[PD] Refactor parallel sizes and add pp support for mooncake (#8571)

Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
This commit is contained in:
Shangming Cai
2025-08-05 11:18:11 +08:00
committed by GitHub
parent 08f8f49016
commit d98a4913ea
4 changed files with 257 additions and 139 deletions

View File

@@ -103,6 +103,8 @@ class PrefillBootstrapQueue:
kv_args_class = get_kv_class(self.transfer_backend, KVClassType.KVARGS)
kv_args = kv_args_class()
kv_args.engine_rank = self.tp_rank
kv_args.pp_rank = self.pp_rank
kv_args.system_dp_rank = self.scheduler.dp_rank
kv_args.decode_tp_size = self.decode_tp_size // self.decode_dp_size
kv_args.prefill_pp_size = self.pp_size
kv_data_ptrs, kv_data_lens, kv_item_lens = (