| ▲ | sroussey 8 hours ago | |
I thought these days thinking tokens sent my the model (as opposed to used internally) were just for the users benefit. When you send the convo back you have to strip the thinking stuff for next turn. Or is that just local models? | ||