[Bugfix] use blockmanagerv1 for encoder-decoder (vllm-project#9084)

Co-authored-by: Roger Wang <[email protected]>
Signed-off-by: Amit Garg <[email protected]>

vllm/engine/arg_utils.py

@@ -903,6 +903,11 @@ def create_engine_config(self) -> EngineConfig:
                     "--enable-prefix-caching is currently not "
                     "supported for multimodal models and has been disabled.")
             self.enable_prefix_caching = False
+        if model_config.is_encoder_decoder_model:
+            logger.warning(
+                "Block Manager v2 does not support encoder-decoder models"
+                " currently. Using Block Manager v1 as fallback.")
+            self.use_v2_block_manager = False
 
         cache_config = CacheConfig(
             block_size=self.block_size if self.device != "neuron" else