Needless text
April 4, 2025
Dynamic shapes are one of the more distinctive parts of torch.compile. Rather than specializing a graph to static shapes (which works in many cases!), PyTorch’s approach allows a single graph to work for a variety of sizes, so things like sequence length or batch size can vary. It does this by reasoning about shapes symbolically: instead of using fixed shape values, it uses placeholders and infers rules that constrain those shapes.
April 3, 2025
April 1, 2025
March 29, 2025
When data is in the global memory on a GPU it’s usually in row-major or column-major order. Loading from global memory is quite slow though, so for performance we want to move the data to shared memory for the threads in a warp to work on.
March 28, 2025
March 27, 2025
March 23, 2025
Last year Keller Jordan at OpenAI beat some of the existing NanoGPT speedrun records thanks to some optimizer improvements. Towards the end of the year the work was formalized as the Muonoptimizer, and it’s making waves in a bunch of areas now
March 21, 2025
| [What if your worry problem is really a planning problem? | Psyche Ideas](https://psyche.co/ideas/what-if-your-worry-problem-is-really-a-planning-problem) |
March 20, 2025
A couple of conversations I was in last week around agentic system design reminding me of Brandon Smith’s excellent write libraries, not frameworks post
March 19, 2025