Add ONNX model conversion docs (#2136)

Describes converting and optimizing SPLADE++ from HuggingFace to ONNX; Ref #2127

docs/onnx-conversion.md

@@ -0,0 +1,75 @@
+# ONNX Conversion for SPLADE++
+This MD file will describe steps to convert particular PyTorch models (i.e., [SPLADE++](https://doi.org/10.1145/3477495.3531857)) to ONNX models and options to further optimize compute graph for Transformer-based models. For more details on how does ONNX Conversion work and how to optimize the compute graph, please refer to [ONNX Tutorials](https://github.com/onnx/tutorials#services).
+
+## Setups
+- pip requirements:
+    ```
+    onnx                          1.13.1
+    onnxoptimizer                 0.3.10
+    onnxruntime                   1.11.1
+    ```
+- SPLADE [Git repo](https://github.com/naver/splade):
+   ```bash
+   git clone https://github.com/naver/splade.git
+   ```
+
+## Converting from PyTorch models to ONNX model
+The following sections will describe how to convert SPLADE++ model to ONNX model. The steps are as follows:  
+
+### Load the corresponding HuggingFace model
+```python
+from splade.models.transformer_rep import Splade
+import torch
+
+model_type_or_dir = "naver/splade-cocondenser-ensembledistil"
+model = Splade(model_type_or_dir, agg="max", fp16=True)
+model.eval()
+```
+
+### Creat dummy input variables to be used for ONNX conversion
+```python
+input_ids = torch.randint(1,100, size=(1,50))
+token_type_ids = torch.full((1,50), 0)
+attention_mask = torch.full((1,50), 1)
+inputs = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": attention_mask}
+
+```
+
+### Export the model to ONNX
+```python
+torch.onnx.export(model, inputs, 'splade-pp-ed.onnx', input_names=['input_ids', 'token_type_ids','attention_mask'], 
+                  output_names=['output_idx','output_weights'], dynamic_axes={'input_ids': {0: 'batch_size', 1: 'length'}, 
+                                                        'token_type_ids': {0:'batch_size', 1: 'length'}, 
+                'attention_mask': {0: 'batch_size', 1: 'length'}}, opset_version=12)
+```
+
+### Optimize the ONNX model (Optional)
+```python
+from onnxruntime.transformers import optimizer
+
+optimized_model = optimizer.optimize_model(
+    'splade-pp-ed-optimized.onnx',
+    model_type='bert',
+    num_heads=12,
+    hidden_size=768
+)
+
+# Optional: convert model to float16 (if not in fp16)
+# optimized_model.convert_float_to_float16()
+
+optimized_model.save_model_to_file('splade-pp-ed-optimized-fp16.onnx')
+```
+
+## Inference with ONNX Runtime
+The following sections will describe how to run inference with ONNX Runtime. The steps are as follows:
+```python
+import onnxruntime
+
+splade = onnxruntime.InferenceSession('path-to-splade-pp-ed-optimized-fp16.onnx')
+tokenizer = AutoTokenizer.from_pretrained(model_type_or_dir)
+
+inputs = tokenizer("This is a sample input", return_tensors="np")
+splade.run(None , {"input_ids": inputs['input_ids'], 
+                   "token_type_ids": inputs['token_type_ids'], 
+                   "attention_mask": inputs['attention_mask']})
+```