Updated cohere notebook for video

[maxhniebergall]

building on https://www.elastic.co/search-labs/blog/elasticsearch-cohere-rerank

[gitnotebooks]

Found 1 changed notebook. Review the changes at https://gitnotebooks.com/elastic/elasticsearch-labs/pull/317

[joemcelroy]

nitpick: rename the notebook to cohere-elasticsearch-inference-api.ipynb

You also need to exclude this notebook from our tests https://github.com/elastic/elasticsearch-labs/blob/main/bin/find-notebooks-to-test.sh#L3

Will review once CI passes! You might have issue with formatting on CI so follow contributing guide

[jeffvestal]

This example uses an API key, not a password.

[jeffvestal]

When I ran the ingest step I got an index error. The rest of the notebook worked.

---------------------------------------------------------------------------
BulkIndexError                            Traceback (most recent call last)
[<ipython-input-7-044c75f1a943>](https://localhost:8080/#) in <cell line: 15>()
     16     if i % 1000 == 999:
     17         print(i, "of", 13000)
---> 18         print(helpers.bulk(client, documents))  # Use the bulk endpoint to index
     19         print("last doc:", documents[-1])
     20         documents = []

3 frames
[/usr/local/lib/python3.10/dist-packages/elasticsearch/helpers/actions.py](https://localhost:8080/#) in _process_bulk_chunk_success(resp, bulk_data, ignore_status, raise_on_error)
    272 
    273     if errors:
--> 274         raise BulkIndexError(f"{len(errors)} document(s) failed to index.", errors)
    275 
    276 

BulkIndexError: 305 document(s) failed to index.

[jeffvestal]

Looks good!
My only small suggestion, take or leave it, would be to explain the last part a little more. You do above the cell but inline comments may help people unfamiliar with this type of content

Something like

...
# Use semantic search with our previously configured reranker to find the best set of matching recipes
    recipes = extract_recipe_as_string(semantic_search_with_reranking(query))

    if len(recipes) > 0:
        # if there are one or more recipes that match, combine them together in a string along with our "personal assistant" instructions
        input.append("Recipes:" + "\n\n".join(recipes))

        # pass the full combined instructions and recipes to the Command+R model 
        chat_completion = client.inference.inference(
            inference_id="cohere_completion", input="\n\n\n".join(input)
        )

        # print the response for the user to answer their recipe question
        print(chat_completion["completion"][0]["result"])
    else:
        print("No hits")```

[maxhniebergall]

@jeffvestal , were you using a cohere production API key or a trial key?

[jeffvestal]

@jeffvestal , were you using a cohere production API key or a trial key?

I think it was a prod key, but let me run through it again with a new prod key to double-check.

[maxhniebergall]

Thanks! I am just trying to think of reasons that some of the documents would fail to ingest. It worked every time I tried it.

[jeffvestal]

Alright, @maxhniebergall it must have been me using the wrong key 🫠
I ran it again on a fresh serverless es project with a new prod cohere key and it worked with no errors!

[maxhniebergall]

Thanks Jeff!