Skip to content

[bugfix] Fix wrong chunk cache error #25562

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
HaiShaw merged 1 commit into from
May 18, 2026
Merged

[bugfix] Fix wrong chunk cache error #25562

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 2 additions & 3 deletions python/sglang/srt/configs/model_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -450,17 +450,16 @@ def _derive_hybrid_model(self):
if not self.is_hybrid_swa:
self.has_attention_sinks = self._detect_attention_sinks()
self.is_hybrid_swa_compress = False
self.is_swa_with_compressed_attention = False
return

logger.info(f"Hybrid swa model: {self.hf_config.architectures=}")

self.is_swa_with_compressed_attention = any(
self.is_deepseek_v4_arch = any(
arch in ["DeepseekV4ForCausalLM", "DeepseekV4ForCausalLMNextN"]
for arch in self.hf_config.architectures
)

if self.is_hybrid_swa and not self.is_swa_with_compressed_attention:
if not self.is_deepseek_v4_arch:
self.swa_attention_layer_ids, self.full_attention_layer_ids = (
get_hybrid_layer_ids(
self.hf_config.architectures,
Expand Down
5 changes: 1 addition & 4 deletions python/sglang/srt/managers/scheduler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -866,10 +866,7 @@ def init_cache_with_memory_pool(self):
)

if effective_chunked_prefill_size is not None and self.disable_radix_cache:
is_v4_compressed = getattr(
self.model_config, "is_swa_with_compressed_attention", False
)
if not self.is_hybrid_swa or is_v4_compressed:
if not self.is_hybrid_swa:
from sglang.srt.mem_cache.chunk_cache import ChunkCache

self.tree_cache = ChunkCache(params)
Expand Down
Loading