Others Formamorph [v1.1.11] [FieryLion]

[Serpream]

I was so so so dam close to having a consistent and detailed flowing transformation narrative with full working tables and everything.
Now I am stuck in loops again after making a few edits, but worse LM Studio also updated, and I am getting endless loops after 8 pages now.
No idea if something I did or the LM Studios update >.>

 

[Breedking]

I was so so so dam close to having a consistent and detailed flowing transformation narrative with full working tables and everything.
Now I am stuck in loops again after making a few edits, but worse LM Studio also updated, and I am getting endless loops after 8 pages now.
No idea if something I did or the LM Studios update >.>

I think it's LM studio, I'm getting the same issue. Rolling back and slightly changing the prompt seems to work 50/50.

 

[X_Tigershark_X]

I was so so so dam close to having a consistent and detailed flowing transformation narrative with full working tables and everything.
Now I am stuck in loops again after making a few edits, but worse LM Studio also updated, and I am getting endless loops after 8 pages now.
No idea if something I did or the LM Studios update >.>

The same issue, but I noticed one pattern. If I do not write anything in the "Notes" field, then everything works fine and events are generated further, but early details begin to be forgotten until events from the past was completely disappears from "story". If I write notes in the "Notes" field - then after the 7th page the program starts to give out text from the second page and falls into a loop. If I do not fill in the "Notes" field and generate more than 8 pages (for example 12), and then make a record in "Notes" - the program immediately starts to give out text from 2nd page and fall into a loop again.
I'm using "LM studio" and "Mistral Nemo instruct 2407 Q8-0" by the way, but with "Qwen2.5 7b instruct" i have the same problem.

 

[ScribiltyBibilty]

The same issue, but I noticed one pattern. If I do not write anything in the "Notes" field, then everything works fine and events are generated further, but early details begin to be forgotten until events from the past was completely disappears from "story". If I write notes in the "Notes" field - then after the 7th page the program starts to give out text from the second page and falls into a loop. If I do not fill in the "Notes" field and generate more than 8 pages (for example 12), and then make a record in "Notes" - the program immediately starts to give out text from 2nd page and fall into a loop again.
I'm using "LM studio" and "Mistral Nemo instruct 2407 Q8-0" by the way, but with "Qwen2.5 7b instruct" i have the same problem.

For folks running into repetition issues, what context size are you using and if you're running locally, what models, quants, and inference settings (temperature, top-k, min-p, top-p..)? I can try to replicate in my local.

For reference, I run 24b models with 24k context and can get about a hundred or so prompts before I run out of context.

 

[Breedking]

I'm running the abliterated q5 nemo with 25k tokens in LM, 20k in the game so there's room for issues. I mostly think it's the AI getting goofy for me more than anything since rolling back and changing the prompt slightly fixes it for me.

 

[X_Tigershark_X]

For folks running into repetition issues, what context size are you using and if you're running locally, what models, quants, and inference settings (temperature, top-k, min-p, top-p..)? I can try to replicate in my local.

For reference, I run 24b models with 24k context and can get about a hundred or so prompts before I run out of context.

Model repository: lmstudio-community/Mistral-Nemo-Instruct-2407-GGUF
Model name: Mistral-Nemo-Instruct-2407-GGUF/Mistral-Nemo-Instruct-2407-Q8_0.gguf

Spoiler: Info

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

 

[ScribiltyBibilty]

Model repository: lmstudio-community/Mistral-Nemo-Instruct-2407-GGUF
Model name: Mistral-Nemo-Instruct-2407-GGUF/Mistral-Nemo-Instruct-2407-Q8_0.gguf

Spoiler: Info

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

Couple of things here:

• Low context size
• Some Mistral models start to favor phrases and repeat them ad nauseam to the point where I don't use them anymore

Loaded up Slime Outbreak and managed to start getting it to repeat similar phrases like "In the labyrinthine bowels of Ashburn" and "In the gloomy recesses of Ashburn's forgotten underbelly" at the start of each scene. This started happening once I ran out of context, about 15 prompts in. Increasing the context size to something more substantial abated this issue but that's only an option if you have the vram for it. Try increasing it in LM Studio until you can't load the model at all, then set the game's context size to something slightly smaller.

 

[X_Tigershark_X]

Model repository: lmstudio-community/Mistral-Nemo-Instruct-2407-GGUF
Model name: Mistral-Nemo-Instruct-2407-GGUF/Mistral-Nemo-Instruct-2407-Q8_0.gguf

Spoiler: Info

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

Well... I set the context length to 30k, increased GPU Offload, doubled evaluation batch size, increased temperature from 0.8 to 0.95, and set repeat penalty 1,5 instead of 1,1. And now i don't see any looping with "Notes" after 20 generated pages. Really don't know is this trouble with context length or repeat penalty parameters. I don't know anything about LM studio or models at all, and only now I saw that i can change the parameters

Couple of things here:

• Low context size
• Some Mistral models start to favor phrases and repeat them ad nauseam to the point where I don't use them anymore

Loaded up Slime Outbreak and managed to start getting it to repeat similar phrases like "In the labyrinthine bowels of Ashburn" and "In the gloomy recesses of Ashburn's forgotten underbelly" at the start of each scene. This started happening once I ran out of context, about 15 prompts in. Increasing the context size to something more substantial abated this issue but that's only an option if you have the vram for it. Try increasing it in LM Studio until you can't load the model at all, then set the game's context size to something slightly smaller.

Yes, now I realized that the problem may be somewhere in LM parameters. Now everything works more or less smoothly, thank you for your time and help

 

[ScribiltyBibilty]

I'm running the abliterated q5 nemo with 25k tokens in LM, 20k in the game so there's room for issues. I mostly think it's the AI getting goofy for me more than anything since rolling back and changing the prompt slightly fixes it for me.

I've never found an abliterated model that I've liked. Always too lobotomized for instruct purposes. Since you're in the 12b camp, I'd suggest mradermacher/Neona-12B-GGUF or TheDrummer/Tiger-Gemma-12B-v3.

I use this as a quick way to find intelligent and willing AI models to run locally:

You must be registered to see the links

Found lots of gems here. Main thing I sort by is Willingness and UGI. (Based on what my hardware can run - 64GB of vram)

 

[Breedking]

I've never found an abliterated model that I've liked. Always too lobotomized for instruct purposes. Since you're in the 12b camp, I'd suggest mradermacher/Neona-12B-GGUF or TheDrummer/Tiger-Gemma-12B-v3.

I use this as a quick way to find intelligent and willing AI models to run locally:

You must be registered to see the links

Found lots of gems here. Main thing I sort by is Willingness and UGI. (Based on what my hardware can run - 64GB of vram)

I'll give those a try and report back.

 

[roddy111.]

Just asking did the game stop working for anybody else this is just my speculation but did the server go down again I could be wrong.

 

[Veilwolf]

Kinda reminds me of bethesda's stuff fun but the longer you play it the more unstable it gets. I thought AI was supposed to be smart it this programming makes a kid with down's syndrome look like a fucking genius almost makes me want to get onboard with keeping AI data centers out of my state not only a strain on the power grid but useless with complex task.

 

[roddy111.]

Well it's down again where to get something to eat guess I'm not playing a game they really need to find a better server or do something they're gonna lose players if this keeps happening.

 

[fierylion]

server died ya, should be back now

they're gonna lose players if this keeps happening.

tbh I would prefer if everyone use their own AI to play, but I know most folks don't have the technical know how or just can't afford to use LLM so I setup a free server for that. The default AI isn't the smartest tho

 

[Veilwolf]

fierylion in no way am I dissing you, I am a big fan of your works since I ran across "Streaming with Portals on Newgrounds. Your the first that I have seen to create an interactive story using ai. What I am saying I think the AI your using is having too many hiccups, may I suggest using AI Horde.

 

[fierylion]

fierylion in no way am I dissing you, I am a big fan of your works since I ran across "Streaming with Portals on Newgrounds. Your the first that I have seen to create an interactive story using ai. What I am saying I think the AI your using is having too many hiccups, may I suggest using AI Horde.

I don't have enough kudos to run this on the horde, maybe I can make a guide so those with kudos can use horde endpoint instead. The default AI is gptq mistral nemo model which isnt very smart yeah, but its what I can offer for free...

 

[Veilwolf]

I don't have enough kudos to run this on the horde, maybe I can make a guide so those with kudos can use horde endpoint instead. The default AI is gptq mistral nemo model which isnt very smart yeah, but its what I can offer for free...

AiHorde is free I believe ChatGPT is also free but I don't run ChatGPT too many varibles. You should look in to Open Source AI programs there are a wide array to choose from hopefully you can find a model that can fit your programming style.

 

[fierylion]

AiHorde is free I believe ChatGPT is also free but I don't run ChatGPT too many varibles. You should look in to Open Source AI programs there are a wide array to choose from hopefully you can find a model that can fit your programming style.

the game is hosted using an open source AI ya, mistral nemo, also aihorde using anon key is extremely limited, especially for text gen

 

[Serpream]

Seems it was another bug again >.<

LM Studio update
Build 3

• Fixed a bug where some Mistral model's correct tool calls would fail to parse

I was so so so dam close to having a consistent and detailed flowing transformation narrative with full working tables and everything.
Now I am stuck in loops again after making a few edits, but worse LM Studio also updated, and I am getting endless loops after 8 pages now.
No idea if something I did or the LM Studios update >.>

 

[tacosnap123]

any more base updates for this or will it get marked as completed