Tag

transformers.

2 writings found

Latest Archives

How a 66MB Test Model Caught a Two-Month Bug in Transformers

Model Integration Contracts let small fixture models catch regressions on CPU that full integration tests miss. Here's how Transformers uses them to ship faster.

Profiling Attention in PyTorch: From Naive to FlashAttention

A deep dive into profiling attention mechanisms in PyTorch, comparing naive, SDPA math, efficient, and FlashAttention backends using GPU traces.
View all rollups →