BETA - this page is not finished: more explanations and changes are to come, but it should be functional.
Not sure what to choose? Go for an imatrix quant, then make sure the table is sorted by quality, and then find the highest quality quant that will fit in your memory (model overhead can be large and depend on many factors, so 16GB VRAM might only be good for a 10GB quant or worse). More detailed instructions will follow.
If there are any issues with this page, feel free to open a discussion to notify us about it.
Privacy note: The current implementation of this page does not send any model names or other such information to our servers. When downloading the page itself, your IP address is recorded for technical reasons. Huggingface, cloudflare, amazon and probably others can still see what you are downloading, though.
production / test version of this page.