[HiCache] Add PP Support with suffix pp rank (#15175)

Co-authored-by: Xuchun Shang <xuchun.shang@gmail.com>
Co-authored-by: ybyang <10629930+whybeyoung@users.noreply.github.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
This commit is contained in:
Teng Ma
2026-01-03 13:49:21 +08:00
committed by GitHub
co-authored by Xuchun Shang ybyang Shangming Cai
parent b23e7ed13c
commit bdde949619
6 changed files with 37 additions and 6 deletions
@@ -259,6 +259,8 @@ class HiCacheController:
prefetch_threshold: int = 256,
model_name: Optional[str] = None,
storage_backend_extra_config: Optional[dict] = None,
pp_rank: int = 0,
pp_size: int = 1,
):
self.mem_pool_device_allocator = token_to_kv_pool_allocator
self.mem_pool_device = token_to_kv_pool_allocator.get_kvcache()
@@ -267,6 +269,8 @@ class HiCacheController:
self.page_size = page_size
self.io_backend = io_backend
self.enable_storage = False
self.pp_rank = pp_rank
self.pp_size = pp_size
if storage_backend is not None:
self.storage_backend_type = storage_backend
@@ -394,6 +398,8 @@ class HiCacheController:
return HiCacheStorageConfig(
tp_rank=self.tp_rank,
tp_size=self.tp_size,
pp_rank=self.pp_rank,
pp_size=self.pp_size,
is_mla_model=is_mla_backend,
is_page_first_layout=self.mem_pool_host.layout == "page_first",
model_name=model_name,
+2
View File
@@ -642,6 +642,8 @@ class Scheduler(
enable_metrics=self.enable_metrics,
enable_kv_cache_events=self.enable_kv_cache_events,
enable_mamba_extra_buffer=server_args.enable_mamba_extra_buffer(),
pp_rank=self.pp_rank,
pp_size=self.pp_size,
)
if (