ubergarm

/

GLM-5.1-GGUF

like 20

Appreciate your kind words!

imatrix just finished, uploaded, and continuing with the process!

cheers!

i concur!

The folder IQ2_KS might be for IQ3_KS.

Before moving on to experiments with larger quants, could I request smol-IQ1_KT for 2x96 GB? Your GLM 5 quant in 2x96 GB has been great.

Thanks again for all the quants!

Edit: Oops, just saw your message at the top, my mistake.

@ndroidph

Thanks, i renamed it properly now to IQ3_KS.

The smol-IQ1_KT is uploaded now. It ran well out to 65k in one short opencode test.

Waiting on the BF16 perplexity now to generate the graphs!

I went ahead and released a larger quant this time too, the smol-IQ4_K is quite nice at just over 1% more perplexity in almost half the size of Q8_0

Its a good model, if you can run it!