Add new colpali example

[jobergum]

I confirm that this contribution is made under the terms of the license found in the root directory of this repository's source tree and that I have the authority necessary to make this contribution on behalf of its copyright owner.

[jobergum]

Input welcome on this, both language and code, also would be nice to see if we can get this into CI/CD

[thomasht86]

Very cool! 🤩
Was able to run it on Colab ✅

Some comments:

• Making it run on CI
  As a start, we could add

!sudo apt-get install poppler-utils -y

Should also move the note about poppler dependency there, and maybe link to other install options. .
But, it will most likely take prohibitively long to run this on CPU. Possible to scale down number of docs/images to a minimum if running on CPU?

• The cell below
  

Takes 1.5*15=22.5 mins to run in Colab with A100 GPU. 🐌
(Didn't bother to wait with T4.) Perhaps add a note about that. ☕

• Could be nice to add to the top (I did this for the other cloud nbs in the big batch update)

• The cell below

Causes linting error in Colab (and older versions of eg. pylance Consider using https://pyvespa.readthedocs.io/en/latest/reference-api.html#vespa.application.Vespa.feed_async_iterable for feeding at least.

• tenant_name = "samples" would need to change to vespa-team to run in CI.

[jobergum]

Great, thanks for this! Will iterate

Takes 1.5*15=22.5 mins to run in Colab with A100 GPU.

This is super weird as it takes just 12-15 seconds per batch on my M1 with MPS

[thomasht86]

Looks like the latest run encountered this hf issue

[thomasht86]

Great. Ran successfully in 25mins - we can add it to CI. 👍
(Wonder what my issue on Colab was though).

Consider less printout of b64-strings and embeddings (will force the reader to scroll a lot).

[thomasht86]

Made this for my own understanding.
May be useful for others. 😄
Feel free to add (or dm if you want to modify).

[jobergum]

It's a good overview diagram! Removed the larger output and also looks like cd passed

[thomasht86]

🙌