3 ms·
In lots of applications, all the computations already happen inside native libraries, e.g. Numpy, PyTorch, TensorFlow, JAX etc. And if you have a complicate co
by albertzeyer 3y ago
In lots of applications, all the computations already happen inside native libraries, e.g. Numpy, PyTorch, TensorFlow, JAX etc.
And if you have a complicate computation graph, there are already JITs on this level, based on Python code, e.g. see torch.compile, or TF XLA (done by default via tf.function), JAX, etc.
It's also important to do JIT on this level, to really be able to fuse CUDA ops, etc. A generic Python JIT probably cannot really do this, as this is CUDA specific, or TPU specific, etc.