If you need really good compression , I’ve used on ChatGPT “SentenceSqueezer”, it usually compresses in neighborhood 60-85% of instruction sets and it’s so simple, guaranteed not to lose context because it makes an acronym out of each sentence or line , tell it to leave periods in there place or whatever you desire, gives pre and post count with a mapping table, and it may do meta compression using one symbol. The other one is “Tokenizer GPT Instruction Compressor” also in the same platform but uses single placeholders for words that repeat more than 3 times I. BElieve, outputs same stats on compression mapping table etc, may be worth a try, I’m new so I may be way off if this will work for you, hopefully so. But if they both have worked for me in not having to ever worry about how technical I need my instructions to be