 Georgi Gerganov
|
4d98d9a656
sync : ggml (SD ops, tests, kernels) (#4444)
|
2 years ago |
 slaren
|
799a1cb13b
llama : add Mixtral support (#4406)
|
2 years ago |
|
Georgi Gerganov
|
fe680e3d10
sync : ggml (new ops, tests, backend, etc.) (#4359)
|
2 years ago |
|
Georgi Gerganov
|
bcc0eb4591
llama : per-layer KV cache + quantum K cache (#4309)
|
2 years ago |
|
Georgi Gerganov
|
ef47ec18da
ggml : add ggml_soft_max_ext (#4256)
|
2 years ago |
|
Georgi Gerganov
|
3d68f364f1
ggml : sync (im2col, GPU conv, 32-bit arm compat) (#4060)
|
2 years ago |
|
Georgi Gerganov
|
183b3fac6c
metal : fix build errors and kernel sig after #2268 (#3898)
|
2 years ago |
 cebtenzzre
|
898aeca90a
llama : implement YaRN RoPE scaling (#2268)
|
2 years ago |
|
Georgi Gerganov
|
e16b9fa4ba
metal : multi-simd softmax (#3710)
|
2 years ago |
|
Georgi Gerganov
|
469c9addef
metal : handle ggml_scale for n%4 != 0 (close #3754)
|
2 years ago |
 Jhen-Jie Hong
|
c67fe68e41
metal : implement q5_0 and q5_1 kernels (#3648)
|
2 years ago |
|
Georgi Gerganov
|
fcca0a7004
refact : fix convert script + zero out KV cache to avoid nans (#3523)
|
2 years ago |
|
Georgi Gerganov
|
b0ec5218c3
metal : support MTLGPUFamily < Apple7, formatting, style (#3524)
|
2 years ago |
 Phillip Kravtsov
|
0e797c2fc5
llm : support Adept Persimmon 8B (#3410)
|
2 years ago |
 Jiahao Li
|
f56e1baec3
metal : alibi for arbitrary number of heads (#3426)
|
2 years ago |
|
Georgi Gerganov
|
ec893798b7
llama : custom attention mask + parallel decoding + no context swaps (#3228)
|
2 years ago |
|
Georgi Gerganov
|
c6f1491da0
metal : fix bug in soft_max kernels (out-of-bounds access) (#3194)
|
2 years ago |
|
Georgi Gerganov
|
8c00b7a6ff
sync : ggml (Metal F32 support + reduce ggml-alloc size) (#3192)
|
2 years ago |
|
Georgi Gerganov
|
a51b687657
metal : relax conditions on fast matrix multiplication kernel (#3168)
|
2 years ago |
 Kawrakow
|
f31b6f4e2d
metal : PP speedup (#3084)
|
2 years ago |
|
Kawrakow
|
ba7ffbb251
metal : Q3_K speedup (#2995)
|
2 years ago |
|
Kawrakow
|
be8c9c245b
metal : parallel RoPE on Metal (#3024)
|
2 years ago |
|
Kawrakow
|
be6beeb8d7
metal : correct fix of kernel_norm (#3060)
|
2 years ago |
|
Georgi Gerganov
|
c4f496648c
metal : fix kernel_norm (fixes Falcon on Metal) (#3057)
|
2 years ago |
|
Georgi Gerganov
|
b7f2aa9e51
metal : restore 363f0bf and fix reduce in F16_F32 kernels (#2986)
|
2 years ago |
|
Georgi Gerganov
|
d9151e6f57
metal : revert 6af0bab until we fix it
|
2 years ago |
|
Kawrakow
|
ca82cf7bac
metal : more optimizations (#2959)
|
2 years ago |
|
Georgi Gerganov
|
13268c5331
metal : slight speed-up for add and mul kernels (#2917)
|
2 years ago |
|
Kawrakow
|
e8d9158925
metal: somewhat faster f16 x f32 matrix multiply kernel (#2951)
|
2 years ago |
|
Georgi Gerganov
|
d67777c202
metal : add Q8_0 support (#2763)
|
2 years ago |
