Create quantization_flow.md using inpage editor

Summary:
This diff has been automatically generated by the inpage editor.

NOTE: If you want to update this diff, go via the preview link inside the static docs section below.
Ensure you are editing the same page that was used to create this diff.

Reviewed By: mergennachin

Differential Revision: D47457075

fbshipit-source-id: 003ae829afdb198e0312b2fa3350fcb00bf5dbb4

Author: jerryzh168

docs/website/quantization_flow.md

@@ -0,0 +1,53 @@
+# Quantization Flow in Executorch
+
+High level flow for short term quantization flow in exeuctorch looks like the following: https://docs.google.com/document/d/1UuktDffiMH0rXRuiL0e8bQaS3X0XfkIHEF1VtpFmA5A/edit#heading=h.mywdosyratgh
+
+## 1. Capture the model with `exir.capture`
+### Process
+The flow uses `PyTorch 2.0 Export Quantization` to quantize the model, that works on a model captured by `exir.capture`. If the model is not traceable, please follow the [User Guide](TBD) to make changes to model, and see [here](https://pytorch.org/docs/main/generated/exportdb/index.html) for supported constructs in `exir.capture`.
+
+```
+# program capture
+from executorch import exir
+from executorch.exir.tracer import ExirDynamoConfig
+from executorch.exir import CaptureConfig
+dynamo_config = exir.tracer.ExirDynamoConfig(
+    capture_scalar_outputs=True,
+    guard_nn_modules=True,
+    dynamic_shapes=enable_dynamic_shape,
+    specialize_int=True,
+    verbose=True,
+)
+capture_config = exir.CaptureConfig(
+    pt2_mode=True,
+    _dynamo_config=dynamo_config,
+    enable_dynamic_shape=enable_dynamic_shape,
+)
+exported_program = exir.capture(m, example_inputs, capture_config)
+m = exported_program.graph_module
+```
+### Result
+The result in this step will be a capturable model (`fx.GraphModule`)
+## 2. Quantization
+### Process
+Please take a look at the [pytorch 2.0 export post training static quantization tutorial](https://pytorch.org/tutorials/prototype/fx_graph_mode_ptq_static.html) to learn about all the steps of quantization. Main APIs that's used to quantize the model would be:
+* `prepare_pt2e`: used to insert observers to the model, it takes a backend specific `Quantizer` as argument, which will annotate the nodes with informations needed to quantize the model properly for the backend
+* (not an api) calibration: run the model through some sample data
+* `convert_pt2e`: convert a observed model to a quantized model, we have special representation for selected ops (e.g. quantized linear), other ops are represented as (dq -> float32_op -> q), and q/dq are decomposed into more primitive operators.
+
+### Result
+The result after these steps will be a reference quantized model, with quantize/dequantize operators being further decomposed. Example:
+
+(TODO): update
+```
+# Reference Quantized Pattern for quantized add
+x = torch.ops.quantized_decomposed.dequantize_per_tensor(x, x_scale, x_zero_point, x_qmin, x_qmax, torch.uint8)
+y = torch.ops.quantized_decomposed.dequantize_per_tensor(y, y_scale, y_zero_point, y_qmin, y_qmax, torch.uint8)
+out = x + y
+out = torch.ops.quantized_decomposed.quantize_per_tensor(out, out_scale, out_zero_point, out_qmin, out_qmax, torch.uint8)
+```
+
+see https://docs.google.com/document/d/17h-OEtD4o_hoVuPqUFsdm5uo7psiNMY8ThN03F9ZZwg/edit#heading=h.ov8z39149wy8 for some operators that has integer operator representations.
+
+## 4. Lowering to Executorch
+TODO: link