Скачать с ютуб Training and Testing an Italian BERT - Transformers From Scratch #4 в хорошем качестве

Training and Testing an Italian BERT - Transformers From Scratch #4

We need two things for training, our DataLoader and a model. The DataLoader we have — but no model. For training, we need a raw (not pre-trained) RobertaForMaskedLM. To create that, we first need to create a RoBERTa config object to describe the parameters we’d like to initialize FiliBERTo with. Once we have our model, we set up our training loop and train! Post-training, we'll test the model with Laura, who is Italian - and hope for the best. Part 1:    • How-to Use HuggingFace's Datasets - T...   Part 2:    • Build a Custom Transformer Tokenizer ...   Part 3:    • Building MLM Training Input Pipeline ...   --- 📙 Medium article: https://towardsdatascience.com/how-to... 📖 If membership is too expensive - here's a free link: https://towardsdatascience.com/how-to... 🤖 70% Discount on the NLP With Transformers in Python course: https://bit.ly/3DFvvY5 👾 Discord   / discord   🕹️ Free AI-Powered Code Refactoring with Sourcery: https://sourcery.ai/?utm_source=YouTu... 00:00 Intro 00:35 Review of Code 02:02 Config Object 06:28 Setup For Training 10:30 Training Loop 14:57 Dealing With CUDA Errors 16:17 Training Results 19:52 Loss 21:18 Fill-mask Pipeline For Testing 21:54 Testing With Laura