add step3-vl tuning

benchmarks/kernels/benchmark_moe.py

@@ -682,6 +682,11 @@ def main(args: argparse.Namespace):
         topk = config.num_experts_per_tok
         intermediate_size = config.moe_intermediate_size
         shard_intermediate_size = 2 * intermediate_size // tp_size
+    elif config.architectures[0] in ("Step3VLForConditionalGeneration"):
+        E = config.text_config.moe_num_experts
+        topk = config.text_config.moe_top_k
+        intermediate_size = config.text_config.moe_intermediate_size
+        shard_intermediate_size = 2 * intermediate_size // tp_size
     else:
         # Support for llama4
         config = config.get_text_config()

vllm/config.py

@@ -4948,4 +4948,4 @@ def get_layers_from_vllm_config(vllm_config: VllmConfig,
         for layer_name, layer in
         vllm_config.compilation_config.static_forward_context.items()
         if isinstance(layer, layer_type)
-    }
+    }
\ No newline at end of file