@nabendu82 https://t.co/DYUqR066q3 is a good start on continued pretraining of LLMs to make them looped transformers, but my goal is to retrofit existing models like Qwen 3.6 35b into looped MoEs, since pretraining is expensive compute-wise.
I’m new on X and am looking for people who are interested in AI, neural nets, and building things.
If you’re verified and interested in those topics, let’s connect!