Thilina Rajapakse comments

Results 57 comments of


Thilina Rajapakse

An example for classification fine tuning

It's finetuning.

sentencepiece tokenizer issue

Does this happen for newly trained models as well? Newly trained as in, if you train a model with the latest library versions and try to load it.

sentencepiece tokenizer issue

Is this happening when loading a T5 model with Simple Transformers? e.g.: ``` from simpletransformers.t5 import T5Model model = T5Model("t5", "t5-base") ```

Binary classification .predict raises a ValueError: could not broadcast input array from shape (2,2) into shape (1,2)

1. Does this only happen with the latest version? 2. Does it only happen with the model that you trained with an earlier version when trying to use it with...

Loading a fine tuned Seq2Seq MarianMT model gives wrong predictions

Do you get any warnings when you reload the model? (Set up logging if you haven't: `logging.basicConfig(level=logging.INFO)`) Does it work as expected if you reload the model with Simple Transformers...

Loading a fine tuned Seq2Seq MarianMT model gives wrong predictions

To load with ST, you'd do: ``` model = Seq2SeqModel( encoder_decoder_type="marian", encoder_decoder_name="", args=model_args, use_cuda=True, ) ``` In theory, `Seq2SeqModel.from_pretrained()` is also supported since ST uses a Huggingface model under the...

model.train_model(train_df, eval_data=eval_df, **eval_metrics) is giving error "All arrays must be of the same length"

For a sequence-to-sequence model like mT5, you also need to set `evaluate_generated_text = True` in your model args. This could however slow down evaluation significantly.