[Model] Add ModelConfig class for GraniteMoeHybrid to override default max_seq_len_to_capture (#20923)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>

Author: tdoublep

vllm/model_executor/models/config.py

@@ -205,6 +205,19 @@ def verify_and_update_config(vllm_config: "VllmConfig") -> None:
         }
 
 
+class GraniteMoeHybridModelConfig(VerifyAndUpdateConfig):
+
+    @staticmethod
+    def verify_and_update_config(vllm_config: "VllmConfig") -> None:
+        config = vllm_config.model_config
+        config.max_seq_len_to_capture = config.max_model_len
+        logger.info(
+            "Setting max_seq_len_to_capture to %d "
+            "to ensure that CUDA graph capture "
+            "covers sequences of length up to max_model_len.",
+            config.max_model_len)
+
+
 class HybridAttentionMambaModelConfig(VerifyAndUpdateConfig):
 
     @classmethod
@@ -297,4 +310,5 @@ def verify_and_update_config(cls, vllm_config: "VllmConfig") -> None:
     "Qwen3ForSequenceClassification": Qwen3ForSequenceClassificationConfig,
     "XLMRobertaModel": JinaRobertaModelConfig,
     "JinaVLForRanking": JinaVLForSequenceClassificationConfig,
+    "GraniteMoeHybridForCausalLM": GraniteMoeHybridModelConfig,
 }