Gemma 3 models sadly don't work #10
Description
I would have loved to use this repo for speeding up my synthetic data generation for a German language dataset, but (after debugging some import issues in utils.py) I got an error saying that gemma3 isn't supported. Just to give it a shot I renamed some variables to trick the script into thinking gemma 3 is the same as gemma 1, but of course this just led to this error, at which point I gave up as it is past my expertise:
TypeError: ModelArgs.__init__() missing 8 required positional arguments: 'hidden_size', 'num_hidden_layers', 'intermediate_size', 'num_attention_heads', 'head_dim', 'rms_norm_eps', 'vocab_size', and 'num_key_value_heads'
With that said I attached the fixed utils.py