Thanks, Sam, for the informative videos on using Google Colab! I have a question regarding incremental training and saving checkpoints when working with the 24-hour notebook runtime limit on Colab Pro. For those on a budget who want to train LLMs over an extended period, the ability to save checkpoints and resume training from those checkpoints is crucial. This would allow us to train incrementally across multiple Colab sessions without losing progress. I'm wondering if you could cover strategies for implementing this in a future video? Specifically, how to save model checkpoints to a storage location like HuggingFace, Google Storage, or Google Drive before the 24-hour limit, and then reload those checkpoints to continue training in the next session. While I understand this would require some manual tweaking, having a reliable approach for incremental training on Colab Pro would be incredibly valuable for those of us working with limited resources. I'd really appreciate it if you could explore this topic more in-depth.

Good topic selection, great video! :) 1) would you share with us your Colab snippets? :) 2) topic of an Ollama server hosted on Colab with mounted Google Drive with permanent LLM storage could be useful for most of us (being able to deploy it quickly and easily on different GPUs).

Great tips. Thanks

Very useful, thank you for it all, Sam.

That is very helpful thank you Sam

whats the differemce with github Codespaces?

Do you use this primarily for experimenting or is it possible to deploy from Collab to a production app on Google Cloud?

the AI assistant is not avail in EU for the moment (but can reach with VPN, as usual)

Can you use Colab AI in Thailand yet? Last time I checked (a few months ago), you cannot. The current documentation says it's only available in US. (the docs might not be updated, because you said it's available in other countries too).