mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-09 22:31:41 +00:00
1d55ef8684
The dashboard suite ran every tier in the forks pool, which pays a child process plus a fresh module graph per test file. Worker threads reuse the process and cut the wall clock by about 20% on the same hardware, with no change to how the tiers are isolated: vitest still gives each file its own environment, so the per-file globals, module registry and jsdom document stay exactly as they were. Measured on the 743-file suite: 86s, 88s and 88s on forks against 71s, 66s and 70s on threads, all 8,588 tests passing in every run. The pool now lives in vitest.config.ts so local runs match CI, and the workflow keeps pinning the worker count for the 16-core runner. |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||