[Bugfix][LoRA] Use an explicit capability flag for tower connector LoRA (#53092)

Signed-off-by: linitra24 <Joy25810@foxmail.com>
This commit is contained in:
linitra24
2026-08-21 09:45:54 +00:00
committed by GitHub
parent 36bad1b90c
commit e85d1b69cf
21 changed files with 36 additions and 2 deletions
+2 -2
View File
@@ -201,8 +201,8 @@ class LoRAModelManager:
self.punica_wrapper_mapping[lm_prefix] = llm_punica_wrapper
# First, determine if the model supports tower connector LoRA.
self.supports_tower_connector_lora = self.supports_mm and hasattr(
self.model, "get_num_mm_encoder_tokens"
self.supports_tower_connector_lora = (
self.supports_mm and self.model.supports_tower_connector_lora
)
# Then, handle the case where the feature is disabled in the config.