SGLang + Megatron PPO takes a long time to reshard when generate_sequences #2419
Description
It takes a long time for SGLang+Megatron to reshard when generate_sequences @Moonlight-16B-A3B & 8xH20 & TP=4, comparing to vllm, releated to sgl-project/sglang#6762
Here lies a possible solution. #2418