Update the example of exporting Bart + BeamSearch to ONNX module to resolve comments.

examples/onnx/pytorch/translation/README.md

@@ -13,14 +13,17 @@ See the License for the specific language governing permissions and
 limitations under the License.
 -->
 
-# Usage
+# Bart + Beam Search to ONNX
 
-This folder contains an example of exporting Bart + Beam Search generation(BartForConditionalGeneration) to ONNX file.
+
+## Usage
+
+This folder contains an example of exporting Bart + Beam Search generation (`BartForConditionalGeneration`) to ONNX.
 
 Beam Search contains a for-loop workflow, so we need to make them TorchScript-compatible for exporting to ONNX. This example shows how to make a Bart model to be TorchScript-compatible by wrapping up it into a new model. In addition, some changes were made to the beam_search() function for making it TorchScript-compatible.
 
 
-# How to run the example
+## How to run the example
 
 To make sure you can successfully run the latest versions of the example scripts, you have to **install the library from source** and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
 

examples/onnx/pytorch/translation/bart_onnx/generation_onnx.py

@@ -13,14 +13,17 @@ def _convert_past_list_to_tuple(past):
     """
     The type of past_key_values is tuple(tuple(torch.FloatTensor)) which is not TorchScript-compatible.
     This will convert past values from a list to tuple(tuple(torch.FloatTensor)) for the inner decoder.
-    """
 
+    According to the definition of past_key_values, each inner tuple(torch.FloatTensor) has 4 tensors,
+    so we convert every 4 elements in the list as a tuple(torch.FloatTensor).
+    """
+    count_of_each_inner_tuple = 4
     results = ()
     temp_result = ()
-    count_n = len(past) // 4
+    count_n = len(past) // count_of_each_inner_tuple
     for idx in range(count_n):
-        real_idx = idx * 4
-        temp_result = tuple(past[real_idx : real_idx + 4])
+        real_idx = idx * count_of_each_inner_tuple
+        temp_result = tuple(past[real_idx : real_idx + count_of_each_inner_tuple])
         results += ((temp_result),)
 
     return results

examples/onnx/pytorch/translation/run_onnx_exporter.py

@@ -20,7 +20,6 @@
 import logging
 import os
 import sys
-from datetime import datetime
 
 import numpy as np
 import torch
@@ -196,7 +195,7 @@ def main():
     if args.output_file_path:
         output_name = args.output_file_path
     else:
-        output_name = "onnx_model_{}.onnx".format(datetime.now().utcnow().microsecond)
+        output_name = "onnx_model.onnx"
 
     logger.info("Exporting model to ONNX")
     export_and_validate_model(model, tokenizer, output_name, num_beams, max_length)