Inference providers reportedly map medium/high settings to 50
A user says many providers appear not to follow a model’s intended mapping and urges them to serve it as designed.
TLDR
A user reports seeing inference providers map medium/high settings to 50, saying many appear not to follow the intended mapping. They speculate that the model’s extensive reasoning could explain the mismatch but urge providers to follow the intended setup.
Combined views
—
1 Source, first seen ago
