Write the reference before debugging the shader

Automated excerpt

The Flash Attention 2 path was easier to finish after the algorithm existed in plain JavaScript first. The forward pass used one workgroup per batch, head, and query tile. The backward pass recomputed attention from q, k, and a saved log-sum-exp value instead of reading a cached attention matrix. At the recorded Mega-class shape, removing that cached-matrix traffic avoided about 67 MB of global memory movement per layer per step. That separated algorithm errors from shader and integration errors.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.