fig (aka:[phil]) @bad-example.com · Apr 11 changing zstd compression level (without dictionary training) makes a surprisingly small difference. very interesting how 1/3 and 6/9 start bunched, then fan out evenly as data increases. i guess they need more to compress for certain benefits to show? (y-axis doesn't go to zero warning)
2 likes 1 replies
?
Replies
dictionary training is sometimes significant, and very tricky rocks has an amount-of-training-data config, you *need* to train enough if using that setting, or you'll get zero benefit (or sometimes negative) leaving it unset (only setting dict size) works consistently but less well