Like Ra's Naughty Forum

Full Version: Writing hypno-scripts with AI
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13
(03 Nov 2023, 13:29 )Like Ra Wrote: [ -> ]
(03 Nov 2023, 09:37 )shinybambi Wrote: [ -> ]... but this one keeps endlessly repeating itself most of the time.
Can you increase the temperature? Or whatever it's called in Llama? To decrease the inferring predictability.

Yes, but it doesn't fix the response loop, something went wrong with the model during training. I'll try different settings and base models.
(03 Nov 2023, 13:44 )shinybambi Wrote: [ -> ]something went wrong with the model during training.
Ah, OK. The learning rate was too high?
(03 Nov 2023, 13:52 )Like Ra Wrote: [ -> ]
(03 Nov 2023, 13:44 )shinybambi Wrote: [ -> ]something went wrong with the model during training.
Ah, OK. The learning rate was too high?

Or loss was too high. Or it was fast tokenizer issue. I have no idea, but that last script was probably the best hypno sample it managed to produce.

Lowered the learning rate and changed base model, we'll know if it'll change anything in an hour or so...
Finished training LoRA, new Q4_K_M quantized test model https://drive.google.com/file/d/1SyQ5U5d...sp=sharing
Used Xwin-MLewd-13B-V0.2 as base.
Context size is 4096, Alpaca instruct prompt format.

Training config:
Show Content
(03 Nov 2023, 17:03 )shinybambi Wrote: [ -> ]Context size is 4096
WOW! 👍
Tried it with slightly modified prompt from another AI thread. 

Here's the result:
Show Content

Show Content

Looks like it's heavily influenced by EMG scripts. I'll try to refine and extend dataset with some lewd stories and train on different base model later...
(04 Nov 2023, 11:18 )shinybambi Wrote: [ -> ]Yes that is what is going to happen now, and you cannot stop it or avoid it in any way shape or form. It simply is and there is nothing you can do about it. Nothing at all.

(04 Nov 2023, 11:18 )shinybambi Wrote: [ -> ]Looks like it's heavily influenced by EMG scripts.
Ha! Yes, indeed! I even hear his voice!!!

Still a bit repetitive, but so is EMG!

(04 Nov 2023, 11:18 )shinybambi Wrote: [ -> ]Your hands are covered by soft black gloves with a few small rings at the end of each finger to indicate how slutty and obedient Bambi's uniform truly is.
What a weird AI hallucination!

The text looks much "looser", definitely with a higher temperature. If edited manually, or put through another round of AI with a lower temp to fix grammar and a bit of a sense, those scripts can be used to screw up the imagination. Great results, I should say!
Some ideas. If you have a draft of a hypno script. It can be enhanced, right?

Also, would it be possible for this model to generate side and/or background voices along the scenario? With instructions, like: use the script below to generate additional subliminal text for left and right audio channels to go along with the main text, using the format {left channel: text for the left channel}, {right channel: text for the right channel}
(04 Nov 2023, 23:05 )Like Ra Wrote: [ -> ]Some ideas. If you have a draft of a hypno script. It can be enhanced, right?
Also, would it be possible for this model to generate side and/or background voices along the scenario? With instructions, like: use the script below to generate additional subliminal text for left and right audio channels to go along with the main text, using the format {left channel: text for the left channel}, {right channel: text for the right channel}
I don't think that's possible with current model. I just showed it vaguely what hypnosis script is, as a whole. It doesn't understand where induction or awakener are, so i think asking it for something complex as background channels would be too much Lol


Show Content
Looks like the original script got ignored completely, and the NN was caught in its own hallucination 😁
Pages: 1 2 3 4 5 6 7 8 9 10 11 12 13