justheuristic b0c7b5c30f wip: implement grad wrt logits 5 жил өмнө
..
__init__.py c58d08cc06 remove run_and_await_k completely, rename gating_function to moe 5 жил өмнө
expert.py 6fb99c8746 wip: parallel fault-tolerant moe backward pass 5 жил өмнө
moe.py b0c7b5c30f wip: implement grad wrt logits 5 жил өмнө