My alternative quantizations.

#1
by ZeroWw - opened

These are my own quantizations (updated almost daily).

The difference with normal quantizations is that I quantize the output and embed tensors to f16.
and the other tensors to 15_k,q6_k or q8_0.
This creates models that are little or not degraded at all and have a smaller size.
They run at about 3-6 t/sec on CPU only using llama.cpp
And obviously faster on computers with potent GPUs

Owner

Thank you, I didn't have time to look before, that interesting!

Thank you, I didn't have time to look before, that interesting!

the quantization naming is a mess.. anyway it's explained here: https://huggingface.co/RobertSinclair

Sign up or log in to comment