Training in Python, serving in Rust
Learning how inference works has made me separate the model from the server that runs it.
Read noteLearning in public
I’m a senior software engineer learning to build models from the ground up. This is a record of the maths, experiments, mistakes, and small discoveries along the way. Some entries are quick notes; others take a little more time.
Learning how inference works has made me separate the model from the server that runs it.
Read noteImplementing gradients for two hidden neurons made backpropagation feel less abstract.
Read noteWhy building a TypeScript language model made me rethink a familiar compiler concept.
Read noteBuilding my first language model is teaching me that the dataset is just as important as the architecture.
Read noteWhy stacking layers isn't enough to make a neural network learn complex patterns.
Read noteIn my journey into neural networks I've finally learned about neurons that stop learning.
Read noteLinear regression models and calculating probability has taught me a lot about the meaning of zero.
Read noteA small data preparation mistake that broke my first model.
Read noteLearning about checkpointing and how vital it is in training models has been fascinating.
Read note