diff --git a/js/web/lib/wasm/wasm-core-impl.ts b/js/web/lib/wasm/wasm-core-impl.ts
index 36328d5f7e459..9cf2d75f318d6 100644
--- a/js/web/lib/wasm/wasm-core-impl.ts
+++ b/js/web/lib/wasm/wasm-core-impl.ts
@@ -79,8 +79,8 @@ export const createSessionAllocate = (model: Uint8Array, weights?: ArrayBuffer):
 
   let weightsFile: FSNode|undefined;
   if (weights) {
-    weightsFile = wasm.FS.create('/home/web_user/weights.pb');
-    weightsFile.contents = new Uint8Array(weights);
+    weightsFile = wasm.FS.create('/home/web_user/model.onnx.data');
+    weightsFile.contents = weights;
     weightsFile.usedBytes = weights.byteLength;
     wasm.FS.chdir('/home/web_user');
   }
@@ -149,7 +149,7 @@ export const createSessionFinalize =
         }
         allocs.forEach(alloc => wasm._free(alloc));
         if (modelData[2]) {
-          wasm.FS.unlink('/home/web_user/weights.pb');
+          wasm.FS.unlink('/home/web_user/model.onnx.data');
         }
       }
     };
diff --git a/onnxruntime/python/tools/transformers/models/stable_diffusion/optimize_pipeline.py b/onnxruntime/python/tools/transformers/models/stable_diffusion/optimize_pipeline.py
index a8e3c69332339..b3691cf288773 100644
--- a/onnxruntime/python/tools/transformers/models/stable_diffusion/optimize_pipeline.py
+++ b/onnxruntime/python/tools/transformers/models/stable_diffusion/optimize_pipeline.py
@@ -146,7 +146,7 @@ def optimize_sd_pipeline(
             logger.info("Convert %s to float16 ...", name)
             op_block_list = ["RandomNormalLike"]
             m.convert_float_to_float16(
-                keep_io_types=False,
+                keep_io_types=True,
                 op_block_list=op_block_list + force_fp32_operators[name],
             )