0dac2ddeac
* llama4 support for linearized experts * clean up fsdp2 sharding to prevent hang * add yaml config * cleanup example [skip ci]
2.0 KiB
2.0 KiB
0dac2ddeac