Dirk Groeneveld

Results 84 issues of Dirk Groeneveld

WiC is one of the tasks of the [SuperGLUE](https://super.gluebenchmark.com) benchmark. The task is to re-trace the steps of Facebook's RoBERTa paper (https://arxiv.org/pdf/1907.11692.pdf) and build an AllenNLP config that reads the...

Contributions welcome
Models
easy

COPA is one of the tasks of the [SuperGLUE](https://super.gluebenchmark.com) benchmark. The task is to re-trace the steps of Facebook's RoBERTa paper (https://arxiv.org/pdf/1907.11692.pdf) and build an AllenNLP config that reads the...

Contributions welcome
Models
easy

CommitmentBank is one of the tasks of the [SuperGLUE](https://super.gluebenchmark.com) benchmark. The task is to re-trace the steps of Facebook's RoBERTa paper (https://arxiv.org/pdf/1907.11692.pdf) and build an AllenNLP config that reads the...

Contributions welcome
Models
easy

In the referring expressions task, the model is given an image and an expression, and has to find a bounding box in the image for the thing that the expression...

Contributions welcome
Models
hard

[DETR](https://github.com/facebookresearch/detr) is an interesting object detection model that fits into our `RegionDetector` abstraction. This task is about porting the DETR model to AllenNLP and trying it on all the vision+language...

Contributions welcome
Models
medium

This is a project in computer vision, rather than natural language processing. It is here because we have found this `RegionEmbedder` to be important for down-stream tasks that combine vision...

Contributions welcome
Models
hard

When you use `numpy.zeros` to create the `embed_arr`, you can't later add it to `embed_vector`, because `embed_vector` might not be a numpy array. This re-works the code such that the...

gpu

I have some sample code here: https://gist.github.com/dirkgr/bc0e8b547015db9e67c7 In short, the code creates a `scala.collections.mutable.HashMap` and adds a default value. When round-tripping it through a chill file, it loses the default...

I want to keep a list of features we're removing for performance reasons. I suspect a lot of these can come back when we start running the demo on GPUs....

... or remove them. Right now we think it's just fine-grained NER and BiDAF that still use ELMo.