DavidAU

/

L3-SthenoMaidBlackroot-9.56B-V1-BRAINSTORM-8x-GGUF

like 4

I don't mind what people do with my models. If DavidAU is willing to make a Brainstorm version of it, then he's free to do so.

@Dunjeon @Casual-Autopsy

This repo has an 8x brainstorm model; this one however is a 40x.
You might want to check it out to see if this is what you are looking for... ;

https://huggingface.co/DavidAU/L3-DARKEST-PLANET-16.5B-GGUF

If you prefer 8x, no problem ;

Also; I will make you a deal:

I will create some quants (please advise specific ones) of L3-Super-Nova-RP-8B with Brainstorm , but I ask that you test it/them in the new software (I wrote this ) which enhances output of all models (mine and all other models) I just released here:

https://huggingface.co/DavidAU/AI_Autocorrect__Auto-Creative-Enhancement__Auto-Low-Quant-Optimization__gguf-exl2-hqq-SOFTWARE

Let me know; thanks David.

@DavidAU

I am a fan of your dark planet series, but this one was unknown to me. I am excited to test it out -- currently working with your 8x8b Death Star.

As to the other (Super Nova), that model is also of interest. I have cloned your enhancements script (mild for now) and have it locked and loaded. (Extremely impressive work, this. Thanks for showing it to me regardless of anything else!)

I will take your deal, and gladly do the tests on all the stuff and things.

I wonder, (am I shooting the moon here?) if I can ask for an 8x, and 40x? - default to 8x if no because of file size. (While I have 128gb of system ram and 16gb of vram + 64gb swap, I am also running an XTTS server on the GPU alongside the game Skyrim and the Mantilla AI Mod - the AI manages radiant and player to npc conversations - all at the same time.)

Quants of 2_k, 4_k_m,6_k, 8_0 would be great + max-D(?)

Thanks for your effort!

Okay , I will be working on this mon local time; repo will show up automatically during uploads.
40x will be significantly different from 8x ; but just as or more stable.
Note; I can only do one (upload bandwidth is slow here) - so I will go with 8x, which will be same formula as this repo.

Please clarify "max-d" - you mean with extended "bit" output tensor+ embed tensor? IE MAX Q8 ?

8x is perfect! And can we get a IQ4x_s? (seems to be the way to go(?))

And yes, I was reading your work on max... I may have been confused on designations - I thought one was more cpu intensive at the cost of tps? (I should review your docs...) Some of your ggufs have a CPU while others are missing it, but they have a D. I was opting out of the cpu option when I selected max-D (or so I thought).

You're the master here, do your thing! I will do the testing and get you the data for the page.

@Dunjeon

Repo is uploading now:

https://huggingface.co/DavidAU/L3-Super-Nova-RP-8B-V1-exp8-11-GGUF

I made 3 IQ4XS -> reg, MAX CPU, and MAX.
Allow a few hours for uploads;

Thanks!

I will begin testing this afternoon and get the data back to you in - in a community tab on that model. Thanks again!

Excellent.
Heads up; There will be an update in a day or so to the software: Beta V 1.1/1.2