From the course: Deep Learning: Getting Started

Using available open-source models

From the course: Deep Learning: Getting Started

Using available open-source models

“

- [Instructor] How do we select and use available open-source neural network models? We will briefly touch upon some key points in this video. Several open source models and their versions are available for use by anyone wishing to develop a neural network for their use case. These are fully trained models with updated parameters and corresponding hyper parameters. Training data and code will also be shared in many cases. Repositories like Hugging face and GitHub are popular locations where such models can be found. Universities also host models that are created from their research. These models are easy to download. They can be pretty large files though, and would need bandwidth and storage. Popular machine learning frameworks like PyTorch and TensorFlow provide capabilities to quickly load and use these models. How do we pick the right open-source model for our use case? First, when evaluating an open-source model, understand their original purpose and use case. This helps us understand what this model is good at and whether this is applicable for our use case. It is also important to understand the type of data that this model is trained on. Data could be public data or maybe focused on a specific use case. There can also be privacy and legal concerns that needs to be looked at. Explore the model's, popularity and usage. This usually can be found by the number of downloads, forks and blocks related to this model. Review licensing requirements for the model, even though it's open source, different categories of licenses determine how they can be used, and if attribution is required. Then proceed to download the model and build fine tuning or inference pipelines around it. Test the model with training data that is specific to your use case, to ensure that the model performs well in your specific scenario. We briefly touched upon open-source neural network models, and I recommend that you explore more about them and try using one of them.

Contents