open-llms
open-llms copied to clipboard
Adding new entries
Hello,
I'm queueing up some updates for new entries and have some questions:
- Do you prefer pull requests with multiple new entries, or separate pull requests for each new entry?
- There seem to be some releases which claim free commercial licensing, if you register or notify them. E.g.: https://github.com/InternLM/InternLM?tab=readme-ov-file#license Do you want do add such entries?
first, thank you for offering to contribute!
one PR per entry might be simpler to review and approve.
for those that require registration/notification, lets leave out for now? unless they're llama3 quality it may make the list hard to navigate
Ok.
For the licenses I was thinking about adding some columns with symbols or simple checkmarks for various use-cases. Because, e.g. for synthetic dataset creation it is already very difficult to find a model. "Requires registration" could be just another part of this information.
The registration models seem to be primarily from China, but I have not checked their benchmarks in detail.
I came across a number of additional questions:
- WizardLM 2 was released in April, but the official weights were removed. However, there are unofficial Huggingface repos with copies of the files. Should I add these?
- There is stuff like Starling-LM-7B-beta, which is under an open license, but trained of OpenAI outputs, so is "not allowed to compete with OpenAI" - whatever that means... What is your take on adding such models?
- For MoE models it is common to add the number of active parameters in addition to the total parameters. Should these be added?