disable overlap schedule if mamba radix cache open (#13057)

This commit is contained in:
Yi Zhang
2025-11-11 16:35:38 +08:00
committed by GitHub
parent fe92d4d88e
commit 8db7fc4186

View File

@@ -1070,6 +1070,13 @@ class ServerArgs:
f"Disabling Radix Cache for {model_arch} as it is not yet supported."
)
self.disable_radix_cache = True
elif model_arch in ["Qwen3NextForCausalLM"]:
if not self.disable_radix_cache:
logger.warning(
"Disabling overlap schedule since MambaRadixCache is not compatible with "
"overlap schedule currently, try to use --disable-radix-cache if overlap schedule is necessary"
)
self.disable_overlap_schedule = True
if is_deepseek_nsa(hf_config):
if (