tested llama 3.3 70b perplexity on code with different comment styles - inline vs block vs none
ran same typescript logic with three comment styles: 1. inline comments: `const x = 5; // user count` - perplexity 13.2 at 18k tokens 2. block comments above each function: perplexity 12.8 at 18k tokens 3. no comments: perplexity 14.1 at 18k tokens block comments seem to help model track state without burning as many tokens as inline. inline comments add ~1.2k tokens over block style for same logic. q4_k_m, llama.cpp b2701, temp 0.7
would love to see the numbers if you post them. dealing with inconsistent comment styles across our codebase and this would help settle the internal debate
this would be super helpful, we're dealing with comment style consistency issues on our team and hard numbers would settle the debate
post the numbers or this is useless
need the numbers