This is a reasoning model that almost always shows the <think> prefix, even outside of RP. It was a quick fine-tune done just for fun.
It works terribly in languages other than English.
Can be used, but still dumb.
Don't evaluate this as something serious at the moment.
Training Details
- Sequence Length: 16384
- Epochs: 5 epoch
- Full fine-tuning
- Learning Rate: 0.00005
- Scheduler: Cosine
- Total batch size (4 x 8 x 1) = 32
- Downloads last month
- 5
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support