Detailed conversation about Declarative AutoML with Piero Molino author of Ludwig and co-founder Predibase. 00:00 Intro 00:50 First meeting Piero at Strata in Moscone Center 10:50 Why Declarative AutoML? 12:00 Introducing Declarative ML Systems 16:44 Ludwig: declarative ML systems on PyTorch 18:06 Github Statistics for Ludwig 19:51 Ludwig training example via CLI 21:09 pip install ludwig and using the Programmatic API 26:32 How does Ludwig work? 26:53 Training with Ludwig 27:34 Predicting with Ludwig 29:59 Running Experiments with Ludwig 31:04 ludwig experiment –dataset 32:09 Input - Encoder - Decoder - Output 34:15 ParallelCNN encoding 35:42 Pretrained Transformers: bert distilbert t5 roberta gpt-2 40:44 concat combiner 41:18 Number features decoding 41:37 Vector featuers decoding 41:52 Sequence features decoding 42:30 Training parameters: batch_size, epochs, learning_rate, … 43:07 Preprocessing parameters 43:33 Speaker Verification 44:03 Expected Time of Delivery 44:35 Summarization 45:04 Distributed Training, Ludwig on Ray 45:23 Running Ludwig on Ray 47:18 Ludwig Hyperpot with RayTune (Advanced) 48:51 Ludwig on Kubernetes 49:15 Managed Ludwig in Predibase 49:54 Predibase: Low-code ML, High-Performance