This is actually quite easy to implement at the harness level and the NER can be way more naive because of the typical nature of LLM dialogue (programming, long running tasks etc).
This is actually quite easy to implement at the harness level and the NER can be way more naive because of the typical nature of LLM dialogue (programming, long running tasks etc).
Can confirm. I did a clean-room implementation of this in Rust just by reading the paper. Will update once they release the reference implementation. https://github.com/ptaranat/zeromem
Care to elaborate?