Text Generation
Transformers
PyTorch
llama
Not-For-All-Audiences
nsfw
text-generation-inference
Inference Endpoints
Solid model. Tends to write too much.
#5
by
snombler
- opened
Here's a MikuMaid sample:
Calla Log
Model tends to write too much, even when instructed not to. The first two gens in the log above were clipped down to two paragraphs in hopes it would mimic, but no such luck. Consistent problem across all cards tested with all prompt templates tested. Miqu and older Mistral/Mixtral models often have the same problem.
Attention seems to be solid. Word choice is decent. Just can't throw stopping tokens after a reasonable period.
snombler
changed discussion status to
closed