Yeming Wen

Not sure for the model.fit, but for the functional api, returning randomvariable is fine in eager mode in my local script.

I did something like `WANDB_MODE=disabled python finetune.py --args` when doing a testing run.

same error here with A40.

Wondering if you have tried lower the gradient_accumulation_steps, with larger batch size, the accumulation steps can be smaller.

Thanks for the hardware requirement info! It seems only have the requirement for inference. I wonder if there is anything for fine-tuning on downsteam tasks?

Hi thanks for the suggestion! I also share a question with the other issue where the evaluation is slow. Running python seq2seq/run_seq2seq.py configs/eval.json takes 9 hours to complete on a...

I am looking into this but it will take some time for me to figure it out.

Hi, I am trying to time each example generation time. I found the generate method wrapper for the SpiderModel in https://github.com/ElementAI/picard/blob/main/seq2seq/utils/picard_model_wrapper.py However, I couldn't find the code which makes use...

oh, generate is invoked in the Seq2SeqTrainer's prediction_step method.