + unroll
2 years ago
d51fc3ee0a
ggml : use vDSP_sve and vDSP_maxv from Accelerate
2 years ago
f82a7dd019
ggml : make gcc happy (minor)
2 years ago
a62170c656
ggml : add SSE3 and fp16 conversion lookup table ( #368 )
...
* Improves WASM performance:
On MacBook M1 Pro, I observe 25% faster using Firefox and 35% faster using Chrome
* Add support for SSE3 SIMD
* Add SSE3 to system information
* Add Imath support for fp16-fp32 conversions
* Add Imath to system information
* Wrap Imath calls to avoid static function warnings
* Drop Imath; Add lookup table for f16 -> f32 conversions
* Remove TODO comments
* Update SSE3 to new macro arguments
* Correct updated macro definitions
* Prefer static inline where possible
* ggml : static inlines + add public f16 <-> f32 conversions
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2 years ago
1944e7c33e
whisper : document POWER VSX support
2 years ago
49a8dd6732
ggml : reorganize POWER9 ppc64le SIMD code
2 years ago
8c7f642286
ggml : change f16 load and store macro arguments
2 years ago
0a0cfa7985
ggml : add void to argument-less functions
2 years ago
d51c5eb906
ggml : define MIN / MAX only if not defined (minor)
2 years ago
424c410c42
ggml : improve f16 acceleration for POWER9 ppc64le
2 years ago
4e0b2069e7
ggml : barrier refactor + static functions
2 years ago
ac521a566e
ggml : simplify the SIMD code ( #324 )
...
* ggml : simplify the SIMD code
* ggml : generic reduce for all register sizes + comments
2 years ago
7282e2109e
ggml : use vaddvq_f32 for slightly more efficient reduce
2 years ago
466ceebb78
ggml : add f16 acceleration for POWER9 ppc64le
2 years ago
493d94130d
ggml : make consts static ( #317 )
...
These shouldn't be able to be referenced outside the compilation unit.
2 years ago
fa463313ad
minor : small code cleanups ( #302 )
...
* Small code cleanups
- fix indentation
- remove extra semicolons
- remove extra break after returns in case statements
- remove unnecessary call to .data() on string
- use empty() instead of checking size()
- no need to check for nullptr before free
- remove unnecessary initialization of string to ""
* minor : switch case always break
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2 years ago
e1432dd91a
Check for both __ARM_NEON and __ARM_FEATURE_FMA so that the project can be compiled for armv7a.
...
Android armeabi-v7a's NEON support doesn't support FMA unless configured with `-mfpu=neon-fp-armv8`, which would need runtime checks.
* Also removed ABI filter from Android project.
2 years ago
419b8a6402
Add AVX,AVX2 support for ggml_vec_scale_f32
2 years ago
a7047b2a28
ggml : implement ggml_compute_forward_dup_f16() special cases
2 years ago
0f11759406
ggml : make more compatible with c99 ( #262 )
2 years ago
f66ac6dc4f
ggml : fix indentation
2 years ago
9955fa4ed7
ggml : make compatible with c99 ( #262 )
2 years ago
e70d47baab
Remove C++20 requirement ( #257 )
...
* Remove C++20 requirement
* Roll back C features not supported in VS2017
2 years ago
3b1aacbe6d
talk : talk with AI in the terminal
2 years ago
50a061b313
ggml : add alternative cblas_sgemm call
2 years ago
04a16bbf11
fix compilation on haiku
2 years ago
b6597539f9
ggml : fix typo in previous commit
2 years ago
9a4b7a916e
ggml : use macros to inline FP16 <-> FP32 conversions
2 years ago
f8ec718b76
ggml : add F16C CPU flag check
2 years ago
35b40a93b9
add fp16/fp32 convert intrinsics
2 years ago
061fc81bd6
ggml : remove inline specifier from fp16 <-> fp32 converters
2 years ago
388e9f79ad
ggml : fix the fix
2 years ago
35cd29ce1f
ggml : fix cross-compile Linux -> Window with mingw ( #168 )
2 years ago
804f36aa2c
ggml: change inline ggml_fp16_to_fp32, ggml_fp16_t ggml_fp32_to_fp16
2 years ago
83456076f0
add AVX support
2 years ago
2065572a11
ggml : fix Windows build
2 years ago
0bfe728b84
Fix the Windows pthread_create shim
...
The current implementation doesn't actually set the out parameter,
and it returns 0 on failure instead of on success.
2 years ago
75171c2b79
ggml : multi-thread the ggml_add operator
2 years ago
137321915f
ggml : fix the check for NEON support ( #7 )
...
Was using the wrong preprocessor macro
2 years ago
Georgi Gerganov