It seems like model have serious repetition issues (both gguf and on openrouter)
#8
by
roadtoagi
- opened
Dry may help, but it will kill reasoning mode...
and in reasoning mode, model starts to repeat itself from second-third chat round.
Using recommended settings.
With some more testing even dry doesn't really help. Deepseek is just much better.
Apologies @roadtoagi which quant did you use, the old ones may have to be deprecated due not being compatible with imatrix quantization
We deleted the ones that are wrong and only left the ones that work
I used Q2_K, but the problem isn't in quants, but in model itself. Openrouter has same issues.
I used Q2_K, but the problem isn't in quants, but in model itself. Openrouter has same issues.
I think it was a chat template issue. We just fixed them 2 hours ago. would you mind checking them again?