complete

docs/source/community/contributing.mdx

@@ -12,7 +12,7 @@ specific language governing permissions and limitations under the License.
 
 # Adding support for new architectures
 
-You want to export and run a new model on AWS Trainium or Inferentia? Check the guideline, and submit a pull request to [🤗 Optimum Neuron's github repo](https://github.com/huggingface/optimum-neuron/)! 
+You want to export and run a new model on AWS Inferentia or Trainium? Check the guideline, and submit a pull request to [🤗 Optimum Neuron's github repo](https://github.com/huggingface/optimum-neuron/)! 
 
 To support a new model architecture in the Optimum Neuron library here are some steps to follow:
 
@@ -23,33 +23,31 @@ To support a new model architecture in the Optimum Neuron library here are some
 
 ## Implement a custom Neuron configuration
 
-To support the [export of a new model to Neuron compatible format](../), the first thing to do is to define a Neuron configuration, describing how to expot a PyTorch model to a Neuron model, by  specifying:
+To support the export of a new model to Neuron compatible format, the first thing to do is to define a Neuron configuration, describing how to expot a PyTorch model to a Neuron compatible format, by specifying:
 
 1. The input names.
 2. The output names.
-3. The dummy inputs used to trace the model. This is needed by the Neuron Compiler to record the computational graph and embed it to a TorchScript module.
+3. The dummy inputs used to trace the model. (This is needed by the Neuron Compiler to record the computational graph and embed it to a TorchScript module.)
 4. The compilation arguments used to control the trade-off between hardware efficiency (latency, throughput) and accuracy.
 
-Depending on the choice of model and task, we represent the data above with _configuration classes_. Each configuration class is associated with
+Depending on the choice of model and task, we represent the data above with configuration classes. Each configuration class is associated with
 a specific model architecture, and follows the naming convention `ArchitectureNameNeuronConfig`. For instance, the configuration which specifies the Neuron
 export of BERT models is `BertNeuronConfig`.
 
 Since many architectures share similar properties for their Neuron configuration, 🤗 Optimum adopts a 3-level class hierarchy:
 
 1. Abstract and generic base classes. These handle all the fundamental features, while being agnostic to the modality (text, image, audio, etc).
-2. Middle-end classes. These are aware of the modality, but multiple can exist for the same modality depending on the inputs they support.
-   They specify which input generators should be used for the dummy inputs, but remain model-agnostic.
+2. Middle-end classes. These are aware of the modality. Multiple config classes could exist for the same modality depending on the inputs they support. They specify which input generators should be used for generating the dummy inputs, but remain model-agnostic.
 3. Model-specific classes like the `BertNeuronConfig` mentioned above. These are the ones actually used to export models.
 
 ### Example: Adding support for ESM models
 
-Here we take the suppport of [ESM models](https://huggingface.co/docs/transformers/model_doc/esm#esm) as an example. Create a `EsmNeuronConfig` class in the `optimum/exporters/neuron/model_configs.py`.
+Here we take the suppport of [ESM models](https://huggingface.co/docs/transformers/model_doc/esm#esm) as an example. Let's create a `EsmNeuronConfig` class in the `optimum/exporters/neuron/model_configs.py`.
 
-When it interprets as a text encoder, we are able to inherit from the middle-end class [`TextEncoderNeuronConfig`](https://github.com/huggingface/optimum-neuron/blob/v0.0.18/optimum/exporters/neuron/config.py#L36). 
-Since the modeling and configuration of Esm is almost the same as BERT when it is interpreted as an encoder, we can use the `NormalizedConfigManager` with `model_type=bert` to normalize the configuration 
-in order to generate dummy inputs for tracing the model.
+When a Esm model interprets as text encoder, we are able to inherit from the middle-end class [`TextEncoderNeuronConfig`](https://github.com/huggingface/optimum-neuron/blob/v0.0.18/optimum/exporters/neuron/config.py#L36). 
+Since the modeling and configuration of Esm is almost the same as BERT when it is interpreted as encoder, we can use the `NormalizedConfigManager` with `model_type=bert` to normalize the configuration in order to generate dummy inputs for tracing the model.
 
-And one last step, use the `register_in_tasks_manager` decorator to register your newly created Neuron config to the Optimum's [TasksManager](https://huggingface.co/docs/optimum/main/en/exporters/task_manager#optimum.exporters.TasksManager), with supported tasks.
+And one last step, since `optimum-neuron` is an extension of `optimum`, we need to register the Neuron config that we create to the [TasksManager](https://huggingface.co/docs/optimum/main/en/exporters/task_manager#optimum.exporters.TasksManager) with the `register_in_tasks_manager` by specifying the model type and supported tasks.
 
 ```python
 
@@ -68,10 +66,24 @@ class EsmNeuronConfig(TextEncoderNeuronConfig):
 
 With the Neuron config that you implemented, now do a quick test if it works as expected:
 
+* Export
+
 ```bash
 optimum-cli export neuron --model facebook/esm2_t33_650M_UR50D --task text-classification --batch_size 1 --sequence_length 16 esm_neuron/
 ```
 
+And then validate the outputs of your exported Neuron model by comparing to the results of PyTorch on CPU.
+
+```python
+from optimum.exporters.neuron import validate_model_outputs
+
+validate_model_outputs(
+    neuron_config, base_model, neuron_model_path, neuron_named_outputs, neuron_config.ATOL_FOR_VALIDATION
+)
+```
+
+* Inference
+
 ```python
 from transformers import AutoTokenizer
 from optimum.neuron import NeuronModelForSequenceClassification
@@ -84,11 +96,17 @@ logits = model(**inputs).logits
 
 ## Add tests
 
-Add the model to the exporter test in tests/exporters/exporters_utils.py  
+Add the model to the exporter test in [`optimum-neuron/tests/exporters/exporters_utils.py`](https://github.com/huggingface/optimum-neuron/blob/v0.0.18/tests/exporters/exporters_utils.py) and the inference test in [`optimum-neuron/tests/inference/inference_utils.py`](https://github.com/huggingface/optimum-neuron/blob/v0.0.18/tests/inference/inference_utils.py).
+
+<Tip>
+
+We usually test a smaller checkpoint to accelerate the CIs, you could find tiny models for testing under the [`Hugging Face Internal Testing Organization`](https://huggingface.co/hf-internal-testing).
+
+</Tip>
 
 ## Contribute to the github repo
 
-Open the pull request and ping one Optimum-neuron maintainer to review your PR.
+Now, just we are all set! Open a pull request and ping one Optimum-neuron maintainers to review your PR.
 
 
 You have made a new model accessible on Neuron for the community! Thanks for joining us in the endeavor of democratizing good machine learning 🤗.