I’m experiencing OoM errors when trying to run the gpt-oss 20b model on Google Colab’s free T4 GPU.
I wonder if there are any good workarounds for this?

Just read this article: magazine.sebastianraschka.com/p/from-gpt-2...
Since I recently tried implementing the GPT-2 architecture from scratch, this article's approach to highlight the differences of gpt-oss from GPT-2 was easy to follow. As the article mentions, GPT-2 is a great starting point.

Recently re-implementing GPT-2 from scratch! Before diving in, I only had a vague understanding of how LLMs work-still learning this field. But after hands-on implementation, my grasp of embeddings, multi-head attention, and GPU parallelization became much clearer. Really grateful for the tutorial!