 AlpinDale
|
d9d287a288
rocm: enable multi-step scheduling for rocm (#1071)
|
hace 1 mes |
|
AlpinDale
|
58aff3771d
core: support prompt logprobs in multi-step (#1060)
|
hace 1 mes |
|
AlpinDale
|
1390915778
multi-step: add support for flashinfer attention backend (#1033)
|
hace 1 mes |
|
AlpinDale
|
c6e8cb058b
fix: lazy init _copy_stream (#1032)
|
hace 1 mes |
|
AlpinDale
|
5c3b94de45
spec decode: move ops.advane_step to flash attention backend (#1005)
|
hace 1 mes |
|
AlpinDale
|
f561a54a43
core: fix async postprocessor in case of preemption (#1000)
|
hace 1 mes |
|
AlpinDale
|
3bb0f07461
chore: rename `task_handler` to `worker` (#985)
|
hace 1 mes |
