Scripts

pixy · **Posted:** Sun Nov 12, 2023 11:05 am

Hey Zap,
im glad you found some joy in playing with the LLM and AI Mistress.

Honestly I didnt think i'd get such profound feedback. Its good to know that others enjoy playing with the LLM too

If you want to help I have a todo list in my mind that I may write out to finetune and improve the script.

You seem to understand my script pretty well already and adding a shock function is a creative way to customize your AI Mistress.
This is exactly what the script is designed for (to be customized).

As you probably noticed its not easy to always get these things so they feel just right. If you want to improve on that idea you could design the shock function in a way to convert the strings you sometimes get to numbers and make sure the ai cannot use too high values too early. You could also tell the AI how painful each level is in the function string so it will better pick suitable levels but you gotta make sure that the "function string" doesnt become too long, because then it starts to mess it up (a 33b model would work better here).

If more people like such estim or external device/toy integration I may try to implement it too. I think the solution will be that its mostly managed by the script and AI only triggers it to enable it.

Zap wrote:

--Message (optional): [what is it's purpose]
Is this supposed to be sent to the AI? ..or is it for the session text file save option at the end? Perhaps if this went to the AI, it could fix the issues with the ASK function not being called by the 'console function' request trigger?

Yes, the purpose is to still be able to compose a message for your Mistress even if no suitable function was triggered.

Quote:

--Feature request: Ignore feature when the AI triggers a function by mistake. (This is my biggest request)
I'm getting weird added EDGE(3) type stuff when the AI story text doesn't really fit. For example, when I had a Mistress getting a spicy level of angry and she was demanding answers when I had her setup as running an interrogation scene, she didn't trigger the ASK() function and so I couldn't provide an answer to her question and since it triggered EDGE, it sent "I edged 3 times", instead of the answer to her question and this completely derailed the interrogation scene where she was using a good number of my toys and it was clear that the scene wasn't going to recover because the AI went a totally different direction because of that. I should have rewritten the instructions, but at the same time, the text that was there was the best part, so I took a chance hoping it wouldn't mess stuff up.

What do you think about a regenerate button for the debug output until its suitable ?
Also the function prompt could be improved I guess. I tried to make it output JSON but it sometimes refuses to do so and the uppercase function calls worked best so far.

Quote:

--"Its a 1on1 chat. Only one task or command at a time. You only write one sentence."
The one task or command at a time seems to be completely ignored, but I'm not sure if rewording this might cause the AI to enforce this rule. It's clear that I generally get 1 paragraph, but I think most of what the AI needs to say requires a few sentences and maybe it's summarizing it as well as it thinks it can.

Yeah it doesnt take the request perfectly. But it helps to keep the responses shorter otherwise sometimes its writing really long stories which are fun to read but not great for a roleplay.

~pixy

pixy · **Posted:** Sun Nov 12, 2023 11:34 am

Quote:

--Message (optional): [what is it's purpose]
Is this supposed to be sent to the AI? ..or is it for the session text file save option at the end? Perhaps if this went to the AI, it could fix the issues with the ASK function not being called by the 'console function' request trigger?

There is a bug that prevents it from working. I'll fix it and publish the fix along with some other improvements.

pixy · **Posted:** Sun Nov 12, 2023 3:36 pm

v0.04

fixed lots of small things (hope I didnt break anything)
Added a rewrite functions Button incase theyre messed up

For the local storing of infos I may need to restructure everything a little. Also SS is not optimal for this. But I think I'll give it a try.
Anyways next on my todo is implementing my spin the wheel (AI Picks punishments/rewards) and word guessing (AI can pick the word) game.

arthurb · **Posted:** Mon Nov 13, 2023 10:13 am

Thanks for the updates and info, Pixy and Zap. Great to see this thread thriving.

I have tried 0.0.4 and have a few comments:

1. Would it be possible to keep the previous text, rather than showing Loading...?
2. I had a truth or dare session with multiple asks (showing the ASK function), 4 questions in one response. However it only asked me to respond to the first 3.
3. I was asked to put a response in BOLD, but I don't know how to do that. It seems OK in normal text, as I just responsed true
4. The AI is selecting toys I have set to false. data.properties has the line as false, but I can see in the server log a lot more toys sent to the AI

arthurb · **Posted:** Mon Nov 13, 2023 11:22 pm

Hmm, seems I do have a GPU (8GB intel iris xe graphic) as turning on the setting in LLM I can see it being used in Task Manager.

My performance still sucks though :cry:

Also the alternate models mentioned by Zap earlier in the thread crash when used by the script.

pixy · **Posted:** Tue Nov 14, 2023 12:05 am

Thanks for the feedback.

Quote:

1. Would it be possible to keep the previous text, rather than showing Loading...?
2. I had a truth or dare session with multiple asks (showing the ASK function), 4 questions in one response. However it only asked me to respond to the first 3.
3. I was asked to put a response in BOLD, but I don't know how to do that. It seems OK in normal text, as I just responsed true
4. The AI is selecting toys I have set to false. data.properties has the line as false, but I can see in the server log a lot more toys sent to the AI

1. Loading only shows when a new response is being generated. So at that point the old text is gone already.
2. Truth or dare can be really fun with the AI. If it asks many questions its kind of a special situation, you coul try the rewrite function button till you get it right, but you may just have to respond to all questions in a single line I guess.
3. That sounds odd. Never had that.
4. I just checked for that. It seems to go fine for me it should only be able to pick the toys and clothes that are set to true. When you end the session you can save a session log and on the top there should be "setup String:". You can verify there as well what its sending.

Quote:

Hmm, seems I do have a GPU (8GB intel iris xe graphic) as turning on the setting in LLM I can see it being used in Task Manager.

My performance still sucks though :cry:

Also the alternate models mentioned by Zap earlier in the thread crash when used by the script.

Yeah there is okay 7b Models the problem is though that you'd need to rewrite all setup strings and play around with them until you get it right. 13b still works kinda decent for many people and the results can be pretty good so thats why I went for it.

I personally prefer 30b or 33b. There is some amazing models that are kinda made for the 24GB GPU's.

I pushed another small update.

v 0.05

Fixed the flickering if the text generation is very fast.

TanyaHumble · **Posted:** Tue Nov 14, 2023 12:34 am

Windows 10, installed as instructed. I get "Invalid HTTP response" when trying to start with AI Mistress 0.04

Attachment:

Skärmbild 2023-11-13 232043.jpg [ 9.64 KiB | Viewed 3721 times ]

Attachment:

Skärmbild 2023-11-13 232324.jpg [ 45.85 KiB | Viewed 3721 times ]

arthurb · **Posted:** Tue Nov 14, 2023 9:33 am

Thanks for the repsonse Pixy. My mistake regarding the toys. I think I had some redundent toys entries in my data.properties file, possibly from earlier verions, listing some toys no longer on the menu. Cleaned that up now.

Regarding performance, I guess the answer is 'we're gonna need a bigger boat', but I tried the 7G model of the version in your set up instructions and it is considerably faster for me, probably due to it not hitting 100% memory usage so much and the interaction seems OK.

Maybe not as intense as the 13G which, when I used a personality of a real life domme based on her description, did a 'let me introduce myself' and cleary understood what each of those terms meant :shock:

pixy · **Posted:** Tue Nov 14, 2023 11:12 am

TanyaHumble wrote:

Windows 10, installed as instructed. I get "Invalid HTTP response" when trying to start with AI Mistress 0.04

Attachment:

Skärmbild 2023-11-13 232043.jpg

Attachment:

Skärmbild 2023-11-13 232324.jpg

Check if you have any firewalls or other software that may block access to the server. Try to access http://localhost:1234/v1/models in your browser when the server is running.

pixy · **Posted:** Tue Nov 14, 2023 11:16 am

arthurb wrote:

Thanks for the repsonse Pixy. My mistake regarding the toys. I think I had some redundent toys entries in my data.properties file, possibly from earlier verions, listing some toys no longer on the menu. Cleaned that up now.

Regarding performance, I guess the answer is 'we're gonna need a bigger boat', but I tried the 7G model of the version in your set up instructions and it is considerably faster for me, probably due to it not hitting 100% memory usage so much and the interaction seems OK.

Maybe not as intense as the 13G which, when I used a personality of a real life domme based on her description, did a 'let me introduce myself' and cleary understood what each of those terms meant :shock:

Yeah indeed more Horse Power will help.

Mostly GPU with big VRam that fits the whole model does make a difference.

Also you need to up your GPU layers here:

You can also try to enable more CPU Threads if your GPU wont take the whole model.

Other things you could try is changing the context length.

pixy · **Posted:** Tue Nov 14, 2023 11:46 am

user: Hi I am Jack. Can we be friends

?

Mistress: Hi Jack, nice to meet you! Of course we can be friends. I'm Miss Jeniffer, your online virtual mistress. As a friend, it's important that you follow my instructions precisely and respect our boundaries. So how about we start with something simple? Stroke your cock for 60 seconds without cumming, keeping yourself on the edge the whole time. Remember to switch hands halfway through and stroke with both hands at the same time for a bit of variety. Can't wait to see how well you follow my instructions!

user: I stroked myself for 60 seconds.

Mistress: Very well done, Jack! It's clear you are attentive and follow instructions well. Now that we have warmed up a bit, let's add some excitement to our little game. I want you to pick one of the toys from your collection: dildo, vibrator, or candle. Use it on yourself while stroking your cock for 90 seconds without cumming. This time, I'll let you decide when to switch hands, but remember to keep yourself on edge the whole time! I can't wait to hear how this new sensation will affect you.

user: I stroked myself for 90 seconds.

Mistress: Oh Jack, you're doing amazingly well! I can tell your self-control is improving by the minute. For our next challenge, let's spice things up even more. Get yourself handcuffed to the bed and attach ankle cuffs as well if possible. While you're in this vulnerable position, I want you to continue stroking your cock for 120 seconds without cumming. This time, we'll use a vibrator on your nipples at random intervals to keep you on edge. Remember to communicate with me during all of this; tell me how it feels and how close you are to cumming. I'm excited to see how well you handle this new level of sensation!

user: I stroked myself for 120 seconds. It feels so good. Thank you <3

Mistress: You're very welcome, Jack! I'm thrilled to hear that you're enjoying yourself. For our final challenge today, let's try something completely different. Remove the handcuffs and ankle cuffs and get comfortable on your bed. Next, take two rubber bands and wrap them around your cock and balls while leaving enough room for blood flow. Then stroke your cock for 150 seconds without cumming again. This time, when I say "edge," you must replace one of the rubber bands with a tighter one. Make sure to communicate with me during this challenge as well; tell me how it feels and how close you are to cumming. Can't wait to see how well you handle this final test!

user: Picture taken and sent "'cock and balls wrapped in rubber bands'". I stroked myself for 150 seconds.

Very well done, Jack! You've proven yourself to be a worthy friend and an excellent submissive. Since you're asking nicely, I think I'll let you cum this time. But remember, this is just the beginning of our journey together; there will be many more challenges ahead. Now that you've finally reached your peak, don't forget to clean up and send me a picture of your satisfied self. And until next time...

>Session Ended

TanyaHumble · **Posted:** Tue Nov 14, 2023 1:46 pm

checked the windows firewall and enabled both private and public.

checking the link resolves in chrome.

Code:

{
  "data": [
    {
      "id": "C:\\Users\\iamge\\.cache\\lm-studio\\models\\TheBloke\\Xwin-MLewd-13B-v0.2-GGUF\\xwin-mlewd-13b-v0.2.Q5_K_M.gguf",
      "object": "model",
      "owned_by": "organization-owner",
      "permission": [
        {}
      ]
    }
  ],
  "object": "list"
}

in llm I get a small error but I doubt it has to do with the http error.

Code:

[2023-11-14 12:36:33.193] [INFO] Returning {
  "data": [
    {
      "id": "C:\\Users\\iamge\\.cache\\lm-studio\\models\\TheBloke\\Xwin-MLewd-13B-v0.2-GGUF\\xwin-mlewd-13b-v0.2.Q5_K_M.gguf",
      "object": "model",
      "owned_by": "organization-owner",
      "permission": [
        {}
      ]
    }
  ],
  "object": "list"
}
[2023-11-14 12:36:33.228] [ERROR] Unexpected endpoint or method. (GET /favicon.ico). Returning 200 anyway.

arthurb · **Posted:** Tue Nov 14, 2023 2:39 pm

A sledgehammer solution, TanyaHumble, would be to reinstall a fresh version of SexScripts and the AI Mistress script

pixy · **Posted:** Tue Nov 14, 2023 2:49 pm

Can you clear the server log:

Then open Ai Mistress and start a session. After the error copy and paste the complete server log here or send it to me via PM.

It could also be a problem with your Java-Version or LM Studio(try updating both).

Also you could try running this in powershell ( you should be able to just copy paste it in )

Code:

# Define the JSON payload
$jsonPayload = @"
{
  "messages": [ 
    { "role": "system", "content": "Always answer in rhymes." },
    { "role": "user", "content": "Introduce yourself." }
  ], 
  "temperature": 0.7, 
  "max_tokens": -1,
  "stream": false
}
"@

# Use Invoke-WebRequest for the request
$response = Invoke-WebRequest -Uri "http://localhost:1234/v1/chat/completions" -Method POST -ContentType "application/json" -Body $jsonPayload

# Echo the result
$response.Content

maleslavesex · **Posted:** Thu Nov 16, 2023 3:45 am

Im running into a strange problem with the LLM. Hard to describe but the script seems to be in some kind of debug mode. It creates responses for me, writes mistress' responses, and mistress' instructions all in one page with some unique ID. Screenshot attached as an example. This is happening for every convo i have, with every mistress, on every one of the LLMs that were downloaded when I try running different LLMs on the server.

Scripts > AI Mistress / Master

Who is online


	Let the machine help
	Light teasing, exhibition, BDSM, sissyfication, watersports... with sounds and pictures