New files... why?

#1
by FlareRebellion - opened

Hi Artefact2,

I notice you're reuploading everything.

Was there a bug in the previous quants? Do we need to redownload?

I accidentally quantized the base model to a f16 gguf, but the base weights are bf16 and it's a lossy conversion. The new files are quantized from a f32 (bf16 to f32 is lossless) gguf.

I'm not sure if the difference is enough to be noticeable.

Sign up or log in to comment