Text Generation
Transformers
Safetensors
qwen2
conversational
text-generation-inference
Inference Endpoints

Recommended Samplers?

#1
by saishf - opened

So far the model seems interesting, but i'm struggling to find samplers that are reliable. I've figured out that high temps make it go insane but beyond that i can't figure it out.
I've tried all the ST defaults + many custom presets ~ 30ish πŸ˜Άβ€πŸŒ«οΈ

flammen.ai org

"Naive" seems to work well for the Mahou models, but I'm still trying to figure that out too haha

"Naive" seems to work well for the Mahou models, but I'm still trying to figure that out too haha

I wonder if throwing some boring instruct data into the tuning would help, as it tends to go off on it's own adventure mid rp 😭
It's rather interesting when using Mahou-1.3-Gemma2 in lmstudio, it struggles to follow instructions with chatml and fails with gemma presets to the point of roleplaying an assistant in 3rd person
Like with sao10k's Ana-Dataset
Or it could maybe the extra training is too strong and affecting the pretraining instruct following? i don't really understand the whole training thing πŸ˜Άβ€πŸŒ«οΈ

Sign up or log in to comment