[+] New quark-e-4.0-100k-mini ML model for KillAura detection that outperforms the previous model by 8.5x
[#] Fixed issues with UseEntity
[#] Upgraded Millennium 5 ML library
[#] Fixed issue with UseEntity listener
[+] /mx ml 7 th (decisionThreshold) command
[+] Contributors in plugin.yml
[+] Model upgrade from quark-e-1.0-56k-public to quark-e-1.5-56k-public
[+] Avg Loss / Accuracy while training
[#] Many fixes in Millennium 5 library
[+] Millennium 5 RNN Library:
- ML upgrade from 001-premium-complex-18k to
quark-e-1.0-56k-public neural model
- /mx ml <index> <param> <value> (Params: lr, dr, rdr, wd, gc, ls, im, pm) for RNN ML configuration
- /mx train <modelIndex> <epochs> for easy train
- /mx dataset <legit|cheat|off> <player|all> for samples collect
[+] New Baritone checks
- Generic
- Long-term
[+] Sprint Heuristic check
[#] Fixes
- Fixed some perfomance issues
- FIxed some teleport issues
- New versions support
[#] Fixed goofy issue with Aim (Invalid)
[+] ML Upgrade (001-free-complex-18k -> 001-premium-complex-18k)
Spartan Anticheat users were the first to try out this update.
[#] Fixed minor issue with async scheduler
[+] 001-free-complex-18k ML for Aim-related hacks, based on super-compact Millennium v4.5 model (still in development, can be turned off in checks.yml). Runs locally, no additional files required, just plug & play.
[+] New "prevention" option (config.yml)
Some checks also apply measures,
such as a ban on dealing damage to others for a few seconds.
AGGRESSIVE (3) - rough hit blocking + teleport
CAREFUL (2) - rough hit blocking (full hit block)
LENIENT (1) - a detected cheater can still deal hits, but their outgoing damage will be very small
SILENT (0) - all measures fully disabled
[#] Hex color support
[#] Improved stability of Smooth, Constant and Linear checks
[#] Improved stability of autoclicker checks
[#] Improved efficiency of Factor check
