Notebooks designed to run in Colab. These require a GPU to run scripts. 

- gcloud_zeroshot_vqa.ipnyb: colab cloud adapted notebook for zeroshot evaluation.

	1. Have miniconda installed on Drive. Check python, pip and libraries are correctly installed on an environment. This notebook uses the environment named py374.
	2. Run the cells, mounting miniconda, cloning BiomedGPT repository, and downloading the instruct versions of BiomedGPT.
	3. Upload your dataset (folder with test.tsv) to datasets/finetuning/ and the gcloud_evaluate_unconstrained.sh to scripts/vqa/. 
	4.  Run the script on the environment. Parameters include dataset ($1 test or val), beam size ($2) and dataset ($3). Feel free to modify the script.

- gcloud_finetune_vqa.ipnyb: colab cloud adapted notebook for finetuning vqa dataset.

	1. Have miniconda installed on Drive. Check python, pip and libraries are correctly installed on an environment. This notebook uses the environment named py374.
	2. Run the cells, mounting miniconda, cloning BiomedGPT repository, and downloading BiomedGPT models.
	3. Upload your dataset (.tsv and .pkl file) to datasets/finetuning/dataset and the gcloud_train_X_beam_scale.sh to scripts/vqa/. 
	4.  Run the script on the environment. Parameters include dataset ($1). Feel free to modify the script.

- gcloud_finetune_evaluate.ipnyb: colab cloud adapted notebook for fine tuned evaluation.

	1. Have miniconda installed on Drive. Check python, pip and libraries are correctly installed on an environment. This notebook uses the environment named py374.
	2. Run the cells, mounting miniconda, cloning BiomedGPT repository, and uploading the fine tuned model.
	3. Upload your dataset (.tsv and .pkl file) to datasets/finetuning/dataset and the gcloud_evaluate_X_beam.sh to scripts/vqa/. 
	4.  Run the script on the environment. Parameters include dataset ($1 test or val), and dataset ($2). Feel free to modify the script.