The model gets stuck in a loop and starts writing all sorts of nonsense.
#1
by amaycom - opened
This comment has been hidden (marked as Off-Topic)
I also get stuck in lm studio with default config for GLM-4.7-Flash-MLX-4bit.
- with the following config, the response finally works
- temperature: 0.7
- repeat penalty: 1.05
- top-p: 0.95
- i'm not using temperature 1.0 as recommended, because it often goes into loop. 0.7 works well for me