Fix enable chunked prefill for Llama4 (#5575)

9a7e83e8 · tarinkk · GitHub · 417b44eb · 9a7e83e8
Unverified Commit 9a7e83e8 authored Apr 20, 2025 by tarinkk Committed by GitHub Apr 20, 2025
Show whitespace changes
Inline Side-by-side

Showing with 4 additions and 1 deletion

python/sglang/srt/configs/model_config.py python/sglang/srt/configs/model_config.py +4 -1

No files found.
--- a/python/sglang/srt/configs/model_config.py
+++ b/python/sglang/srt/configs/model_config.py
@@ -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