remove the fa4 page_size hardcode to 128 restriction on mla model arch (#12801)

python/sglang/srt/server_args.py

@@ -1263,9 +1263,9 @@ class ServerArgs:
             raise ValueError(
                 "FA4 backend is only supported for prefill. Please use `--prefill-attention-backend fa4` instead."
             )
-        if self.prefill_attention_backend == "fa4":
+        if self.prefill_attention_backend == "fa4" and not self.use_mla_backend():
             logger.warning(
-                f"FA4 backend only supports page size 128, changing page_size from {self.page_size} to 128."
+                f"FA4 backend only supports page size 128 for non-MLA model architectures, changing page_size from {self.page_size} to 128."
             )
             self.page_size = 128