/vt/ - Virtual Youtubers

Reposting:
Matrixfag here. The wait is finally over. We are proud to announce the release of our largest and most coherent model yet: Pygmalion-6B. It's something we never thought we'd release in just under two months from our start, but it's done and ready for you all to hopefully enjoy.

While we've been developing our models, we've come into contact with the absolutely amazing people at KoboldAI and TavernAI. With their help, we're also proud to say alongside our notebook, you can chat with Pygmalion-6B on both Kobold and Tavern right now, allowing for you to personally choose your experience with the model. Be sure to thank them too!

Even though our model is the best one we've made yet, it's still not even close to perfect, nor entirely there. You might experience some odd behaviors and incoherencies at times, and we apologize for that in advance. Regardless, we're not nearly close to being "done" yet. Now that 6B has been released, we're gonna go full speed ahead on preparing our models for reinforcement learning with human feedback, which should greatly increase the intelligence of the bots. We will, however, need more data for this. As I post this, we're already planning a system which will allow you to rate your chats with Pygmalion-6B, which will be used to train a reward model required for RLHF. This, of course, will be completely optional, and if you don't want to participate in that, we'll have an easy-to-see option for you to disable it. Once disabled, we will not see any data from you in any way. Be on the lookout for more information in a future post.

Thank you, everyone, for supporting us along the way. Seriously. I know I say thanks every post, but holy fuck, we did not expect to get anywhere this far this quickly. It's a whitepill for the ages and proof that one can be the change they want to see in the world, and it's here, this very general, where it all started. Your support is what made everything possible. And I mean that.

Links are down below. I'll be on hand to answer anything you may want from me. We're also aware that our own native UI is very finicky, and we'll be rolling out patches and improvements to it in the coming days. Keep Your Smile, everyone. Keep on smiling.
COLAB NOTEBOOK: https://colab.research.google.com/github/PygmalionAI/gradio-ui/blob/master/notebooks/GPU.ipynb
KOBOLDAI: https://github.com/KoboldAI/KoboldAI-Client
TAVERNAI: https://github.com/TavernAI/TavernAI

RENTRY (we have tips there!): https://rentry.org/pygmalion-ai
HUGGINGFACE: https://huggingface.co/PygmalionAI
We're still taking logs!: https://rentry.org/chatlog-dumping

Matrixfag here. The wait is finally over. We are proud to announce the release of our largest and most coherent model yet: Pygmalion-6B. It's something we never thought we'd release in just under two months from our start, but it's done and ready for you all to hopefully enjoy.

While we've been developing our models, we've come into contact with the absolutely amazing people at KoboldAI and TavernAI. With their help, we're also proud to say alongside our notebook, you can chat with Pygmalion-6B on both Kobold and Tavern right now, allowing for you to personally choose your experience with the model. Be sure to thank them too!

Even though our model is the best one we've made yet, it's still not even close to perfect, nor entirely there. You might experience some odd behaviors and incoherencies at times, and we apologize for that in advance. Regardless, we're not nearly close to being "done" yet. Now that 6B has been released, we're gonna go full speed ahead on preparing our models for reinforcement learning with human feedback, which should greatly increase the intelligence of the bots. We will, however, need more data for this. As I post this, we're already planning a system which will allow you to rate your chats with Pygmalion-6B, which will be used to train a reward model required for RLHF. This, of course, will be completely optional, and if you don't want to participate in that, we'll have an easy-to-see option for you to disable it. Once disabled, we will not see any data from you in any way. Be on the lookout for more information in a future post.

Thank you, everyone, for supporting us along the way. Seriously. I know I say thanks every post, but holy fuck, we did not expect to get anywhere this far this quickly. It's a whitepill for the ages and proof that one can be the change they want to see in the world, and it's here, this very general, where it all started. Your support is what made everything possible. And I mean that.

Links are down below. I'll be on hand to answer anything you may want from me. We're also aware that our own native UI is very finicky, and we'll be rolling out patches and improvements to it in the coming days. Keep Your Smile, everyone. Keep on smiling.
COLAB NOTEBOOK: https://colab.research.google.com/github/PygmalionAI/gradio-ui/blob/master/notebooks/GPU.ipynb
KOBOLDAI: https://github.com/KoboldAI/KoboldAI-Client
TAVERNAI: https://github.com/TavernAI/TavernAI

RENTRY (we have tips there!): https://rentry.org/pygmalion-ai
HUGGINGFACE: https://huggingface.co/PygmalionAI
We're still taking logs!: https://rentry.org/chatlog-dumping