AI article
Running Captum on Cloud TPUs with TorchTPU: a work log
Community description: This is the log of a few days spent running Captum's attribution methods on Cloud TPUs through...
Dev.to | Nov 1, 2026 | Joinal Ahmed
Automated excerpt
Captum works on TorchTPU without modification. On Qwen3-0.6B, Layer Integrated Gradients with 64 steps takes 0.80 s in strict eager mode, 0.36 s with DeferAndFuse and 0.16 s under torch.compile(backend="tpu"), against 4 to 17 s on CPU. Learned: set torch.set_float32_matmul_precision("highest") before any attribution work on TPU.
Selected automatically from source text; not independently written or fact-checked. Read the original for full context.