#R1 hallucinating input.
12 messages · Page 1 of 1 (latest)
The provider for this gen was together
The two previous successful gens were deepinfra
Please let me know if you need any additional information. I just really want to stop my paid deepseek gens from outputting the answers to random leetcode puzzles.
Notably, the incorrect responses never include correctly formatted reasoning
I've noticed this on some of my deepseek calls too, with both deepinfra and fireworks, mostly the former
I've had the opposite experience
After restricting the provider to only deepinfra I've had no further issues
is this from deepinfra?