Hi everyone! I was curious to see if anyone has tried using any of the Nixtla packages with the lag-llama model?

Hello! I am not sure what you mean by using Nixtla with Lag-Llama, but we have benchmarked it against other methods and you can see the fully reproducible results here: https://github.com/Nixtla/nixtla/tree/main/experiments/lag-llama

Awesome, thanks @Marco!

It is to reproduce the same conditions as presented in the original paper of Lag-Llama.

Ok gotcha, that makes sense. Outside of reproducing those conditions, would it be possible to run this model with MLForecast? For example, maybe in a way similar to this example? https://nixtlaverse.nixtla.io/mlforecast/docs/how-to-guides/custom_training.html If there are any other examples or advice you have, please feel free to share!

We have included some other foundation models in neuralforecast, but we don't plan on doing it for Lag-Llama, at least to my knowledge. However, PRs are welcome, and you can always take some inspiration from how we integrated Time-LLM: https://github.com/Nixtla/neuralforecast/blob/main/nbs/models.timellm.ipynb

Thanks so much!