| .. |
|
all_reduce
|
6b75a66c60
fix: unsafe all-reduce sync (#1118)
|
2 долоо хоног өмнө |
|
attention
|
3bb0f07461
chore: rename `task_handler` to `worker` (#985)
|
1 сар өмнө |
|
backup
|
3bb0f07461
chore: rename `task_handler` to `worker` (#985)
|
1 сар өмнө |
|
core
|
f1ea7711bd
core: do not compile ScalarType for torch < 2.4.0 (#938)
|
1 сар өмнө |
|
cpu
|
8976805f90
kernel: asymmetric AQ AZP quantization kernels (#1048)
|
1 сар өмнө |
|
cutlass_extensions
|
93bc863591
feat: Machete Kernels for Hopper GPUs (#842)
|
2 сар өмнө |
|
flash_attn
|
7fffa507ff
build: build flash attention kernels inside aphrodite (#1085)
|
3 долоо хоног өмнө |
|
hadamard
|
5d288aa76c
feat: add fast hadamard transformation kernels (#232)
|
1 жил өмнө |
|
mamba
|
9bdf8d5bfa
mamba: enable continuous batching for mamba kernels (#1055)
|
1 сар өмнө |
|
moe
|
fa84f8102e
kernels: split marlin kernels for faster compile, fix MoE, temporarily remove HQQ (#1119)
|
1 долоо хоног өмнө |
|
prepare_inputs
|
1390915778
multi-step: add support for flashinfer attention backend (#1033)
|
1 сар өмнө |
|
punica
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
quantization
|
fa84f8102e
kernels: split marlin kernels for faster compile, fix MoE, temporarily remove HQQ (#1119)
|
1 долоо хоног өмнө |
|
rocm
|
61aed092a5
rocm: add support for FP8 KV cache in the custom paged attention kkernels (#1066)
|
1 сар өмнө |
|
sampling
|
bfc8988116
feat: add cuda sampling kernels for top_k and top_p (#828)
|
3 сар өмнө |
|
xqa
|
949f974c59
(1/N) XQA: integrate the XQA CUDA kernels within Aphrodite (#1115)
|
3 долоо хоног өмнө |
|
activation_kernels.cu
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
cache.h
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
cache_kernels.cu
|
0256ed236b
feat: windows support (#790)
|
3 сар өмнө |
|
cuda_compat.h
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
cuda_utils.h
|
93bc863591
feat: Machete Kernels for Hopper GPUs (#842)
|
2 сар өмнө |
|
cuda_utils_kernels.cu
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
dispatch_utils.h
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
layernorm_kernels.cu
|
e14223dce5
kernel: use `cub::BlockReduce` instead of custom impl (#895)
|
2 сар өмнө |
|
ops.h
|
525edc1283
build: fix compilation for causal_conv1d_fwd kernel signature (#1057)
|
1 сар өмнө |
|
permute_cols.cu
|
93bc863591
feat: Machete Kernels for Hopper GPUs (#842)
|
2 сар өмнө |
|
pos_encoding_kernels.cu
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
5 сар өмнө |
|
torch_bindings.cpp
|
fa84f8102e
kernels: split marlin kernels for faster compile, fix MoE, temporarily remove HQQ (#1119)
|
1 долоо хоног өмнө |
AlpinDale
fa84f8102e
kernels: split marlin kernels for faster compile, fix MoE, temporarily remove HQQ (#1119)