Matrixfag here. We've got 2 things to share. The first: you can now save and load your chats in our Colab notebook. The second: Our "Experiment 2" version (also called V3) of the 6B model, which has been in the "experimental" branch for a bit, has now officially been ported to the main branch. Reception to V3 has been really positive so far, so we figured it was time to mark it down as a new stable version. As it happened, we managed to get another fine-tune done on new data at the same time we were ready for release. This new model, Experiment 3 (also called V4) is currently being pushed onto the experimental branch. We'll notify you when it's been fully uploaded. Picrel is the changelog for both V3 and V4.
Saving/loading chats has been a thing requested ever since the notebook first came out, and we're finally happy to deliver it. It took me personally a while to wrestle with the code for this, but it should hopefully be working. If it's bugged, let us know so that we can get to fixing that ASAP. Your chat will be saved in a .json format, much like character definitions are. Please keep in mind that character definitions and chats are saved separately, and BOTH must be loaded in order to properly resume a chat! If character defs aren't loaded, you'll get weird things like the bot's future chats starting with just ":" instead of the proper "[CHARACTER]:".
We've published our logbooks (aka model changelogs) as well at
https://github.com/PygmalionAI/logbooks. We hope you all enjoy the new 6B version and saving/loading chats. It's been a little while coming. Your support has been thunderous so far - we've recieved over 11,000 downloads on HuggingFace in just 3 days, absolutely smashing our expectations. Thank you for all of it.
NOTEBOOK:
https://colab.research.google.com/github/PygmalionAI/gradio-ui/blob/master/notebooks/GPU.ipynbRENTRY:
https://rentry.org/pygmalion-aiStill taking chatlogs!:
https://rentry.org/chatlog-dumping 