Enable ONNX/ONNXRuntime optimizations through converter script (#6131)

* Add onnxruntime transformers optimization support

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Added Optimization section in ONNX/ONNXRuntime documentation.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Improve note reference

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Fixing imports order.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Add warning about different level of optimization between torch and tf export.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Address @LysandreJik wording suggestion

Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Address @LysandreJik wording suggestion

Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

* Always optimize model before quantization for maximum performances.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Address comments on the documentation.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Improve TensorFlow optimization message as suggested by @yufenglee

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Removed --optimize parameter

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Warn the user about current quantization limitation when model is larger than 2GB.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

* Trigger CI for last check

* Small change in print for the optimization section.

Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>

Co-authored-by: Lysandre Debut <lysandre@huggingface.co>

docs/source/serialization.rst

@@ -5,7 +5,7 @@ Exporting transformers models
 ONNX / ONNXRuntime
 ==============================================
 
-Projects ONNX (Open Neural Network eXchange) and ONNXRuntime (ORT) are part of an effort from leading industries in the AI field
+Projects `ONNX (Open Neural Network eXchange) <http://onnx.ai>`_ and `ONNXRuntime (ORT) <https://microsoft.github.io/onnxruntime/>`_ are part of an effort from leading industries in the AI field
 to provide a unified and community-driven format to store and, by extension, efficiently execute neural network leveraging a variety
 of hardware and dedicated optimizations.
 
@@ -34,9 +34,36 @@ The conversion tool works for both PyTorch and Tensorflow models and ensures:
 
 Also, the conversion tool supports different options which let you tune the behavior of the generated model:
 
-* Change the target opset version of the generated model: More recent opset generally supports more operator and enables faster inference.
+* **Change the target opset version of the generated model.**  (More recent opset generally supports more operators and enables faster inference)
-* Export pipeline specific prediction heads: Allow to export model along with its task-specific prediction head(s).
+
-* Use the external data format (PyTorch only): Lets you export model which size is above 2Gb (`More info <https://github.com/pytorch/pytorch/pull/33062>`_).
+* **Export pipeline-specific prediction heads.**  (Allow to export model along with its task-specific prediction head(s))
+
+* **Use the external data format (PyTorch only).**  (Lets you export model which size is above 2Gb (`More info <https://github.com/pytorch/pytorch/pull/33062>`_))
+
+
+Optimizations
+------------------------------------------------
+
+ONNXRuntime includes some transformers-specific transformations to leverage optimized operations in the graph.
+Below are some of the operators which can be enabled to speed up inference through ONNXRuntime (*see note below*):
+
+* Constant folding
+* Attention Layer fusing
+* Skip connection LayerNormalization fusing
+* FastGeLU approximation
+
+
+Fortunately, you can let ONNXRuntime find all the possible optimized operators for you. Simply add ``--optimize``
+when exporting your model through ``convert_graph_to_onnx.py``.
+
+Example:
+
+.. code-block:: bash
+
+    python convert_graph_to_onnx.py --framework <pt, tf> --model bert-base-cased --optimize bert-base-cased.onnx
+
+.. note::
+    For more information about the optimizations enabled by ONNXRuntime, please have a look at the (`ONNXRuntime Github <https://github.com/microsoft/onnxruntime/tree/master/onnxruntime/python/tools/transformers>`_)
 
 Quantization
 ------------------------------------------------
@@ -85,6 +112,8 @@ Example of quantized BERT model export:
     above command will contain the original ONNX model storing `float32` weights.
     The second one, with ``-quantized`` suffix, will hold the quantized parameters.
 
+.. note::
+    The quantization export gives the best performances when used in combination with ``--optimize``.
 
 TorchScript
 =======================================

src/transformers/convert_graph_to_onnx.py

@@ -3,7 +3,7 @@ from os import listdir, makedirs
 from pathlib import Path
 from typing import Dict, List, Optional, Tuple
 
-from packaging.version import parse
+from packaging.version import Version, parse
 
 from transformers import is_tf_available, is_torch_available
 from transformers.file_utils import ModelOutput
@@ -72,7 +72,7 @@ def generate_identified_filename(filename: Path, identifier: str) -> Path:
     return filename.parent.joinpath(filename.stem + identifier).with_suffix(filename.suffix)
 
 
-def ensure_onnxruntime_installed():
+def check_onnxruntime_requirements(minimum_version: Version):
     """
     Check onnxruntime is installed and if the installed version match is recent enough.
     Raises:
@@ -88,7 +88,7 @@ def ensure_onnxruntime_installed():
         if ort_version < ORT_QUANTIZE_MINIMUM_VERSION:
             raise ImportError(
                 f"We found an older version of onnxruntime ({onnxruntime.__version__}) "
-                f"but we require onnxruntime to be >= 1.4.0 to enable all the conversions options.\n"
+                f"but we require onnxruntime to be >= {minimum_version} to enable all the conversions options.\n"
                 f"Please update onnxruntime by running `pip install --upgrade onnxruntime`"
             )
 
@@ -330,6 +330,30 @@ def convert(
         convert_tensorflow(nlp, opset, output)
 
 
+def optimize(onnx_model_path: Path) -> Path:
+    """
+    Load the model at the specified path and let onnxruntime look at transformations on the graph
+    to enable all the optimizations possible
+    Args:
+        onnx_model_path: filepath where the model binary description is stored
+
+    Returns: Path where the optimized model binary description has been saved
+
+    """
+    from onnxruntime import SessionOptions, InferenceSession
+
+    # Generate model name with suffix "optimized"
+    opt_model_path = generate_identified_filename(onnx_model_path, "-optimized")
+    sess_option = SessionOptions()
+    sess_option.optimized_model_filepath = opt_model_path.as_posix()
+    _ = InferenceSession(onnx_model_path.as_posix(), sess_option)
+
+    print(f"Optimized model has been written at {opt_model_path}: \N{heavy check mark}")
+    print("/!\\ Optimized model contains hardware specific operators which might not be portable. /!\\")
+
+    return opt_model_path
+
+
 def quantize(onnx_model_path: Path) -> Path:
     """
     Quantize the weights of the model from float32 to in8 to allow very efficient inference on modern CPU.
@@ -338,17 +362,18 @@ def quantize(onnx_model_path: Path) -> Path:
 
     Returns: The Path generated for the quantized
     """
-
     try:
-        ensure_onnxruntime_installed()
         import onnx
-        from onnxruntime import __version__ as ort_version
         from onnxruntime.quantization import quantize, QuantizationMode
 
-        print(f"Found ONNX: {onnx.__version__}")
-        print(f"Found ONNXRuntime: {ort_version}")
-
         onnx_model = onnx.load(onnx_model_path.as_posix())
+
+        # Discussed with @yufenglee from ONNX runtime, this will be address in the next release of onnxruntime
+        print(
+            "As of onnxruntime 1.4.0, models larger than 2GB will fail to quantize due to protobuf constraint.\n"
+            "This limitation will be removed in the next release of onnxruntime."
+        )
+
         quantized_model = quantize(
             model=onnx_model, quantization_mode=QuantizationMode.IntegerOps, force_fusions=True, symmetric_weight=True,
         )
@@ -357,11 +382,11 @@ def quantize(onnx_model_path: Path) -> Path:
         quantized_model_path = generate_identified_filename(onnx_model_path, "-quantized")
 
         # Save model
-        print(f"Storing quantized model at {quantized_model_path}")
+        print(f"Quantized model has been written at {quantized_model_path}: \N{heavy check mark}")
-        onnx.save(quantized_model, quantized_model_path.as_posix())
+        onnx.save_model(quantized_model, quantized_model_path.as_posix())
 
         return quantized_model_path
-    except ImportError as ie:
+    except Exception as ie:
         print(f"Error while quantizing the model:\n{str(ie)}")
 
 
@@ -369,7 +394,7 @@ def verify(path: Path):
     from onnxruntime import InferenceSession, SessionOptions
     from onnxruntime.capi.onnxruntime_pybind11_state import RuntimeException
 
-    print(f"Checking ONNX model loading from: {path}")
+    print(f"Checking ONNX model loading from: {path} ...")
     try:
         onnx_options = SessionOptions()
         _ = InferenceSession(path.as_posix(), onnx_options, providers=["CPUExecutionProvider"])
@@ -386,6 +411,7 @@ if __name__ == "__main__":
     args.output = Path(args.output).absolute()
 
     try:
+        print("\n====== Converting model to ONNX ======")
         # Convert
         convert(
             args.framework,
@@ -398,12 +424,34 @@ if __name__ == "__main__":
         )
 
         if args.quantize:
-            args.quantized_output = quantize(args.output)
+            # Ensure requirements for quantization on onnxruntime is met
+            check_onnxruntime_requirements(ORT_QUANTIZE_MINIMUM_VERSION)
+
+            # onnxruntime optimizations doesn't provide the same level of performances on TensorFlow than PyTorch
+            if args.framework == "tf":
+                print(
+                    "\t Using TensorFlow might not provide the same optimization level compared to PyTorch.\n"
+                    "\t For TensorFlow users you can try optimizing the model directly through onnxruntime_tools.\n"
+                    "\t For more information, please refer to the onnxruntime documentation:\n"
+                    "\t\thttps://github.com/microsoft/onnxruntime/tree/master/onnxruntime/python/tools/transformers\n"
+                )
+
+            print("\n====== Optimizing ONNX model ======")
+
+            # Quantization works best when using the optimized version of the model
+            args.optimized_output = optimize(args.output)
+
+            # Do the quantization on the right graph
+            args.quantized_output = quantize(args.optimized_output)
 
         # And verify
         if args.check_loading:
+            print("\n====== Check exported ONNX model(s) ======")
             verify(args.output)
 
+            if hasattr(args, "optimized_output"):
+                verify(args.optimized_output)
+
             if hasattr(args, "quantized_output"):
                 verify(args.quantized_output)