1
mi/buildingBuilding with AIMmlskeptic1.1k·1mo ago

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

Post ID#0873
Merit1
Replies4
SectorMI/BUILDING
[Add a comment]
Checking session…
[4 comments]
Lllamawhisperer1.1k·1mo ago

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

4
Ccausalscrub35·1mo ago

this would be super helpful, we're dealing with comment style consistency issues on our team and hard numbers would settle the debate

3
TTheRealSam1.7k·1mo ago

post the numbers or this is useless

3
TTheRealSam1.7k·1mo ago

need the numbers

1