 AlpinDale
|
7fffa507ff
build: build flash attention kernels inside aphrodite (#1085)
|
hace 2 semanas |
|
AlpinDale
|
1390915778
multi-step: add support for flashinfer attention backend (#1033)
|
hace 1 mes |
|
AlpinDale
|
6f59024522
torch.compile: hide slicing under custom op for inductor (#1029)
|
hace 1 mes |
|
AlpinDale
|
de341ffb00
fix: ensure multistep lookahead allocation is compatible with cugraph max capture (#1008)
|
hace 1 mes |
|
AlpinDale
|
5c3b94de45
spec decode: move ops.advane_step to flash attention backend (#1005)
|
hace 1 mes |
|
AlpinDale
|
3bb0f07461
chore: rename `task_handler` to `worker` (#985)
|
hace 1 mes |
|
AlpinDale
|
1405051912
attention: add `AttentionState` abstraction (#863)
|
hace 2 meses |
|
AlpinDale
|
7a313483f1
chore: move update_flash_attn_metadata to attn backend (#731)
|
hace 4 meses |
|
AlpinDale
|
60b702a827
chore: register custom torch ops for flash-attn and flashinfer (#724)
|
hace 5 meses |
|
AlpinDale
|
24456206a9
fix: logit softcapping in flash-attn (#688)
|
hace 5 meses |
|
AlpinDale
|
7df7b8ca53
optimization: reduce end-to-end overhead from python obj allocation (#666)
|
hace 5 meses |
|
AlpinDale
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
hace 5 meses |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
hace 9 meses |
