catgrad

You like category theory? You like tinygrad? You love catgrad! ❤️

catgrad is a bit different: instead of using autograd to train, you compile your model's reverse pass into static code. This means your training loop can run without needing a deep learning framework (not even catgrad!)

Here is a linear model in catgrad:

model = layers.linear(BATCH_TYPE, INPUT_TYPE, OUTPUT_TYPE)

catgrad can compile this model...

CompiledModel, _, _ = compile_model(model, layers.sgd(learning_rate), layers.mse)

... into static code like this...

class CompiledModel:
    backend: ArrayBackend

    def predict(self, x1, x0):
        x2 = x0 @ x1
        return [x2]

    def step(self, x0, x1, x9):
        x4, x10 = (x0, x0)
        x11, x12 = (x1, x1)
        x16 = self.backend.constant(0.0001, Dtype.float32)
        # ... snip ...
        x18 = x17 * x5
        x2 = x10 - x18
        return [x2]

... so you can train your model by just iterating step; no autograd needed:

for i in range(0, NUM_ITER):
    p = step(p, x, y)

Catgrad doesn't just compile to Python: I'm working on support for other targets like C++ (ggml), CUDA, FPGAs, and more!

Catgrad uses reverse derivatives and open hypergraphs to transform a model into its backwards pass. For details, see this paper.

Install

pip install catgrad

Examples

Train simple MLPs for the iris dataset:

./data/get-iris-data.sh
python3 -m examples.iris (linear|simple|dense|hidden)

Compilation Targets

Target backends we plan to support soon:

• [x] Python/numpy
• [x] Python/torch
• [ ] Python/tinygrad
• [ ] C++/GGML