Issue not found in original model: model stops after thinking
Once in a while, the model seems to stop generating after thinking. The final answer is contained within the thinking blocks, but outside them, nothing is generated. I've tested and it seems that this doesn't happen in the original model, so it's likely due to this fine-tune. I've tried both Q4 and Q8 quants, same issue.
Seems that literally telling the model to fix itself helps. This system prompt drastically reduces instances of this happening:
"Remember that the final answer should be written outside the thinking blocks."
Seems that literally telling the model to fix itself helps. This system prompt drastically reduces instances of this happening:
"Remember that the final answer should be written outside the thinking blocks."
Thanks! I noticed the same issue. Usually a regen would fix it, but it was still annoying.