 Stephan Walter
|
939ad2d3a5
Fix undefined variables in debug build, remove unused variables (#531)
|
%!s(int64=2) %!d(string=hai) anos |
 slaren
|
459e93cce0
Add AVX2 implementation of dequantize_row_q4_1 (#505)
|
%!s(int64=2) %!d(string=hai) anos |
 Georgi Gerganov
|
a316a425d0
Overhaul the examples structure
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
ecbe466a36
Retire the ggml_mul_mat() branch for transposed src0 (#500)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
09aecbf628
Add AVX2 implementation of dequantize_row_q4_0 (#467)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
6b6dbc8910
Remove obsolete assert and fix compiler warning
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
2a2e63ce05
Fix nasty bug in ggml_compute_forward_mul_mat_f32() and reenable BLAS
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
8520fc310e
Disable BLAS altogether - the bug is not just for qunatized mat mul
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
b3f460e941
Disable BLAS branch in mul_mat - seems there is a bug
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
7a9b6c3a8b
Reduce memory usage and allocate enough memory for largest context (#473)
|
%!s(int64=2) %!d(string=hai) anos |
 Cameron Kaiser
|
481044d50c
additional optimizations for POWER9 (#454)
|
%!s(int64=2) %!d(string=hai) anos |
 comex
|
563cdc391d
Support calling mlock() on loaded model data on Linux and macOS (#453)
|
%!s(int64=2) %!d(string=hai) anos |
|
Stephan Walter
|
69c92298a9
Deduplicate q4 quantization functions (#383)
|
%!s(int64=2) %!d(string=hai) anos |
 Valentyn Bezshapkin
|
97940520e8
fix: add POSIX functionality for Linux compilation (#51)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
f5a77a629b
Introduce C-style API (#370)
|
%!s(int64=2) %!d(string=hai) anos |
 Kevin Lo
|
715d292ee0
Add OpenBSD support (#314)
|
%!s(int64=2) %!d(string=hai) anos |
 Casey Primozic
|
2e664f1ff4
Add initial AVX512 support for dot product on Linux (#320)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
22213a17b5
Change RMSNorm eps to 1e-6 (#173)
|
%!s(int64=2) %!d(string=hai) anos |
|
Stephan Walter
|
367946c668
Don't tell users to use a bad number of threads (#243)
|
%!s(int64=2) %!d(string=hai) anos |
 Matvey Soloviev
|
904d2a8d6a
Q4_1 quantization (#193)
|
%!s(int64=2) %!d(string=hai) anos |
 Nebula
|
9b4a15b17d
Fix RMS norm in GGML (#191)
|
%!s(int64=2) %!d(string=hai) anos |
 hoangmit
|
6eac39ba95
Add RMS norm and use it (#187)
|
%!s(int64=2) %!d(string=hai) anos |
|
hoangmit
|
113e685d18
inline -> static inline for "bytesFromNibbles" (#161)
|
%!s(int64=2) %!d(string=hai) anos |
 Ronsor
|
47857e564c
Don't use vdotq_s32 if it's not available (#139)
|
%!s(int64=2) %!d(string=hai) anos |
 Thomas Klausner
|
41be0a3b3d
Add NetBSD support. (#90)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
84d9015c4a
Use vdotq_s32 to improve performance (#67)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
c80e2a8f2a
Revert "10% performance boost on ARM"
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
54a0e66ea0
Check for vdotq_s32 availability
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
543c57e991
Ammend to previous commit - forgot to update non-QRDMX branch
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
113a9e83eb
10% performance boost on ARM
|
%!s(int64=2) %!d(string=hai) anos |
