model_summary.md - Restore link to Harvard's Annotated Transformer. (#29702)

* model_summary.md - Add link to Harvard's Annotated Transformer.

* model_summary.md - slight wording change + capitalize name of the paper

* model_summary.md - moves the Annotated Transformer link in a praenthesis next to the link to the original paper (great idea, stevhliu!)

* model_summary.md - moves the Annotated Transformer link in a praenthesis next to the link to the original paper (commit pt. 2, accidentally removed "has" in pt. 1)

Author: gamepad-coder

docs/source/en/model_summary.md

@@ -16,7 +16,7 @@ rendered properly in your Markdown viewer.
 
 # The Transformer model family
 
-Since its introduction in 2017, the [original Transformer](https://arxiv.org/abs/1706.03762) model has inspired many new and exciting models that extend beyond natural language processing (NLP) tasks. There are models for [predicting the folded structure of proteins](https://huggingface.co/blog/deep-learning-with-proteins), [training a cheetah to run](https://huggingface.co/blog/train-decision-transformers), and [time series forecasting](https://huggingface.co/blog/time-series-transformers). With so many Transformer variants available, it can be easy to miss the bigger picture. What all these models have in common is they're based on the original Transformer architecture. Some models only use the encoder or decoder, while others use both. This provides a useful taxonomy to categorize and examine the high-level differences within models in the Transformer family, and it'll help you understand Transformers you haven't encountered before.
+Since its introduction in 2017, the [original Transformer](https://arxiv.org/abs/1706.03762) model (see the [Annotated Transformer](http://nlp.seas.harvard.edu/2018/04/03/attention.html) blog post for a gentle technical introduction) has inspired many new and exciting models that extend beyond natural language processing (NLP) tasks. There are models for [predicting the folded structure of proteins](https://huggingface.co/blog/deep-learning-with-proteins), [training a cheetah to run](https://huggingface.co/blog/train-decision-transformers), and [time series forecasting](https://huggingface.co/blog/time-series-transformers). With so many Transformer variants available, it can be easy to miss the bigger picture. What all these models have in common is they're based on the original Transformer architecture. Some models only use the encoder or decoder, while others use both. This provides a useful taxonomy to categorize and examine the high-level differences within models in the Transformer family, and it'll help you understand Transformers you haven't encountered before.
 
 If you aren't familiar with the original Transformer model or need a refresher, check out the [How do Transformers work](https://huggingface.co/course/chapter1/4?fw=pt) chapter from the Hugging Face course.
 
@@ -104,4 +104,4 @@ Optical character recognition (OCR) is a long-standing text recognition task tha
 
 ### Decoder[[rl-decoder]]
 
-The Decision and Trajectory Transformer casts the state, action, and reward as a sequence modeling problem. The [Decision Transformer](model_doc/decision_transformer) generates a series of actions that lead to a future desired return based on returns-to-go, past states, and actions. For the last *K* timesteps, each of the three modalities are converted into token embeddings and processed by a GPT-like model to predict a future action token. [Trajectory Transformer](model_doc/trajectory_transformer) also tokenizes the states, actions, and rewards and processes them with a GPT architecture. Unlike the Decision Transformer, which is focused on reward conditioning, the Trajectory Transformer generates future actions with beam search.
+The Decision and Trajectory Transformer casts the state, action, and reward as a sequence modeling problem. The [Decision Transformer](model_doc/decision_transformer) generates a series of actions that lead to a future desired return based on returns-to-go, past states, and actions. For the last *K* timesteps, each of the three modalities are converted into token embeddings and processed by a GPT-like model to predict a future action token. [Trajectory Transformer](model_doc/trajectory_transformer) also tokenizes the states, actions, and rewards and processes them with a GPT architecture. Unlike the Decision Transformer, which is focused on reward conditioning, the Trajectory Transformer generates future actions with beam search.