SaisExperiments/Gemma-2-2B-Stheno-Filtered
#272
by
SaisExperiments
- opened
I've completed round two of training, The last quants helped as i discovered they leak at the end consistently which i'm 99% sure is because i used the wrong prompt template (Gemma-1 in place of Gemma-2)
This model is trained on the same prompt template as the first so i should have a definitive answer about my prompt templates if they leak too
SaisExperiments/Gemma-2-2B-Stheno-Filtered
Yeah i used the wrong ones x.x
Gemma 2: {{ bos_token }}
Whatever i misused: {{ '<bos>' }}
An extra thanks for the quants, wouldn't have noticed it otherwise ^^
Good to hear it was useful :) Also: queued!
mradermacher
changed discussion status to
closed