https://huggingface.co/AuraIndustries/Aura-MoE-2x4B
#496
by
jeiku
- opened
https://huggingface.co/AuraIndustries/Aura-MoE-2x4B
Trying a MoE this time, FFT on the KTO set to heal. It should work in llama.cpp, I was able to convert with no issues.
Thank you!
Then I am not worried :)
mradermacher
changed discussion status to
closed
Unfortunately Missing importance matrix for tensor blk.0.ffn_gate_exps.weight in a very low-bit quantization, meaning no good imatrix quants.
That happens when the tensors are not covered 100% during imatrix measurements. For whatever reason.
Ah... Something was bound to go wrong with this abomination of a model. No big deal to me to just have static quants for this. Thanks for letting me know!
sure :)