 Justine Tunney
|
a0b3ac8c48
ggml : introduce GGML_CALL function annotation (#4850)
|
%!s(int64=2) %!d(string=hai) anos |
 Georgi Gerganov
|
ddb008d845
cuda : fix dequantize kernel names (#4938)
|
%!s(int64=2) %!d(string=hai) anos |
 Kawrakow
|
4a3156de2f
CUDA: faster dequantize kernels for Q4_0 and Q4_1 (#4938)
|
%!s(int64=2) %!d(string=hai) anos |
 Johannes Gäßler
|
3fe81781e3
CUDA: faster q8_0 -> f16 dequantization (#4895)
|
%!s(int64=2) %!d(string=hai) anos |
 slaren
|
e7e4df031b
llama : ggml-backend integration (#4766)
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
1b280c9fff
CUDA: fix softmax compile for old CUDA versions (#4862)
|
%!s(int64=2) %!d(string=hai) anos |
|
Kawrakow
|
49662cbed3
ggml : SOTA 2-bit quants (add IQ2_XS) (#4856)
|
%!s(int64=2) %!d(string=hai) anos |
 Erik Scholz
|
f34432ca1e
fix : cuda order of synchronization when setting a buffer (ggml/679)
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
8f900abfc0
CUDA: faster softmax via shared memory + fp16 math (#4742)
|
%!s(int64=2) %!d(string=hai) anos |
|
Kawrakow
|
dd5ae06405
SOTA 2-bit quants (#4773)
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
d5a410e855
CUDA: fixed redundant value dequantization (#4809)
|
%!s(int64=2) %!d(string=hai) anos |
 Konstantin Zhuravlyov
|
63ee677efd
ggml : use __builtin_amdgcn_sudot4 in __dp4a for gfx11 (#4787)
|
%!s(int64=2) %!d(string=hai) anos |
 Finn Voorhees
|
1bf681f90e
ggml : add error handling to graph_compute (whisper/1714)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
7bed7eba35
cuda : simplify expression
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
d55356d3ba
cuda : mark I16 and I32 ops as unsupported
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
39d8bc71ed
CUDA: fixed tensor cores not being used on RDNA3 (#4697)
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
a20f3c7465
CUDA: fix tensor core logic for Pascal and HIP (#4682)
|
%!s(int64=2) %!d(string=hai) anos |
 hydai
|
91bb39cec7
cuda: fix vmm oom issue on NVIDIA AGX Orin (#4687)
|
%!s(int64=2) %!d(string=hai) anos |
 bssrdf
|
afc8c19291
ggml : fix some mul mat cases + add tests for src1 F16 (ggml/669)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
dc68f0054c
cuda : fix vmm pool with multi GPU (#4620)
|
%!s(int64=2) %!d(string=hai) anos |
 FantasyGmm
|
77465dad48
Fix new CUDA10 compilation errors (#4635)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
5bf3953d7e
cuda : improve cuda pool efficiency using virtual memory (#4606)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
708e179e85
fallback to CPU buffer if host buffer alloc fails (#4610)
|
%!s(int64=2) %!d(string=hai) anos |
|
Johannes Gäßler
|
e0a4002273
CUDA: fixed row rounding for 0 tensor splits (#4594)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
ba66175132
sync : ggml (fix im2col) (#4591)
|
%!s(int64=2) %!d(string=hai) anos |
|
FantasyGmm
|
a55876955b
cuda : fix jetson compile error (#4560)
|
%!s(int64=2) %!d(string=hai) anos |
 Henrik Forstén
|
6724ef1657
Fix CudaMemcpy direction (#4599)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
48b7ff193e
llama : fix platforms without mmap (#4578)
|
%!s(int64=2) %!d(string=hai) anos |
|
Georgi Gerganov
|
afefa319f1
ggml : change ggml_scale to take a float instead of tensor (#4573)
|
%!s(int64=2) %!d(string=hai) anos |
|
slaren
|
d232aca5a7
llama : initial ggml-backend integration (#4520)
|
%!s(int64=2) %!d(string=hai) anos |
