trunk/f90c091c44cf4e8feffbf5d5afdebd20798a86fa: Update inductor expected accuracy files (#175041)
Core framework changes signal major performance tuning for AI developers...
Deep Dive
The PyTorch team has updated the expected accuracy CSV files for its Inductor compiler benchmarks, based on CI results from commit 93dd7743c6577271a81f2fef0fdeafc5fe06e553. These files in the `benchmarks/dynamo/ci_expected_accuracy/` directories serve as critical reference points for tracking graph breaks and model accuracy across both CUDA and ROCm platforms, ensuring the 97.4k-star framework maintains performance standards.
Why It Matters
These updates are essential for developers relying on PyTorch's Inductor for stable, high-performance model compilation and deployment.