mrseeker

Green Devil live on Pirr

As some of you have noticed, I have been quite busy lately, and it seems not only me but also the community.

I can finally publically announce the public arrival of "Green Devil" exclusively on the Pirr app. They managed to bugfix some appar...

OPT-30B-Erebus-V1 (pipelined)

It took me a while, but I am currently training Erebus-v1 op OPT-30B. I wish I could get higher, but this is a good start to seeing how my training schedule behaves on bigger batches.

I have to say, I was shocked at seeing how much this wou...

Erebus V2 & future developments

New year, new chances to deliver one of the best fine tunes I can offer.

I started working today on a "test version" of Erebus-v2. This one has been on the shelf since my holiday, but I haven't had time to test it yet. The dataset is three t...

Holiday season

I know it's been a while since I posted a message here, but I want to tell you all that I am just swamped in the background with other things other than working on KoboldAI models. Work and life are stretching me to the point where I want some fre...

Data size literotica

Someone asked me, why does it take so long to get a dataset going?

This is what I currently have from Literotica:

36G     literotica/high-quality
57G     literotica/medium-quality
18G &nbs...

Status update

I have been on a hiatus lately; sorry about that. After I released Erebus-13B, the model went viral on 4chan and people loved it. Hearing that my model performed well, I felt like I was onto something great.

I know a lot of you are currently...

New dataset: "Erebus"

I have finally done it: I am currently cleaning and compiling a new dataset called "Erebus".

Some specifications:
Size: 4Gb (200k stories)  (before cleaning)
Contains:
- Dataset G (Pixiv)
- Literotica (4.5 or higher)
- S...

New model type

Just two days ago, one of the projects I am involved in pushed out an update that I eagerly awaited since it meant that my AI models would become much better than they were before. It also meant that I now have a project similar to my "ModronAI" p...

License issues

As some have noticed, I have released OPT-6B-Nerys-v2 on KoboldAI. I am seeing that BLOOM also has a modified license with restrictions sparked some thoughts. Why should I be doing finetunes on those models? And I want to write a clear message to ...

Nerys 13B version 2

(sorry for the delay)
Hello all,

I am looking for some people that are willing to test out my v2 model of Nerys. This one is similar to the v1 model, but I have removed a lot of chapter headings and empty spaces that could be found in the...

Facebook OPT finetuning

Yes, you read the title right, I am currently in the process of making "yet another" model. This time I am working on the facebook OPT models. They do have a difference compared to my normal models, they are released under a CC-BY-NC-SA licence. T...

Hybrid models incoming!

I couldn't be more proud of this one: I have a new fairseq-dense model in the pipeline called "Nerys".  It's a combination of the "Pike" dataset (20% bigger than Janeway) with some Asian novels, the CYS dataset AND some good stories from Shin...

Narrator

It's been a while since I worked on KoboldAI models but that does not mean I have been busy behind the scenes. One of my goals my the "bucket list" is to have an AI give a TED talk, most likely about questioning the philosophy of an AI being a "se...

Big, Bigger, Biggest!

I am happy to announce that we have now an entire family of models (thanks to Vast.AI), ready to be released soon! In the coming days, the following models will be released to KoboldAI when I can confirm that they are functional and working. If yo...

April update

I wish I could write this sooner, but it takes quite some time to write posts. With all the people pinging me up to the point where my wife starts complaining, I am taking more time talking to people than actually doing work.

Either way, I f...