TheBloke

/

Yi-34B-Chat-AWQ

like 11

This one was made by me, and that one was made by them!

Beyond that, I can't say. I don't know what dataset they used for AWQ calibration. I used an English-only dataset, VMware Open Instruct, which is a very good English dataset but doesn't contain any non-English samples.

My guess is that there's likely not much difference between them. They're both AWQ 4-bit 128g, so any difference from the calibration dataset is likely to be quite minimal.