metaseq icon indicating copy to clipboard operation
metaseq copied to clipboard

convert_to_singleton doesn't seem to handle bias properly

Open punitkoura opened this issue 1 year ago • 0 comments

  1. Take a 125m pretrained checkpoint.
  2. Consolidate the checkpoint using convert_to_singleton.py
  3. Try loading the model behind the metaseq API.

RuntimeError: Error(s) in loading state_dict for TransformerLanguageModel: Missing key(s) in state_dict: "decoder.layers.0.fc2.bias", "decoder.layers.1.fc2.bias",

punitkoura avatar Mar 22 '23 20:03 punitkoura