Hi I’m the maintainer of the top trending GGUF weights on HF right now.
Why is my download count zero? These are the flagship weights for a recent Mozilla launch that’s received a lot of public attention over the last few weeks. I know people are downloading it. But I have no way of knowing how many.
Is my count zero because I tell my users to download the weights using wget rather than the Python tools?
It seems they only count if someone is using
.from_pretrained() which is a problem for other models that are not exactly used that way.
I even tested, even if you use
huggingface_hub cli to download the GGUF file it won’t count it.
There are repos dedicated to GGUF models and they have download counts, I don’t know how is that possible do be honest. (maybe somebody by mistake used
from_pretrained(), but that should immediately fail)
UPDATE: they actually count config.json download, which if you download a file directly like *.gguf mode, you would never need that file. Hence the count 0.
The reason for that is because there’s no integration being set up yet for GGUF models.
Read more about enabling that here: Models Download Stats. Usually we have a tight integration with the software, e.g. in the Transformers library for instance, the download count is based on the number of times a user calls
from_pretrained to load a config.json.
Read more about how to integrate this into your software here: Integrate your library with the Hub