config: three cache and pool readers take the bags (#36791)

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Cheng Wan
2026-08-28 10:21:32 -07:00
committed by GitHub
co-authored by Claude Opus 5
parent 43c63a22ff
commit 7bc3204117
4 changed files with 19 additions and 14 deletions
@@ -169,7 +169,7 @@ class DefaultPoolConfigurator(MemoryPoolConfigurator):
self._zero_kv_max_tokens = (
torch.iinfo(torch.int64).max
if has_kv_on_another_pp_stage
else kvc.server_args.max_total_tokens or kvc.model_config.context_len
else get_schedule().max_total_tokens or kvc.model_config.context_len
)
# EAGLE/STANDALONE: scale cell_size to account for draft model KV cache.
@@ -792,7 +792,7 @@ class DSV4PoolConfigurator(MemoryPoolConfigurator):
self.disaggregation_decode_extra_slots = (
get_disagg().disaggregation_decode_extra_slots or 0
)
if kvc.server_args.enable_hisparse:
if get_memory().enable_hisparse:
from sglang.srt.mem_cache.sparsity import parse_hisparse_config
self.c4_shrink_factor = parse_hisparse_config(