Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[tune/release] Use STRICT_PACK in air_benchmark_tune_torch_mnist (ray…
…-project#29133) We are using STRICT_PACK here to do an apples to apples comparison. PyTorch defaults to using multithreading, so if the workers are spread, they are able to utilize more resources. We would effectively be comparing X tune runs with 2 CPUs per worker vs. 1 tune run with up to 8 CPUs per worker. Using STRICT_PACK avoids this by forcing all workers to be co-located. Signed-off-by: Kai Fricke <[email protected]>
- Loading branch information