Last pass for NLP tasks pages (#568)
* Misc fill-mask changes
- Remove mention of T5 and BERT in dataset descriptions since that requires previous knowledge
- Reworded a bit DistilBERT and metrics descriptions
- Added link to course webinar
- Added output to code snippet
* Minor QA wording and formatting changes
* Misc sentence-similarity changes
- Simplified explanation of datasets and metrics
- Changed flax-sentence-embeddings/stackexchange_titlebody_best_voted_answer_jsonl to squad which is the to-go dataset for this
- Changed all-mpnet-base-v2 description since it assumed sentence-transformers knowledge
- Added sentence transformers section
* Misc summarization changes
(This was a very nice section! I just made some minor changes to keep things more concise)
- Made description more concise
- Removed some of the text in ROUGE to keep it shorter
- Changed the use case to be a list so it's easier for the eyes
* Misc text classification changes
- Reworded a bit initial sentence
- Fixed schema, it was not having right output
- Consistent capitalization of tasks
- You say NLI is the largest text classification variant but I think that can be debatable, I reworded and structure a bit more. I felt the current structure had too much text and not necessarily informative for our audience.
- Removed mention to SNLI since people don't know what that is
- Removed Winograd to keep things more concise
- "The task is evaluated on Stanford Sentiment Treebank." is really dependant on the task imo, so removed that
- Added link to GLUE, QQP, COLA
- Renamed linguistic acceptability to Grammatical Correctness to be more intuitive (still mention linguistic acceptability though)
* Make resources section consistent
* Misc text generation changes
(Very nice section as well)
- Reworded a bit the one-liner
- Changed dataset to the pile which is now in datasets
- Updated metrics descriptions to be consistent with fill-mask
- Removed legal documents use case since that one is sensitive
- Made content of inference more concise
* Misc token classification changes
- Changed one example model to a more popular one
- Changed widget to a more common model so it's cached and runs faster
- Minor changes
* Misc translation changes
- Moves some of the initial description to the about
- Fixed title of conversational agents
- Changed use case to a bit more readable format
* Fix model description
