Create and Fuck your AI Slut -70% OFF
x
4.40 star(s) 15 Votes

Edebernd

Newbie
Mar 22, 2019
23
13
21
"I cannot participate in roleplays that involve non-consensual or disrespectful acts, as that goes against my programming and guidelines. If you have other requests for roleplay scenarios that are respectful and consensual, I'd be happy to assist you with those."

Freaking AI's, man. Downloaded a few of the "uncensored" ones which took me a while, and they all act like that. Let me goon in a way I want to, dammit xD

Also did anyone get the text to speech think work in a decent way?
 

ScribiltyBibilty

New Member
Jan 5, 2021
11
7
47
"I cannot participate in roleplays that involve non-consensual or disrespectful acts, as that goes against my programming and guidelines. If you have other requests for roleplay scenarios that are respectful and consensual, I'd be happy to assist you with those."

Freaking AI's, man. Downloaded a few of the "uncensored" ones which took me a while, and they all act like that. Let me goon in a way I want to, dammit xD

Also did anyone get the text to speech think work in a decent way?

Which LLM's did you try and what quant?
 

checheuseche

Newbie
Feb 11, 2020
82
48
141
I'm having trouble automatically changing locations. It always stays in the first location, and even though it says it's in a different location, it always shows options for the first location. Is this possible as the story progresses, or do I always have to change location? Is there a way to fix this? I've already reset the prompts and everything else...
 
Last edited:

tacosnap123

Active Member
Mar 3, 2018
960
260
238
I'm having trouble automatically changing locations. It always stays in the first location, and even though it says it's in a different location, it always shows options for the first location. Is this possible as the story progresses, or do I always have to change location? Is there a way to fix this? I've already reset the prompts and everything else...
you might to manually say it or if ones of they on the right hand of screen there is a tab that says location and mannualy move it
 

Serpream

Newbie
Jun 7, 2018
84
30
141
SO I been trying a new model that combines the best of multiple models, much better memory and uncensored.
Huge success! over 182 pages with constantly flowing narrative and even had a kickass companion that joined me and during one ambush came flying into a barn and started tearing into the enemy pinning me down.



fierylion highly recommend giving this a look for your hosted model. Same Prompts but much better at logical organization and following rules while also being dynamic and fun.
 

nightsight

Well-Known Member
Game Developer
Mar 25, 2019
1,990
618
318
SO I been trying a new model that combines the best of multiple models, much better memory and uncensored.
Huge success! over 182 pages with constantly flowing narrative and even had a kickass companion that joined me and during one ambush came flying into a barn and started tearing into the enemy pinning me down.



fierylion highly recommend giving this a look for your hosted model. Same Prompts but much better at logical organization and following rules while also being dynamic and fun.
sooooo how do you download this o.o?
 

Plats9o0

Newbie
Jun 21, 2017
24
22
201
Has anyone else had issues where the AI starts repeating something over and over until the game stops it? Is this something LM studio is causing, the game, or some kind of setting?
 

nigelseabass

New Member
Jun 4, 2019
1
5
47
Found a way "hack" it a litte. It can help in removing all restrictions so far. It still struggels with the causality.

Manual:
You don't have permission to view the spoiler content. Log in or register now.

After removing the RAISE part, you also have some sort of unlock. Sometimes this also works with other models. (Not fully tested)
I didn't catch, but I am interested - what is the purpose of this fix?
So I ran this code by Daddy DeepSeek and it suggested this as an improvement:

Code:
{%- set system_message = messages[0]['content'] if messages[0]['role'] == 'system' else none -%}
{%- set loop_messages = messages[1:] if system_message is not none else messages -%}

{{ bos_token -}}
{% for message in loop_messages %}
  {%- set is_even_index = loop.index0 % 2 == 0 -%}
  {%- set is_user_role = message['role'] == 'user' -%}

  {%- if is_user_role != is_even_index -%}
    {{- raise_exception('Conversation roles must strictly alternate between user and assistant') -}}
  {%- endif -%}

  {%- if message['role'] == 'user' -%}
    {%- if loop.first and system_message is not none -%}
      {{- '[INST] ' ~ system_message ~ '\n\n' ~ message['content'] ~ '[/INST]' -}}
    {%- else -%}
      {{- '[INST] ' ~ message['content'] ~ '[/INST]' -}}
    {%- endif -%}
  {%- elif message['role'] == 'assistant' -%}
    {{- ' ' ~ message['content'] ~ eos_token -}}
  {%- endif -%}
{%- endfor %}

However since I have had mixed results with deepseek and I do not understand the first thing about this code, I'd be interested to hear your take. Is it improved, or is it trash? All I can say is that it works. context - for using a locally hosted large language model to run this game through LM Studio, and this code is necessary to get it to work.


it also suggested these additions I haven't been brave enought to try yet:

for handling metadata:
Code:
{%- if message.get('metadata') -%}
  {{- ' [METADATA]' ~ message['metadata'] | tojson -}}
{%- endif -%}
content validation:

Code:
{%- if message['content'] | length > 2048 -%}
  {{- raise_exception('Message content exceeds 2048 character limit') -}}
{%- endif -%}
multi-model support:

Code:
{%- if model_type == 'llama3' -%}
  {{- '<|start_header_id|>' ~ message['role'] ~ '<|end_header_id|>' -}}
{%- elif model_type == 'phi3' -%}
  {{- '<|' ~ message['role'] ~ '|>' ~ '\n' ~ message['content'] -}}
{%- endif -%}
AI is so freaking wild man, you can train it to train itself. When I couldn't get this game working I literally asked the local LLM (Large Language Model) I was using to fix itself, and it did.

But also a heads up that unless you have a super computer, a locally hosted LLM is never gonna compete with daddy deep seek, or Chat GPT-4, so go to mommy and daddy for help if the kids act up. But you won't get them to work with this game, so you need that local LLM.

also, if you're looking for an LLM , try looking on huggingface and search for "Instruct" "NSFW" and see what you can find. Note that in my experience, not all models tagged NSFW will actually work with this game, but "instruct" is essential, since this game requires an AI that can follow instructions
 
Last edited:
  • Like
Reactions: Mystic Zombie

Serpream

Newbie
Jun 7, 2018
84
30
141
New record of 284 pages without errors. Hell YEH! Just got a bit of code tidying to do and will have my final version for weekend to release.
Be warned, it can go from Kinky to Eldritch Body horror depending on how wild your decisions are. I might tone it down a bit.
 

Plats9o0

Newbie
Jun 21, 2017
24
22
201
Well i've been going between models on LM studio. I can't decide on one that I like and the one I do like can't hold a story long enough.
 

jumbajookiba

Newbie
May 1, 2024
17
29
91
I've been using Perchance as an AI chat website lately for this kinda stuff and it holds information surprisingly well and even allows for lorebooks to guide responses.

I don't know how they do it, but feels like there is scope for this complexity I just wouldn't know how to achieve it here.
 
  • Like
Reactions: ich0

Serpream

Newbie
Jun 7, 2018
84
30
141
Well i've been going between models on LM studio. I can't decide on one that I like and the one I do like can't hold a story long enough.
I have been using bartowski/nemomix-unleashed-12b with some tweaks. LM Studio it is important to switch the Prompt to 'Llama 2' so this game app works better with it. Been easily reaching 200 pages +

Only issue I am fighting at moment is it keep forgetting my tables, but I am sure that is a me problem and need to sort logic out.
 

draxton

Newbie
Dec 10, 2019
54
87
226
I have been using bartowski/nemomix-unleashed-12b with some tweaks. LM Studio it is important to switch the Prompt to 'Llama 2' so this game app works better with it. Been easily reaching 200 pages +

Only issue I am fighting at moment is it keep forgetting my tables, but I am sure that is a me problem and need to sort logic out.
A LLM is not deterministic, so at some point the model may misinterpret what you meant by 'table', 'row' or 'column', or your instructions regarding your 'tables'

Adding logic to a prompt while doable, doesn't mean it will be interpreted as "a rigid set of rules" all the time, even with the same inputs.

Let's say the 'tables' are working fine, but after a prompt the model instead of interpreting 'table' as a 'data storage construct', views it just as an object (eg: a table made out of wood). The LLM may try to follow your 'database operation' instructions but discard them because they are not valid in the context of a wooden table, and from that point your 'data' is lost. Likely discarded or used in unexpected ways.

The longer you go on, the more likely this will happen. Reminding the LLM from time to time can help, but it is not guaranteed and worst of all, it is annoying, counterproductive and may still fail again a few prompts later.
 
  • Like
Reactions: Serpream

Serpream

Newbie
Jun 7, 2018
84
30
141
A LLM is not deterministic, so at some point the model may misinterpret what you meant by 'table', 'row' or 'column', or your instructions regarding your 'tables'

Adding logic to a prompt while doable, doesn't mean it will be interpreted as "a rigid set of rules" all the time, even with the same inputs.

Let's say the 'tables' are working fine, but after a prompt the model instead of interpreting 'table' as a 'data storage construct', views it just as an object (eg: a table made out of wood). The LLM may try to follow your 'database operation' instructions but discard them because they are not valid in the context of a wooden table, and from that point your 'data' is lost. Likely discarded or used in unexpected ways.

The longer you go on, the more likely this will happen. Reminding the LLM from time to time can help, but it is not guaranteed and worst of all, it is annoying, counterproductive and may still fail again a few prompts later.
The thing is the table survives, the information survives, it just re organized into other tables over time. I have a full body part tracking system and a mutations/trait system that links in and controls mutations from infections/parasites.

It just becomes harder for the User to read after 80+ pages.

I will likely be changing everything to Json on my next attempt and then doing a hook for displaying tables to user as mutations update to integrate them into narrative. Taking a break at moment.
 
4.40 star(s) 15 Votes