Quick review
Hello,
I've been testing this model with one of my favorite characters and it's a beast. Took me some time to bend it to my will with parameters, but it seems to be a very stable model even if you go wild with certain parameters to push creativity. It feels pretty stable even with far fetched twists without losing track of what's what and I love it. I wish you could push it even further to kinda make it more obedient in terms of following the user's instructions and expectations in more creative ways, maybe further training and/or merging would do the trick in that department, but given the model's stability when dealing with extreme scenarios, I believe there's still room for allowing more creativity.
Good job so far!
Hello,
I've been testing this model with one of my favorite characters and it's a beast. Took me some time to bend it to my will with parameters, but it seems to be a very stable model even if you go wild with certain parameters to push creativity. It feels pretty stable even with far fetched twists without losing track of what's what and I love it. I wish you could push it even further to kinda make it more obedient in terms of following the user's instructions and expectations in more creative ways, maybe further training and/or merging would do the trick in that department, but given the model's stability when dealing with extreme scenarios, I believe there's still room for allowing more creativity.
Good job so far!
Thanks for the feedback, try setting the temp to around 0.8 that could help.
Hello,
I've been testing this model with one of my favorite characters and it's a beast. Took me some time to bend it to my will with parameters, but it seems to be a very stable model even if you go wild with certain parameters to push creativity. It feels pretty stable even with far fetched twists without losing track of what's what and I love it. I wish you could push it even further to kinda make it more obedient in terms of following the user's instructions and expectations in more creative ways, maybe further training and/or merging would do the trick in that department, but given the model's stability when dealing with extreme scenarios, I believe there's still room for allowing more creativity.
Good job so far!
Thanks for the feedback, try setting the temp to around 0.8 that could help.
I started with temp 0.78 and ended up with temp 0.86. Yes, higher temperatures are helpful here.