AmberChat in top 5 most open LLM projects
Just a note of appreciation: AmberChat came in at #3 on the openness leaderboard at https://opening-up-chatgpt.github.io/ — a rare example (in our experience) of a new system increasing the average openness of the field rather than decreasing it.
Thank you so much for your note! We appreciate the work you are doing in the space.
As a side note, we released the training code last week. Please reach out with any feedback or other ideas on how LLM360 can meaningfully support open source.
Thanks for sharing this. The LLM360 team would like to make our model as open as possible. We design to release our base models in a open manner, and this leaderboard also makes me realize the additional efforts needed to make the chat models open. If you are interested, check out also the Crystal series
Btw, I notice the leaderboard indicate AmberChat's base to be Llama. AmberChat actually uses our own pretrained base model (https://huggingface.co/LLM360/Amber), though sharing the same architecture as Llama.
Updated — AmberChat now at position 2. Impressive work!
Thank you! And this framework is really helpful to us since it points out problems in our open-source attempt. Tbh, LLM360 is designed to target at base model openness and we didn't realize the efforts missing for chat model openness. We will keep improving on this.