[Quant] Move parts of BackendConfig tutorial

[andrewor14]

Summary: This commit moves the API specification and the Data
Type Restriction sections to the docstrings. See
pytorch/pytorch#91999. This commit also
reorganizes the end-to-end example to make it easier to follow.

Reviewers: jerryzh168, vkuzo

Subscribers: jerryzh168, vkuzo

[jerryzh168]

For the tutorial example, it would be good to lay out a step by step guide to user as well, start with a scenario, e.g. user have a backend that they want to implement quantized linear and quantized conv2d_relu operator, how they figure out the BackendConfig step by step.

e.g. 1. decide the list of quantized operators in the backend 2. write down the reference pattern for each quantized operator 3. for each operator, write the BackendPatternConfig based on the reference pattern of the quantized operator, and what types of fusions they will need etc. 4. combine BackendPatternConfig and get a BackendConfig object 5. write custom qconfig mapping 6. use it in prepare and convert

some of pieces are already there in the tutorial, but it might be better to just list the steps more explicitly, similar to https://github.com/pytorch/tutorials/blob/main/prototype_source/fx_graph_mode_ptq_static.rst

[andrewor14]

For the tutorial example, it would be good to lay out a step by step guide to user as well, start with a scenario, e.g. user have a backend that they want to implement quantized linear and quantized conv2d_relu operator, how they figure out the BackendConfig step by step.

e.g. 1. decide the list of quantized operators in the backend 2. write down the reference pattern for each quantized operator 3. for each operator, write the BackendPatternConfig based on the reference pattern of the quantized operator, and what types of fusions they will need etc. 4. combine BackendPatternConfig and get a BackendConfig object 5. write custom qconfig mapping 6. use it in prepare and convert

some of pieces are already there in the tutorial, but it might be better to just list the steps more explicitly, similar to https://github.com/pytorch/tutorials/blob/main/prototype_source/fx_graph_mode_ptq_static.rst

Sounds good. I reorganized them based on your suggestion.

[andrewor14]

Hi @svekars, can you help me merge this? Thanks.