Csanád Póda

Requirements misleading

1

In your requirements, you're only listing Python 3.7+. However, having Python 3.7+ is not enough. I have recently tried to use this tool with Tensorflow 2.1.0 installed in addition to...

Error: "Make sure `_init_weights` is implemented for <class 'donut.model.DonutModel'>"

6

I've tried using the official pretrained CORD model, both v1 and v2. I set up a virtual environment with Python 3.7, and add the dependencies as specified in the Readme:...

training_loss doesn't change

5

I'm trying to pretrain Donut on Hungarian, and I'm following the steps described in the documentation. I've successfully used SynthDOG to generate a dataset, and I'm running the provided script...

How did training with a batch size of 8 fit onto a single A100?

2

In the "Training" section, you mention you used a single A100 with the attached config yaml. An A100 has either 40 or 80GB of VRAM. The batch size is set...

Please expand wiki with task_names!

1

I wanted to check how the v2 version of the cord model performs, and I couldn't get it to run. It just returned some "text_sequence" with only " " characters...

Continuing from checkpoint results in: ValueError: loaded state dict contains a parameter group that doesn't match the size of optimizer's group

1

I've wanted to pretrain the model to a new language, so I ran it on a dataset for 30 epochs. When training, the logger showed 200 M trainable params. After...

Trying to `train.py` a new language based on a corpora I generated with synthDOG, running the command `python train.py --config config/base.yaml --exp_version "base"` on up-to-date Windows 11 inside a conda...

_sentencepiece.SentencePieceProcessor_LoadFromFile can't load existing file.

1

I'm trying to run Donut on my Windows machine. As per the documentation, I1m trying to run the following code: ``` import re from transformers import DonutProcessor, VisionEncoderDecoderModel from datasets...