Separate swa and local attention chunk cache eviction (#15820)

This commit is contained in:
Ke Bao
2025-12-26 09:34:22 +08:00
committed by GitHub
parent 2f66b0671b
commit 7b7e357f61
8 changed files with 50 additions and 29 deletions
+2 -4
View File
@@ -648,10 +648,8 @@ class Scheduler(
else:
from sglang.srt.mem_cache.chunk_cache import SWAChunkCache
params.is_local_attention = (
"Llama4ForConditionalGeneration"
in self.model_config.hf_config.architectures
)
params.sliding_window_size = self.model_config.sliding_window_size
params.attention_chunk_size = self.model_config.attention_chunk_size
self.tree_cache = SWAChunkCache(params)
else: