Modify the examples optimization level (#230)

* Modify the examples optimization level

* Update readme

Author: echarlaix

README.md

@@ -154,8 +154,10 @@ Then let's take a look at applying _graph optimizations_ techniques such as oper
 ```python
 from optimum.onnxruntime.configuration import OptimizationConfig
 
-# optimization_config=99 enables all available graph optimisations
-optimization_config = OptimizationConfig(optimization_level=99)
+# Here the optimization level is selected to be 1, enabling basic optimizations such as redundant
+# node eliminations and constant folding. Higher optimization level will result in a hardware
+# dependent optimized graph.
+optimization_config = OptimizationConfig(optimization_level=1)
 ```
 
 Next, we load an _optimizer_ to apply these optimisations to our model:

docs/source/quickstart.mdx

@@ -95,8 +95,8 @@ As a final example, let's take a look at applying _graph optimizations_ techniqu
 ```python
 >>> from optimum.onnxruntime.configuration import OptimizationConfig
 
->>> # optimization_config=99 enables all available graph optimisations
->>> optimization_config = OptimizationConfig(optimization_level=99)
+>>> # Here the optimization level is selected to be 1, enabling basic optimizations such as redundant node eliminations and constant folding. Higher optimization level will result in a hardware dependent optimized graph.
+>>> optimization_config = OptimizationConfig(optimization_level=1)
 ```
 
 Next, we load an _optimizer_ to apply these optimisations to our model:

examples/onnxruntime/optimization/multiple-choice/README.md

@@ -18,11 +18,12 @@ limitations under the License.
 
 The script [`run_swag.py`](https://github.com/huggingface/optimum/blob/main/examples/onnxruntime/quantization/multiple-choice/run_swag.py) allows us to apply graph optimizations using [ONNX Runtime](https://github.com/microsoft/onnxruntime) for multiple choice tasks.
 
-The following example applies graph optimizations on a BERT fine-tuned on the SWAG dataset.
+The following example applies graph optimizations on a BERT fine-tuned on the SWAG dataset. Here the optimization level is selected to be 1, enabling basic optimizations such as redundant node eliminations and constant folding. Higher optimization level will result in hardware dependent optimized graph.
 
 ```bash
 python run_swag.py \
     --model_name_or_path ehdwns1516/bert-base-uncased_SWAG \
+    --optimization_level 1 \
     --do_eval \
     --output_dir /tmp/optimized_bert_swag
 ```

examples/onnxruntime/optimization/question-answering/README.md

@@ -22,13 +22,13 @@ allows us to apply graph optimizations using [ONNX Runtime](https://github.com/m
 Note that if your dataset contains samples with no possible answers (like SQuAD version 2), you need to pass along 
 the flag `--version_2_with_negative`.
 
-The following example applies graph optimizations on a DistilBERT fine-tuned on the SQuAD1.0 dataset.
+The following example applies graph optimizations on a DistilBERT fine-tuned on the SQuAD1.0 dataset. Here the optimization level is selected to be 1, enabling basic optimizations such as redundant node eliminations and constant folding. Higher optimization level will result in hardware dependent optimized graph.
 
 ```bash
 python run_qa.py \
     --model_name_or_path distilbert-base-uncased-distilled-squad \
     --dataset_name squad \
-    --optimization_level 99 \
+    --optimization_level 1 \
     --do_eval \
     --output_dir /tmp/optimized_distilbert_squad
 ```

examples/onnxruntime/optimization/text-classification/README.md

@@ -22,13 +22,13 @@ The script [`run_glue.py`](https://github.com/huggingface/optimum/blob/main/exam
 allows us to apply graph optimizations and fusion using [ONNX Runtime](https://github.com/microsoft/onnxruntime) for sequence classification tasks such as 
 the ones from the [GLUE benchmark](https://gluebenchmark.com/).
 
-The following example applies graph optimization on a DistilBERT fine-tuned on the sst-2 task.
+The following example applies graph optimization on a DistilBERT fine-tuned on the sst-2 task. Here the optimization level is selected to be 1, enabling basic optimizations such as redundant node eliminations and constant folding. Higher optimization level will result in hardware dependent optimized graph.
 
 ```bash
 python run_glue.py \
     --model_name_or_path distilbert-base-uncased-finetuned-sst-2-english \
     --task_name sst2 \
-    --optimization_level 99 \
+    --optimization_level 1 \
     --do_eval \
     --output_dir /tmp/optimized_distilbert_sst2
 ```

examples/onnxruntime/optimization/token-classification/README.md

@@ -19,13 +19,13 @@ limitations under the License.
 The script [`run_ner.py`](https://github.com/huggingface/optimum/blob/main/examples/onnxruntime/optimization/token-classification/run_ner.py)
 allows us to apply graph optimizations using [ONNX Runtime](https://github.com/microsoft/onnxruntime) for token classification tasks. 
 
-The following example applies graph optimizations on a DistilBERT fine-tuned on the CoNLL-2003 task
+The following example applies graph optimizations on a DistilBERT fine-tuned on the CoNLL-2003 task. Here the optimization level is selected to be 1, enabling basic optimizations such as redundant node eliminations and constant folding. Higher optimization level will result in hardware dependent optimized graph.
 
 ```bash
 python run_ner.py \
     --model_name_or_path elastic/distilbert-base-uncased-finetuned-conll03-english \
     --dataset_name conll2003 \
-    --optimization_level 99 \
+    --optimization_level 1 \
     --do_eval \
     --output_dir /tmp/optimized_distilbert_conll2003
 ```

tests/onnxruntime/test_onnxruntime.py

@@ -53,7 +53,7 @@ def test_optimize(self):
             "roberta-base",
             "google/electra-small-discriminator",
         }
-        optimization_config = OptimizationConfig(optimization_level=99, optimize_with_onnxruntime_only=False)
+        optimization_config = OptimizationConfig(optimization_level=2, optimize_with_onnxruntime_only=False)
         for model_name in model_names:
             with self.subTest(model_name=model_name):
                 with tempfile.TemporaryDirectory() as tmp_dir: