paq8px v217
Version by @GotthardtZ
https://encode.su/threads/342-paq8px?p=89137&viewfull=1#post89137
- New Mixer:
- It has now 3 layers instead of the traditional 2
- Optimized mixer parameters
- Removed the stale and partially broken adaptive learning rate (-a) option
- JPEG model enhancements and fixes
- 1-bit image model enhancements
- Fixes in SSE stage
- Improved/simplified LSTM model contexts
- Reduced memory size of compression level 10 and above. Now level 12 should not cause swapping with 32 GB of RAM