search

time-series-foundation-models / lag-llama

Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting
Apache License 2.0
1.08k stars 121 forks source link

Make Lag-Llama pip-installable #80

Closed araghukas closed 1 week ago

araghukas commented 1 week ago

Hey there!

Some context:

We recently used lag-llama to write a tutorial for Covalent, which is the compute platform that we develop at Agnostiq. What we did there was to host the lag-llama model as a Covalent service, then use it to reproduce a simple result from your zero-shot google colab notebook.

In order to do this, we had to add a pyproject.toml file and edit the existing requirements.txt (some classes have moved around in gluonts and were no longer importable as per the notebook). These changes (mainly the pyproject file) made lag-llama pip-installable (via the github repo URL or a local clone, either way, in the usual manner).

This PR:

The changes here include the aforementioned. No library code was modified. Only the handful of changes to make the project pip-installable.

I was curious if there was a good reason for not having this on your end? Please let me know if you can.

Cheers.

ashok-arjun commented 1 week ago

Hi @araghukas ! Thank you so much for the PR!

Making lag-llama pip-installable was on our roadmap; it just got sidetracked by other stuff :)

Thanks for the link to the tutorial too - it looks exciting! Would also be separately curious to know if it works well for you (i.e. on your data or usecases) or not, so we can inculcate the feedback as we are designing new foundation models ;)

Merging the PR as it looks good as is 🚀 I will also add an update acknowledging the contribution in the README. Thanks again! CC @kashif.

ashok-arjun commented 1 week ago

@araghukas Just to confirm, is it correct that now we can install lag-llama using pip install git+https://github.com/time-series-foundation-models/lag-llama.git?

I can add a note on the README about this.