| .. |
|
all_reduce
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
10 сар өмнө |
|
attention
|
ced1b36b8b
feat: support head size of 192
|
7 сар өмнө |
|
backup
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
11 сар өмнө |
|
cpu
|
ba02fb65c9
fix: pos encodings for CPU
|
7 сар өмнө |
|
hadamard
|
5d288aa76c
feat: add fast hadamard transformation kernels (#232)
|
1 жил өмнө |
|
moe
|
00acf371f9
rocm: fused topk softmax
|
7 сар өмнө |
|
punica
|
3bdeb3e116
fix: clang formatting for all kernels (#558)
|
7 сар өмнө |
|
quantization
|
e32f506e17
chore: gpu arch guard for cutlass w8a8 kernels
|
7 сар өмнө |
|
activation_kernels.cu
|
3d6695cfbb
feat: add approximate gelu activation kernels (#370)
|
11 сар өмнө |
|
cache.h
|
3bdeb3e116
fix: clang formatting for all kernels (#558)
|
7 сар өмнө |
|
cache_kernels.cu
|
3bdeb3e116
fix: clang formatting for all kernels (#558)
|
7 сар өмнө |
|
cuda_compat.h
|
00acf371f9
rocm: fused topk softmax
|
7 сар өмнө |
|
cuda_utils.h
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 жил өмнө |
|
cuda_utils_kernels.cu
|
31c95011a6
feat: FP8 E5M2 KV Cache (#226)
|
1 жил өмнө |
|
dispatch_utils.h
|
f8dfac6372
chore: attention refactor and upstream sync apr01 (#365)
|
11 сар өмнө |
|
layernorm_kernels.cu
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
10 сар өмнө |
|
ops.h
|
696f2cd59c
add phi3_small support with blocksparse attention
|
7 сар өмнө |
|
pos_encoding_kernels.cu
|
e702f587cf
feat: add batched RoPE kernels (#371)
|
11 сар өмнө |
|
pybind.cpp
|
3bdeb3e116
fix: clang formatting for all kernels (#558)
|
7 сар өмнө |
|
reduction.cuh
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
10 сар өмнө |
AlpinDale
e32f506e17
chore: gpu arch guard for cutlass w8a8 kernels