Closed mali-git closed 3 months ago
This PR adds a manual SwiGLU implementation. The original one from xops was imcompatible with activation checkpointing (see issue #14)
General changes:
Breaking changes:
fused_swiglu
swiglu
ActivationType
fixes #14
This PR adds a manual SwiGLU implementation. The original one from xops was imcompatible with activation checkpointing (see issue #14)
General changes:
Breaking changes:
fused_swiglu
toswiglu
inActivationType
(see here for the respective config changes)