LayerScope is a language model tool that loads two language models of the same architecture and parameter size and generates a bar graph detailing which layers have the most difference between the models.
Hi! I have such a problem with GPT-J, OPT models and Pythia (with the same architecture and parameters, but trained differently):
Torch Version: 2.0.1+cpu
Total Layers Found: -1
Script completed, close graph to unload models and return to commandline.
Then there is an empty graph and the colorama does not even start.
Everything works well with Llama models.
Windows.
Am I doing something wrong? I can't figure out what the error is. Or are such models not supported by the script?