Benchmarks on model card and report don't add up

#9
by patrickvonplaten - opened

What benchmark is correct here?

Screenshot 2025-03-12 at 15.04.10.png

Screenshot 2025-03-12 at 15.04.27.png

Pretty much nothing adding up here

One appears to be the IT model and the other appears to be the PT model? I might be wrong, however!

Ah yeah that makes sense - on this instruction-tuned model card we only see PT numbers? Would be good to transfer the IT numbers to this model card I think

The first image shows the benchmark evaluated for 5-shot while the 2nd one is for 0-shot.

Sign up or log in to comment