贡献 / #1456

#1456 bench: Flash Next variant tool-eval campaign

closed · draft · @CC-David-CC · 0 评论 · 去 GitHub 看

BenchmarksModels & quants

说明

## Summary
Draft benchmark campaign for ISTA and Swift 1.5 Q2_0/IQ2_XS. ISTA Q2_0: 80/100 short, 83/100 standard. Q8 is running on three P4s.

## What changed
Progress matrix with blank pending scores, raw traces, resource samples, exact settings and pinned model hashes.

## Extra Notes
Please create as a draft. No engine changes. Required variants are still being tested. Different hardware and background transfers are documented. Benchmark integration fix is a separate PR.

本站相关内容

相关页面的快捷入口。