[HiCache][Dsv4] Don't cache C128 State pool in L3 (#27293)
This commit is contained in:
@@ -439,17 +439,8 @@ def build_deepseek_v4_hicache_stack(
|
|||||||
layout=server_args.hicache_mem_layout,
|
layout=server_args.hicache_mem_layout,
|
||||||
allocator_type=server_args.hicache_storage_backend,
|
allocator_type=server_args.hicache_storage_backend,
|
||||||
)
|
)
|
||||||
c128_state_host_pool = DeepSeekV4StateHostPool(
|
# C128 state pool is intentionally not registered with hicache.
|
||||||
pool_name=str(PoolName.DEEPSEEK_V4_C128_STATE),
|
# page_size=256 % 128 == 0, so state pool is not consumed on load.
|
||||||
state_pools=[
|
|
||||||
kvcache.compress_state_pools[layer_id]
|
|
||||||
for layer_id in c128_state_global_layers
|
|
||||||
],
|
|
||||||
num_host_pages=swa_num_host_pages,
|
|
||||||
swa_page_size=kvcache.swa_page_size,
|
|
||||||
layout=server_args.hicache_mem_layout,
|
|
||||||
allocator_type=server_args.hicache_storage_backend,
|
|
||||||
)
|
|
||||||
entries.extend(
|
entries.extend(
|
||||||
[
|
[
|
||||||
build_pool_entry(
|
build_pool_entry(
|
||||||
@@ -459,13 +450,6 @@ def build_deepseek_v4_hicache_stack(
|
|||||||
layer_mapping=c128_layer_mapping,
|
layer_mapping=c128_layer_mapping,
|
||||||
transfer_layer_num=transfer_layer_num,
|
transfer_layer_num=transfer_layer_num,
|
||||||
),
|
),
|
||||||
build_pool_entry(
|
|
||||||
name=PoolName.DEEPSEEK_V4_C128_STATE,
|
|
||||||
host_pool=c128_state_host_pool,
|
|
||||||
device_pool=None,
|
|
||||||
layer_mapping=c128_state_mapping,
|
|
||||||
transfer_layer_num=transfer_layer_num,
|
|
||||||
),
|
|
||||||
]
|
]
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user