Pierce Lamb

Layoutlmv2 documents with multiple pages

2

Hi Niels, I am attempting to build off of your [Layoutlmv2 example](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/LayoutLMv2/RVL-CDIP/Fine_tuning_LayoutLMv2ForSequenceClassification_on_RVL_CDIP.ipynb) except I am using my own Amazon Textract results instead of the internal OCR engine in LayoutLMv2Processor. The...

allow_val_change parameter is not passed to wandb.config.update in the WandBTracker

3

As far as I know, setting `allow_val_change=True` when calling `wandb.init(...)` allows you to have hyperparameter values that change over time (like a scheduled learning rate). However, for this to work,...

After countless hours of trying to get an `Estimator()` to run on a custom `image_uri` in smdistributed/dataparllel mode (it was failing on trying to import any non-sagemaker-DLC library), I finally...

I am currently loading the checkpoint on huggingface `udop-unimodel-large-224` and immediately trying to `torch.jit.trace` or `torch.onnx.export` it with input data provided for the `input_ids`, `attention_mask`, `labels`, `seg_data`, `visual_seg_data`, `decoder_attention_mask` and...

Can the provided models perform multi-label classification?

1

The provided code examples provide a good example of single label classification (e.g. rvlcdip). But can this be easily extended to multi-label classification? I'm a bit new to models where...

I followed https://github.com/microsoft/i-Code/issues/17#issuecomment-1416369657 in order to load the `UdopTokenizer`. I then followed the code examples for tokenizing text provided in [rvlcdip.py](https://github.com/microsoft/i-Code/blob/main/i-Code-Doc/core/datasets/rvlcdip.py) This amounts to [calling](https://github.com/microsoft/i-Code/blob/7d5810e621a9c98ddce3ec0d5899f8454dbdaaf1/i-Code-Doc/core/datasets/rvlcdip.py#L280) `tokenizer.tokenize(text)` on a word text,...

All I can find relative to distributed evaluation in huggingface is this mention of it in the quick tour: https://huggingface.co/docs/evaluate/a_quick_tour#distributed-evaluation Is there a doc or a code example anywhere that...

Can `VisionEncoderDecoderModel.generate()` work with batched data?

2

Sorry if this is the wrong place to post this. I'm currently trying to finetune Donut using your excellent fine-tuning guide as a starting point. As a test, I am...

Donut finetuning RVL-CDIP ipynb -- add class names to tokenizer as empty strings?

1

The [ipynb](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Donut/RVL-CDIP/Fine_tune_Donut_on_toy_RVL_CDIP_(document_image_classification).ipynb) states: ``` Prepare dataset The first thing we'll do is add the class names as added tokens to the vocabulary of the decoder of Donut, and the corresponding...

Does the recent news about speeding up pytorch model loading apply to huggingface transformers APIs?

2

https://twitter.com/RisingSayak/status/1756634311493890559 Can this apply to using transfomers APIs? Take the below for example: ```python model = DistilBertForSequenceClassification.from_pretrained(model_path, local_files_only=True).to( torch.device("cpu") ) ``` Can this be modified to load faster?