[LoRA] Code cleanup (#60017)

Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
This commit is contained in:
Jee Jee Li
2026-10-05 09:15:53 +00:00
committed by GitHub
parent 4a30c4cad0
commit e3af5bf3ce
7 changed files with 28 additions and 52 deletions
+2 -4
View File
@@ -3022,10 +3022,8 @@ class GPUModelRunner(
)
self.lora_manager.set_active_adapters(lora_requests, tower_mapping)
# Only set connector mapping if the model actually has a connector.
# Some multimodal models inherit a stub `get_num_mm_connector_tokens`
# from `SupportsMultiModal`, which returns None and should not be
# treated as a signal that connector LoRA is supported.
# Only set connector mapping if the model actually has a connector
# and reports connector token counts.
mm_mapping = (
self.model.get_mm_mapping() # type: ignore[attr-defined]
if hasattr(self.model, "get_mm_mapping")