Fix enable chunked prefill for Llama4 (#5575)

This commit is contained in:
tarinkk
2025-04-20 17:01:30 -07:00
committed by GitHub
parent 417b44eba8
commit 9a7e83e899
+4 -1
View File
@@ -73,8 +73,11 @@ class ModelConfig:
)
if enable_multimodal is None:
if self.hf_config.architectures == "Llama4ForConditionalGeneration":
if self.hf_config.architectures[0] == "Llama4ForConditionalGeneration":
enable_multimodal = False
logger.info(
"Multimodal is disabled for Llama4. To enable it, set --enable-llama4-multimodal."
)
else:
enable_multimodal = True