Read use_external_data_format from ORTConfig file (#1917)

When quantizing the models >2Gb, it's important to
set the flag use_external_data_format to 'true',
since otherwise the quantization will fail due to

`ValueError: Message onnx.ModelProto exceeds maximum protobuf size of 2GB`

However, currently there is no way to set the parameter
when using optimum-cli because there is no such
command option. Theoretically, it could be set when
using ORTConfig file with -c comman flag, because one
of the configuration parameters in it is use_external_data_format.
In fact, the optimum code ignores it and does not pass it
in quantize() function.

The goal of this change is to close this gap.

optimum/onnxruntime/subpackage/commands/quantize.py

@@ -77,6 +77,7 @@ def run(self):
 
         save_dir = self.args.output
         quantizers = []
+        use_external_data_format = False
 
         quantizers = [
             ORTQuantizer.from_pretrained(self.args.onnx_model, file_name=model.name)
@@ -96,7 +97,11 @@ def run(self):
                 "TensorRT quantization relies on static quantization that requires calibration, which is currently not supported through optimum-cli. Please adapt Optimum static quantization examples to run static quantization for TensorRT: https://github.com/huggingface/optimum/tree/main/examples/onnxruntime/quantization"
             )
         else:
-            qconfig = ORTConfig.from_pretrained(self.args.config).quantization
+            config = ORTConfig.from_pretrained(self.args.config)
+            qconfig = config.quantization
+            use_external_data_format = config.use_external_data_format
 
         for q in quantizers:
-            q.quantize(save_dir=save_dir, quantization_config=qconfig)
+            q.quantize(
+                save_dir=save_dir, quantization_config=qconfig, use_external_data_format=use_external_data_format
+            )