Mark

icon location ML Researcher. Except of the research, working on @pytorch torchtune.

Results 189 comments of Mark

Random results with 0.0 confidence on demo.py

Same problem

Random results with 0.0 confidence on demo.py

Tried CRNN model, same result

Random results with 0.0 confidence on demo.py

@ashkanmradi I fixed it with reinstalling all libraries and this repo.

Random results with 0.0 confidence on demo.py

@ashkanmradi Yes, I trained new model, old one returned random results with 0.0 confidence.

Random results with 0.0 confidence on demo.py

> I learned a new model

OCR accuracy is very low.

Any updates

[RFC] Long context fine tuning in torchtune

Is this still actual? I don't see any related PRs, but things that are proposed here are pretty reasonable.

Add Phi4

Should we wait until Phi-4 will be on HF?

Add Phi4

We require on changes in only tokenizer actually, as we already used full Attention in Phi3 without sliding one.

Add Phi4

I assume that I need to do run with Phi4