/lab · live compare
Live SOAP compare
Pick a fixture or paste a transcript, select arms, generate side-by-side notes, and read live soap-eval scores (including fluency/PPL when GPT-2 is available). Golden / held-out fixtures are on /golden-sets; historical score boards on /eval; metric definitions on /eval-methods.
Catalog: 0
FT create paused · Together for LLM arms · classic local
0 shown
Arms
Results appear here after a run — quality first, then safety / grounding / structure.