Add Better Benchmark performance summary dashboard

[karthickai]

Summary

Adds a summary view for Better Benchmark results in PyTorch HUD.

• Shows overall model and kernel performance using geomean, median, mean, and comparable-record counts.
• Breaks results down by suite and mode.
• Shows the largest model and kernel gains/regressions, ranked by absolute latency impact.
• Displays comparison coverage and exclusion reasons.
• Links workflow IDs to their GitHub Actions runs.
• Uses lower-is-better green/red semantics with a neutral threshold for small changes.
• Adds a dedicated backend query path for Better Benchmark summary data.

The benchmark records are produced by pytorch/pytorch#193991

Test plan

• yarn tsc --noEmit
• Prettier check for changed files
• ESLint for changed files
• Better Benchmark summary Jest tests
• Validate parsing of kernel, projected-model, coverage, and exclusion records
• Validate suite/mode aggregation and mover ranking
• Validate against uploaded B200 benchmark output

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
torchci	Ready	Preview	Aug 21, 2026 9:04pm