Others Formamorph [v1.1.3] [FieryLion]

[Nidkar]

You need the chat completion, not the regular completition endpoint

You must be registered to see the links

Thanks for the answer, but sadly that still doesn't fully work for Ooba. When i press start game, Ooba does generate something but when it's done Formamorph recieves an error and the text screen remains empty. "Failed to complete action, please try again" and "Failed to parse AI response, please try again". API token/model name shouldn't be required, right?

 

[fierylion]

Thanks for the answer, but sadly that still doesn't fully work for Ooba. When i press start game, Ooba does generate something but when it's done Formamorph recieves an error and the text screen remains empty. "Failed to complete action, please try again" and "Failed to parse AI response, please try again". API token/model name shouldn't be required, right?

That means your model wasn't able to produce the correct response format, the game expects JSON response for example{"game_text": "something something", "stats_updates": [{"Health":-5}]} If the AI writes JSON with syntax error such as missing a quotation mark or closing bracket then the game cannot parse the response. You need a model that understands coding somewhat, most of the instruct models can work. Models that are too heavily quantized may not work.

Also model name should be the exact model name you have loaded on Ooba

 

[Nidkar]

That means your model wasn't able to produce the correct response format, the game expects JSON response for example{"game_text": "something something", "stats_updates": [{"Health":-5}]} If the AI writes JSON with syntax error such as missing a quotation mark or closing bracket then the game cannot parse the response. You need a model that understands coding somewhat, most of the instruct models can work. Models that are too heavily quantized may not work.

Also model name should be the exact model name you have loaded on Ooba

I see, thank you for the answer. You'd think a 123B instruct model (which is 2.75 bpw, so maybe it's too quantised but i'm not sure it matters at this size) would do the formatting correctly first try, i'll do a little more research and see if i can tweak it to work.

 

[fierylion]

I see, thank you for the answer. You'd think a 123B instruct model (which is 2.75 bpw, so maybe it's too quantised but i'm not sure it matters at this size) would do the formatting correctly first try, i'll do a little more research and see if i can tweak it to work.

I mainly test on nemo instruct (gptq) and it works like a charm

 

[Nidkar]

I mainly test on nemo instruct (gptq) and it works like a charm

Unfortunately, i was unable to get it to work. Tried downloading the model you suggested, sadly doesn't work either.
If there was some way i could read the response from ooba in some form (maybe a debug output log or something, or a small internal console) that might help in tracking down the problem. And i've tried the verbose option in Ooba, but that is only for input. For now i'm going to give up. It's probably some specific setting on my PC. Best of luck to the other issues.

 

[fakeaccount7654321]

somethings wrong with the breasts on mine

 

[fierylion]

somethings wrong with the breasts on mine
View attachment 4545517

You got the breast size to go negative?

 

[fierylion]

Unfortunately, i was unable to get it to work. Tried downloading the model you suggested, sadly doesn't work either.
If there was some way i could read the response from ooba in some form (maybe a debug output log or something, or a small internal console) that might help in tracking down the problem. And i've tried the verbose option in Ooba, but that is only for input. For now i'm going to give up. It's probably some specific setting on my PC. Best of luck to the other issues.

I forgot to ask, but did you try chatting with the model? Some models the client (Ooba) may not be able to guess the chat template and you have to provide it. On my end I have to explicitly write down the chat template for the LLM engine (Aphrodite-Engine)

You must be registered to see the links

Try using LMStudio as the API server

 

[krubinka]

The description sounds like a great way to implement AI. I've been waiting for something like this type of AI use, the question is, how well is it implemented. Guess I'll soon find out.

You can use AI Roguelite, attach a NSFW llm (7-8b q4 k_s/m) and instruct model on top of a sd 1.5. That can run with just 8 gigs of vram. Having one model running somewhere else helps a lot. Might also mess around with the local files (all of them are txt) to get what you want.

 

[bluerubber]

You can use AI Roguelite, attach a NSFW llm (7-8b q4 k_s/m) and instruct model on top of a sd 1.5. That can run with just 8 gigs of vram. Having one model running somewhere else helps a lot. Might also mess around with the local files (all of them are txt) to get what you want.

Roguelite looks a little too much RPG focused for my preference, but it is nice to see progress in implementing AI into gaming in general. I am looking forward to openworld story based content with intuitive render based images to CYOA smut stories with less emphasis on RPG. I'm sure it'll be around at some point in the near future.

 

[jackops]

i just played the game for bit or should i say, tried it. 2-3 request per min with Mistral Ai and the immersion is broken...
i'm not fan of base-text game but the concept look interesting. the problem with IA is that there is no really "free" one to use, so even if the game is free you need to pay, also it still in WIP ( the IA) if may say ^^.
you will need a lot of patience and hard work to make the game "playable" for every one. So for me the biggest problem is the reactivity with response. i didn't see much of the game to juge it but it looks promising.
Good luck !

 

[fierylion]

i just played the game for bit or should i say, tried it. 2-3 request per min with Mistral Ai and the immersion is broken...
i'm not fan of base-text game but the concept look interesting. the problem with IA is that there is no really "free" one to use, so even if the game is free you need to pay, also it still in WIP ( the IA) if may say ^^.
you will need a lot of patience and hard work to make the game "playable" for every one. So for me the biggest problem is the reactivity with response. i didn't see much of the game to juge it but it looks promising.
Good luck !

Hi, thanks for trying the game! I couldn't quite understand sorry, are you saying the long wait time breaks your immersion or is it something else?

 

[jackops]

yes i make 2 choice then i need to wait 1 or 2 min, some time more.

 

[Yergavitz]

I've been enjoying the game so far, though I feel as if it would be nice to allow for the possibility to gain Traits in the Trait menu next to your stats through gameplay -- Unless it is possible and I just never happened to succeed, of course. Making Traits in the Trait section of the world editor that are specifically designated to be obtainable through gameplay (Optionally excluded from the starting Traits you can pick from) might be a nice way to spice things up.

In a similar fashion, if possible, allowing for additional Stats in the world editor that are that are not visible in your starting Stat menu but instead can also be gained through gameplay or through Trait acquisition might help to increase replayability.

 

[yunasenna]

Sadly, the game isn't working and I have done every step and I still get errors. I've been trying to fiddle with other possible AIs for the game but nothing. I really wanted to play it too. Good luck on the game.

 

[Nidkar]

I forgot to ask, but did you try chatting with the model? Some models the client (Ooba) may not be able to guess the chat template and you have to provide it. On my end I have to explicitly write down the chat template for the LLM engine (Aphrodite-Engine)

You must be registered to see the links

Try using LMStudio as the API server

Continued failures in Ooba, the chat in Ooba itself works. I even manually set the instruction template to mistral (it was a mistral model after all) but sadly same failed to parse error.
BUT
LMStudio is working!
in fact, this looks a lot nicer than Ooba. I might even jump ship to this program instead.
Thank you for the help, i'll be staying with LMStudio for now, still a shame Ooba refuses to work.

 

[fierylion]

Sadly, the game isn't working and I have done every step and I still get errors. I've been trying to fiddle with other possible AIs for the game but nothing. I really wanted to play it too. Good luck on the game.

You can just use default AI settings which will use the AI on my potato PC

 

[fierylion]

I've been enjoying the game so far, though I feel as if it would be nice to allow for the possibility to gain Traits in the Trait menu next to your stats through gameplay -- Unless it is possible and I just never happened to succeed, of course. Making Traits in the Trait section of the world editor that are specifically designated to be obtainable through gameplay (Optionally excluded from the starting Traits you can pick from) might be a nice way to spice things up.

In a similar fashion, if possible, allowing for additional Stats in the world editor that are that are not visible in your starting Stat menu but instead can also be gained through gameplay or through Trait acquisition might help to increase replayability.

I'm looking into giving the AI more power in-game. But I have a major limitation where smaller AI models just can't do it, giving it too many different game functions will make it confused and misbehave. My plan is to make these mechanics such as gaining, loosing traits in-game to be toggable modules, so players with larger beefier AI models can toggle them on. The default AI can barely run the game as is lol

 

[Nidkar]

Continued failures in Ooba, the chat in Ooba itself works. I even manually set the instruction template to mistral (it was a mistral model after all) but sadly same failed to parse error.
BUT
LMStudio is working!
in fact, this looks a lot nicer than Ooba. I might even jump ship to this program instead.
Thank you for the help, i'll be staying with LMStudio for now, still a shame Ooba refuses to work.

One small thing to add, using LMStudio i was able to run a 123B model, but that didn't want to work the first time presumably because smarter models want to put json formats in code blocks (```)
Changing the prompt to basically say "hey don't do that" managed to fix it and now i can use it. Still, didn't fix it for Ooba, i tried anyway. Must be the instruction template messing with the output. Eh, it's working now.

 

[August_Brinkamann]

what is save direction ?