Previous version
v3 : same old training parameters version
v1 and v2 were abandoned because the outputs are poor, they were designed as lite version,
New version
v4 : version with new training parameters
v5 : most optimized version with new training parameters
v4 and v5 also perform well with simple prompts (last 2 posts for example)
without trigger words the outputs result slightly different, maybe worthwhile to try.
Since new parameters perform better and more normal, next I will train new version for each model before. Hopefully that will be much faster.