| Type - | -Definition - | -Notes - | -
| Tensor - | -Pytorch Tensor type - | -https://pytorch.org/docs/stable/tensors.html#torch.Tensor - | -
| Scalar - | -Any numerical types from Python, including integral types, floating point types, and zero dimensional tensors. - | -float and int argument types should suffice for most algorithms (you should only use Scalar if the operator truly may accept either type) - | -
| int - | -python int (binded as int64_t in C++) - | -- | -
| float - | -python float (binded as double in C++) - | -- | -
| bool - | -python bool - | -- | -
| str - | -python string - | -- | -
| ScalarType - | -Enum type for all permissible dtype - | -https://pytorch.org/docs/stable/tensor_attributes.html#torch.dtype - | -
| Layout - | -Enum type for all permissible Layout - | -https://pytorch.org/docs/stable/tensor_attributes.html#torch.layout - | -
| MemoryFormat - | -Enum type for all permissible MemoryFormat - | -https://pytorch.org/docs/stable/tensor_attributes.html#torch-memory-format - | -
| Device - | -torch.device -> str - | -https://pytorch.org/docs/stable/tensor_attributes.html#torch.device - | -
| Tuple - | -Python tuple - | -- | -
| List - | -Python list - | -- | -
| Dict - | -Python dict with keys Scalar as defined in the table above. - | -- | -
| Format - | -Column Order - | -
| contiguous_format - | -NCHW - | -
| Channels_last - | -NHWC - | -
| (no name) - | -Other permutations - | -
| -Op Set - | -Canonical ATen - | -Custom Op - | -|
| ATen - | -Allowed - | -Allowed, must have meta kernel - | -- | -
| Edge - | -Aten + Type specializations - | -Allowed - | -- | -
- 
-
-
-- **AIBench** - With an ETrecord users can benchmark their model on devices in the AIBench lab. We'll behind the scenes generate an ETdump for you with all the performance data, retrieve it from the device and then display the corresponding URL on the AIBench job result page that the users can open for graph and performance data visualization.
-
-
- 
-
-
-## SDK tutorials:
-The SDK tutorial is overall broken into the following components:
-
-- [Generating an ETrecord](./01_generating_etrecord.md)
-- [Generating an ETdump](./02_generating_etdump.md)
-- [Library and CLI](./03_using_sdk_cli_tools.md)
-- [Visualizing models along with performance data](./04_visualizing_models.mdx)
-- [Debugging with ETDB](./05_ETDB.md)
-- [Using AIBench to profile and visualize models](./06_aibench_sdk.md)
-
-## SDK fundamentals:
-Before we dive into to understand how to use the tooling let's understand some fundamental components of the SDK that are integral for fully leveraging all the functionality it provides. These are:
-
-- **ETrecord**
-- **ETdump**
-
-
-### ETrecord:
-
-
-ETrecord is intended to be the debug artifact that is generated by users ahead of time. To draw a rough equivalent to conventional software development ETrecord can be considered as the binary built with debug symbols that is used for debugging in gdb.
-
-When generating an ETrecord the user will be expected to provide the graph modules that they are interested in working with via our tooling. The user can also provide their ExecuTorch program to store in ETrecord which can then later also be used to run the actual model on device or anywhere the user intends to. Further details on how to generate an ETrecord are provided in the ETrecord section.
-
-
-### ETdump
-
-
-ETdump is the serialized object that is generated by the runtime after running a model. Similarly as above to draw a rough equivalent to conventional software development ETdump could be considered as the coredump of ExecuTorch, but in this case within ETdump we store all the performance and debug data that was generated by the runtime.
-
-ETdump is then provided to our tooling to visualize the performance and debug data generated by the runtime. When ETdump and ETrecord are both passed to our tooling we are able to correlate this performance and debug data with the corresponding nodes in the graph that generated them.
-
-*** NOTE:*** If you only want to visualize the graph then you don't need an ETDump. You will only need an ETDump if you want to visualize the graph along with model performance and debug data.
diff --git a/docs/website/docs/sdk/01_generating_etrecord.md b/docs/website/docs/sdk/01_generating_etrecord.md
deleted file mode 100644
index 8347832513..0000000000
--- a/docs/website/docs/sdk/01_generating_etrecord.md
+++ /dev/null
@@ -1,55 +0,0 @@
-# Generating an ETrecord
-
-Make sure to read and understand what an ETrecord is and why we need it for the SDK in the [getting started page](./00_getting_started.md#etrecord)
-
-
-For a demo on how to generate an ETrecord, please refer to this [notebook](https://www.internalfb.com/intern/anp/view/?id=3799219).
-
-There are two important API's users must be aware of when dealing with ETrecord: `generate_etrecord` and `parse_etrecord`.
-
----
-### Generating an ETrecord
-
-```python
-generate_etrecord(
- etrecord_path: str,
- edge_dialect_program: ExirExportedProgram
- executorch_program: Union[ExecutorchProgram, MultiMethodExecutorchProgram],
- export_modules: Optional[
- Dict[
- str, Union[MultiMethodExirExportedProgram, ExirExportedProgram]
- ]
- ] = None,
-) -> None:
-```
-
-Generates an ETRecord from the given objects and saves it to the given path.
-The objects that will be serialized to an ETRecord are all the graph modules present in the export_modules dict, the graph module present in the edge dialect program object,
-and also the graph module present in the executorch program object, which is the closest graph module representation of what is eventually run on the device.
-
-In addition to all the graph modules we also serialize the program buffer which the users can provide to the ExecuTorch runtime to run the model.
-
-#### Parameters:
-- `etrecord_path` : Path to where the ETRecord file will be saved to.
-- `edge_dialect_program`: ExirExportedProgram for this model returned by the call to to_edge()
-- `executorch_program`: ExecutorchProgram or MultiMethodExecutorchProgram for this model returned by the call to to_executorch()
-- `export_modules`: Dictionary of graph modules with the key being the user provided name and the value is the corresponding exported module. The exported graph modules can be either the output of capture() or to_edge().
-
-#### Returns:
-None
-
----
-
-### Parsing an ETrecord
-
-```python
-parse_etrecord(etrecord_path: str)
-```
-
-Parses an ETRecord file and returns a ETRecord object that contains the deserialized graph modules, program buffer and debug handle map. In the graph map in the returned ETRecord object if a model with multiple entry points was provided originally by the user during ETRecord generation then each entry point will be stored as a separate graph module in the ETRecord object with the name being the original module name + "/" + the name of the entry point.
-
-#### Parameters:
- - `etrecord_path`: Path to the ETRecord file.
-
-#### Returns:
- - ETRecord object
diff --git a/docs/website/docs/sdk/02_generating_etdump.md b/docs/website/docs/sdk/02_generating_etdump.md
deleted file mode 100644
index 0ddc3b74d0..0000000000
--- a/docs/website/docs/sdk/02_generating_etdump.md
+++ /dev/null
@@ -1,49 +0,0 @@
-# Generating an ETdump
-> Note: If you're using AIBench with one of the pre-built binaries then you can skip this step.
-AIBench will handle the retrieval and processing of ETdump for you.
-
-Generating an ETdump from the runtime is very trivial. It consists of two steps.
-
-1) Integrate the etdump generation code into your application to generate an etdump after the model has been executed.
-
-2) Compile your binary with the profiling enabled flag to ensure that the profiling data is written out to etdump.
-
----
-
-#### 1. Integrating ETdump generation code into your application
-
-- Add this header to your file
-```c
-#include
-
-
diff --git a/docs/website/docs/sdk/05_ETDB.md b/docs/website/docs/sdk/05_ETDB.md
deleted file mode 100644
index 2ca32c6a59..0000000000
--- a/docs/website/docs/sdk/05_ETDB.md
+++ /dev/null
@@ -1,101 +0,0 @@
- # Debugging with ETDB
-
-ETDB (ExecuTorch Debugger) is an interactive text-based tool used for investigating models (`ETRecord`) and profiling information (`ETDump`). Features provided include:
-- **Tabular Visualization** of model graphs
-- **Drill-in Selection** of components for detailed investigation
-- **Aggregated Operator Statistics** based on results from ETDump
-- **Module Architecture** grouped by a node's immediate SubModule
-
-To kick off an ETDB instance follow these steps:
-
-
-1. Generate an etrecord with the exported graph modules of your model that you want to visualize. Refer to the [etrecord tutorial](./01_generating_etrecord.md) for more details on how to do this.
-
-2. *Optionally* generate an etdump to also visualize the performance data of your model run along with the graph. Refer to the [etdump tutorial](./02_generating_etdump.md) for more details on how to do this.
-
-3. Initiate the ETDB instance by calling a `debug_etrecord` API, passing in the ETrecord you generated in step 1 and also the optional ETdump you generated in step 2. This will kick off the text tool. See [library documentation](./03_using_sdk_cli_tools.md) for API details
-
-For a complete end-to-end tutorial on how to do this please refer to this [notebook](https://www.internalfb.com/intern/anp/view/?id=3799219).
-
----
-
-## ETDB Flow
-
-There are 3 stages to the ETDB flow:
-1. **Graph Selection**: Prompts for which model graph to investigate
-2. **Graph Overview**: Prints an overview of the selected graph
-3. **Graph Investigation**: Interactive mode; User input is used to generate a focused view of the graph
-
-
-### Graph Selection
-
-Given a list of the graphs found in ETRecord, choose one to investigate.
-
-Note: In the future it will be possible to toggle between graphs during the Graph Investigation stage.
-
-### Graph Overview
-
-Once a graph is selected, a set of overview tables will be displayed. Specifically the following:
-- **Graph Counts**: Graph input + output counts, and a breakdown of node types in the model
-- **Enumerated Nodes**: Tables listing each of the top level model nodes
-- **Aggregated Summaries**: An aggregated summary table grouped by op/module type
-
-If profiling information was attached to this graph, a **Run Summary** table with aggregated run stats will also be shown.
-
-
-### Graph Investigation
-
-Graph Investigation is the interactive segment of ETDB. With reference to the row entries in **Graph Overview**, ETDB takes as input any of the following:
-- **Node Name** (e.g. add\_tensor_)
-- **Module Name** (e.g. l_\_self___model_backbone_encoder_module)
-- **Op Type** (e.g. add.Tensor)
-- **Module Type** (e.g. Linear)
-
-From the selection, a new set of tables containing related information will be generated.
-
----
-
-#### Selection: Node Name
-
-If an node name is provided, a few things are returned:
-- **Related Nodes**: Tables containing the inputs/outputs of the selected Node
-- **Parent Module Summary**: Basic information about this node's parent module (if exists)
-- **Operator Summary**: Count (and summary stats if ETDump is provided) of this node's type
-
-For example, if ETDump was provided and `addmm_default_` was selected:
-- Related Nodes: Tables of Input Nodes, Tables of Output Nodes
-- Parent Module Summary: Summary of `l__self___linear`
-- Operator Summary: # of `addmm.default` instances in this model and runtime stats of those instances (e.g. Avg, P10, P90)
-
----
-
-#### Selection: Module Name
-
-If an module name is provided, a few things are returned:
-- **Related Nodes**: Tables containing the inputs/outputs of the selected module
-- **Module Contents**: Table listing all nodes within this module instance
-- **Parent Module Summary**: Basic information about this node's parent module (if exists)
-- **Module Summary**: Count (and summary stats if ETDump is provided) of this module's type
-
-For example, if ETDump was provided and `l__self___linear` was selected:
-- Related Nodes/Modules: Tables of Input Nodes, Tables of Output Nodes
-- Module Contents: List of all nodes in `l__self___linear`
-- Parent Module Summary: Summary of `l__self___wrapped_linear`
-- Module Summary: # of instances of `Linear`
-
----
-
-#### Selection: Op Type + Module Type
-
-If either an op type or module type are provided, two tables are returned:
-- **Aggregated Stats**: Count (and summary stats if ETDump is provided) for that node type
-- **Enumerated Nodes**: Table listing all nodes of the selected type
-
-For example, if ETDump was provided and `slice_copy.Tensor_out` was selected:
-- Aggregated Stats: # of `slice_copy.Tensor_out` instances and runtime stats of those instances (e.g. Avg, P10, P90)
-- Enumerated Nodes: List of all `slice_copy.Tensor_out` instances in this model
----
-## Ease of Use
-
-Below are a few ease of use features
-- **Undo Investigation**: Entering `b` or `back` while in the Graph Investigation flow, replays the last user input provided
diff --git a/docs/website/docs/sdk/06_aibench_sdk.md b/docs/website/docs/sdk/06_aibench_sdk.md
deleted file mode 100644
index 5d5fcf9a56..0000000000
--- a/docs/website/docs/sdk/06_aibench_sdk.md
+++ /dev/null
@@ -1,23 +0,0 @@
-# Using AIBench along with the SDK
-
-AIBench is a benchmarking framework used to execute binaries on-device and collect output. It's an intuitive tool for on-device workloads, including testing, benchmarking ML models, measuring PyTorch operator performance, and scheduling continuous benchmark runs.
-Get started at the [AIBench wiki](https://www.internalfb.com/intern/wiki/AIBench/Mobile_Overview/Submitting_a_Job/)
-
-If you're interested in benchmarking your model on a device using AIBench and then visualizing the performance data along with the model visualization then the `ExecutorchAIBenchAPI` is your one-stop solution for all of this.
-
-AIBench handles all of the plumbing required to generate an ETdump and also to subsequently generate the visualization along with the performance data.
-Pass in the ETRecord (please refer to [this page](./01_generating_etrecord.md) for instructions on how to generate en ETrecord) along with your device selection to the `ExecutorchAIBenchAPI`.
-
-Behind the scenes we will then run your model on the device you selected, fetch the etdump that was generated by the runtime, generate the visualization along with the performance data visualization on a TensorBoard and the link to this will be populated on the AIBench Job Result page.
-
-### Sample Notebook:
-A sample notebook that goes through the entire flow of generating an ETRecord and then running an AIBench job using that. [Link](https://www.internalfb.com/intern/anp/view/?id=3799219)
-
-### Visualization and ETDump links in AIBench job result page:
-As shown in the image below you can click on the link that says "Open etdump.etdp in TensorBoard" which will take you to the TensorBoard page with the visualization.
-
-Alternatively if you wish to you can also download the etdump from the link thay says "etdump.etdp" and use it in our command line tooling.
-
-
- 
-
diff --git a/docs/website/docs/sdk/README.md b/docs/website/docs/sdk/README.md
deleted file mode 100644
index c22e70e138..0000000000
--- a/docs/website/docs/sdk/README.md
+++ /dev/null
@@ -1,4 +0,0 @@
-# Deprecated folder
-
-This folder is deprecated. For up-to-date comprehensive technical overview of ExecuTorch and step-by-step tutorials,
-please visit our [documentation website](https://pytorch.org/executorch).
diff --git a/docs/website/docs/tutorials/00_setting_up_executorch.md b/docs/website/docs/tutorials/00_setting_up_executorch.md
deleted file mode 100644
index 6277e2c49e..0000000000
--- a/docs/website/docs/tutorials/00_setting_up_executorch.md
+++ /dev/null
@@ -1,163 +0,0 @@
-
-
-# Setting up ExecuTorch
-
-This is a tutorial for building and installing ExecuTorch from the GitHub
-repository.
-
-## Ahead-of-Time Setup [(Open on Google Colab)](https://colab.research.google.com/drive/1m8iU4y7CRVelnnolK3ThS2l2gBo7QnAP#scrollTo=1o2t3LlYJQY5)
-
-This will install an `executorch` pip package to your conda environment and
-allow you to export your PyTorch model to a flatbuffer file using ExecuTorch.
-
-### Step 1: Clone the ExecuTorch repo
-
-```bash
-# Do one of these, depending on how your auth is set up
-git clone https://github.com/pytorch/executorch.git
-git clone git@github.com:pytorch/executorch.git
-```
-
-Ensure that git has fetched and updated the submodules. This is necessary
-anytime commit hash of any of the submodules changes. Thus it is safe and
-necessary at times to apply this step after you pull changes from upstream.
-
-```bash
-cd executorch
-git submodule sync
-git submodule update --init
-```
-
-### Step 2: Set up a dev environment
-
-Install
-[Conda](https://conda.io/projects/conda/en/latest/user-guide/install/index.html)
-and activate a new virtual environment:
-
-```bash
-conda create -yn executorch python=3.10.0
-conda activate executorch
-```
-
-Or alternatively, activate a new Python virtual environment:
-
-```bash
-python3 -m venv .executorch
-source .executorch/bin/activate
-```
-
-### Step 3: Install [Cmake](https://cmake.org/download)
-
-```bash
-conda install cmake
-```
-
-Alternatively:
-
-```bash
-pip install cmake
-```
-
-### Step 4: Install `executorch` pip package and dependencies
-
-Install all required python dependencies and PyTorch dependencies.
-
-```bash
-# Note: if you are behind a firewall an appropriate proxy server must be setup
-# for all subsequent steps.
-./install_requirements.sh
-```
-
-When getting a new version of the executorch repo (via clone, fetch, or pull),
-you may need to re-install a new version the PyTorch nightly pip package. The
-`TORCH_VERSION` value in this document will be the correct version for the
-corresponsing version of the repo.
-
-### Step 5: Expose FlatBuffers compiler
-
-ExecuTorch uses `flatc` to export models and builds it from sources at
-`third-party/flatbuffers`. Make it's available by referring in `$PATH`,
-as prompted by the previous step, or exporting as `$FLATC_EXECUTABLE`
-enironment variable.
-Run `./build/install_flatc.sh` to make sure `flatc` is installed correctly.
-
-## Testing Model Export
-
-Generate a test program file from an `nn.Module` via Python script:
-
-```bash
-# Creates the file `add.pte`
-python3 -m examples.portable.scripts.export --model_name="add"
-
-# Creates the delegated program `composite_model.pte`, other options are "whole" and "partition"
-python3 -m examples.portable.scripts.export_and_delegate --option "composite"
-```
-
-Or via Python interpreter:
-
-```python
-$ python3
->>> import executorch.exir as exir
->>> from executorch.exir.tests.models import Mul
->>> m = Mul()
->>> print(exir.capture(m, m.get_random_inputs()).to_edge())
->>> open("mul.pte", "wb").write(exir.capture(m, m.get_random_inputs()).to_edge().to_executorch().buffer)
-```
-
-Please refer to the [More Examples](./00_setting_up_executorch.md#more-examples)
-section for running with more popular models.
-
-## Runtime Setup
-
-Follow
-[Ahead-of-Time Setup: Step 1](./00_setting_up_executorch.md#step-1-clone-the-executorch-repo)
-above to clone the `executorch` repo if you haven't already.
-
-### Step 1: Install buck2
-
-- If you don't have the `zstd` commandline tool, install it with `pip install zstd`
-- Download a prebuilt buck2 archive for your system from the https://github.com/facebook/buck2/releases/tag/2023-07-18 release page. Note that the version is important, and newer or older versions may not work with the version of the buck2 prelude used by the ExecuTorch repo.
-- Decompress with the following command (filename depends on your system)
-
-```bash
-# For example, buck2-x86_64-unknown-linux-musl.zst or buck2-aarch64-apple-darwin.zst
-zstd -cdq buck2-DOWNLOADED_FILENAME.zst > /tmp/buck2 && chmod +x /tmp/buck2
-```
-
-You may want to copy the `buck2` binary into your `$PATH` so you can run it as `buck2`.
-
-### Step 2: Build a binary
-
-`executor_runner` is an example wrapper around executorch runtime which includes all the operators and backends
-
-```bash
-/tmp/buck2 build //examples/portable/executor_runner:executor_runner --show-output
-```
-
-The `--show-output` flag will print the path to the executable if you want to run it directly.
-
-If you run into `Stderr: clang-14: error: invalid linker name in argument '-fuse-ld=lld'`, do
-```bash
-conda install -c conda-forge lld
-```
-
-### Step 3: Run a binary
-
-```bash
-# add.pte is the program generated from portable/scripts/export.py during AOT Setup Step 3
-/tmp/buck2 run //examples/portable/executor_runner:executor_runner -- --model_path add.pte
-
-# To run a delegated model
-/tmp/buck2 run //examples/portable/executor_runner:executor_runner -- --model_path composite_model.pte
-```
-
-or execute the binary directly from the `--show-output` path shown when building.
-
-```bash
-./buck-out/.../executor_runner --model_path add.pte
-```
-
-## More Examples
-
-The [`executorch/examples`](https://github.com/pytorch/executorch/blob/main/examples) directory contains useful examples with a guide to lower and run
-popular models like MobileNet V3, Torchvision ViT, Wav2Letter, etc. on ExecuTorch.
diff --git a/docs/website/docs/tutorials/README.md b/docs/website/docs/tutorials/README.md
deleted file mode 100644
index c22e70e138..0000000000
--- a/docs/website/docs/tutorials/README.md
+++ /dev/null
@@ -1,4 +0,0 @@
-# Deprecated folder
-
-This folder is deprecated. For up-to-date comprehensive technical overview of ExecuTorch and step-by-step tutorials,
-please visit our [documentation website](https://pytorch.org/executorch).
diff --git a/docs/website/docs/tutorials/aten_ops_and_aten_mode.md b/docs/website/docs/tutorials/aten_ops_and_aten_mode.md
deleted file mode 100644
index 42686d42fa..0000000000
--- a/docs/website/docs/tutorials/aten_ops_and_aten_mode.md
+++ /dev/null
@@ -1,182 +0,0 @@
-# ATen-compliant Operator Registration & ATen mode
-
-
-## Introduction
-
-ExecuTorch supports a subset of ATen-compliant operators.
-ATen-compliant operators are those defined in
-[`native_functions.yaml`](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/native_functions.yaml),
-with their native functions (or kernels, we use these two terms interchangeably)
-either defined in ATen library or other user defined libraries. The ATen-compliant operators supported by ExecuTorch have these traits (actually same for custom ops):
-1. Out variant, means these ops take an `out` argument
-2. Functional except `out`. These ops shouldn't mutate input tensors other than `out`, shouldn't create aliasing views.
-
-To give an example, `aten::add_.Tensor` is not supported since it mutates an input tensor, `aten::add.out` is supported.
-
-ATen mode is a build-time option to link ATen library into ExecuTorch runtime, so those registered ATen-compliant ops can use their original ATen kernels.
-
-On the other hand we need to provide our custom kernels if ATen mode is off (a.k.a. lean mode).
-
-In the next section we will walk through the steps to register ATen-compliant ops into ExecuTorch runtime.
-
-## Step by step guide
-There are two branches for this use case:
-* ATen mode. In this case we expect the exported model to be able to run with ATen kernels .
-* Lean mode. This requires ATen-compliant op implementations using `ETensor`.
-
-In a nutshell, we need the following steps in order for a ATen-compliant op to work on ExecuTorch:
-
-#### ATen mode:
-1. Define a target for selective build (`et_operator_library` macro)
-2. Pass this target to codegen using `executorch_generated_lib` macro
-3. Hookup the generated lib into ExecuTorch runtime.
-
-For more details on how to use selective build, check [Selective Build](https://www.internalfb.com/intern/staticdocs/executorch/docs/tutorials/custom_ops/#selective-build).
-#### Lean mode:
-1. Declare the op name in `functions.yaml`. Detail instruction can be found in [Declare the operator in a YAML file](https://www.internalfb.com/code/fbsource/xplat/executorch/kernels/portable/README.md).
-2. (not required if using ATen mode) Implement the kernel for your operator using `ETensor`. ExecuTorch provides a portable library for frequently used ATen-compliant ops. Check if the op you need is already there, or you can write your own kernel.
-3. Specify the kernel namespace and function name in `functions.yaml` so codegen knows how to bind operator to its kernel.
-4. Let codegen machinery generate code for either ATen mode or lean mode, and hookup the generated lib into ExecuTorch runtime.
-
-### Case Study
-Let's say a model uses an ATen-compliant operator `aten::add.out`.
-
-We can either reuse the kernel written in portable library [here](https://www.internalfb.com/code/fbsource/xplat/executorch/kernels/portable/cpu/op_add.cpp) or choose to write our own.
-
-#### ATen mode
-
-For ATen mode we don't need to define any `functions.yaml` file or write any kernel, since they are already defined in `native_functions.yaml`. All we need to do is to use selective build to choose the ops we want:
-```python
-et_operator_library(
- name = "selected_ops",
- ops = [
- "aten::add.out",
- ],
-)
-
-executorch_generated_lib(
- name = "add_lib",
- deps = [
- ":selected_ops",
- ],
- aten_mode = True,
-)
-```
-
-#### Lean mode
-
-Let's say we like to write our own kernel:
-
-```cpp
-// add.cpp
-namespace custom {
-namespace native {
-
-Tensor& add_out(const Tensor& a, const Tensor& b, const Scalar& alpha, Tensor& out) {
- // do something
- return out;
-}
-
-} // namespace native
-} // namespace custom
-```
-
-The corresponding `functions.yaml` for this operator looks like:
-
-```yaml
-- op: add.out
- dispatch:
- CPU: custom::add_out
-```
-Notice that there are some caveats:
-#### Caveats
-* `dispatch` and `CPU` are legacy fields and they don't mean anything in ExecuTorch context.
-* Namespace `aten` is omitted.
-* We don't need to write `aten::add.out` function schema because we will use the schema definition in `native_functions.yaml` as our source of truth.
-* Kernel namespace in the yaml file is `custom` instead of `custom::native`. This is because codegen will append a `native` namespace automatically. It also means the kernel always needs to be defined under `k)}Uqp?BdsEA1Xm%g;MQr