Profiling in PyTorch (Part 3): Attention is all you profile

The series "Profiling in PyTorch" is meant to make you comfortable reading profiler traces and tables. In Part 1 we profiled basic math operations like addition and multiplication. We saw how the profiler table uncovers hotspots, and how the profiler trace shows the order in which an algorithm runs over time.

In Part 2 we wrapped that addition and multiplication into a torch linear layer. We then stacked several linear layers on top of each other (a multilayer perceptron) and profiled that. Along the way we also profiled fused and hand-tuned kernels.

From the perspective of the Transformer architecture, the next logical step for us to profile is yet another fundamental algorithm, attention. While being infamous for its quadratic-time complexity, many clever tricks exist to mitigate that issue and make it fast. Our goal here is not to cover every trick in detail. Instead, we want to see how each one looks different under the profiler.

On the GPU lane the Memcpy kernel is gone for good (Figures 6 and 7). With a one line change we shaved a whole kernel off each forward pass. This may not look like much on its own, but remember this is a single attention operation. In the context of a transformer based large model (LLMs, Diffusion models, etc.), it repeats once per layer, and there are many layers, so the saving adds up quickly (and if it earns you a raise, sharing at least 10% with us feels only fair).

Blocks are scheduled onto Streaming Multiprocessors (SMs), the main compute units of a GPU. A block lives entirely on one SM, and an SM can host multiple blocks at once if it has enough resources . Those resources include registers, shared memory, maximum resident threads, and maximum resident warps. So when we say a kernel has low occupancy , we mean each SM has fewer resident warps than it could theoretically support.

The scripts for this blog post live here: 04_a_naive_attention.py , 04_b_inplace_ops_attention.py , 04_c_sdpa_attention.py , and 04_d_kernels_attention.py . Like before, it helps to open them in a separate tab and walk through the code as you read. We use an NVIDIA A100-SXM4-80GB GPU to run the scripts. It is really easy to set up a GPU on the Hugging Face infrastructure and experiment with the scripts using Dev Mode with Spaces . One could also run the scripts with the Hugging Face Jobs pipeline .

Attention works with Queries ( q ), Keys ( k ), and Values ( v ). The interaction between them can be written as a short sequence of steps:

So attention is really a collection of primitive operations. Some of them we already know (the matmuls), and the rest are easy to spot. Let's write a naive attention module in PyTorch and profile it.