mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-07 02:59:05 +00:00
Lowers the shipped simple_medium, medium_complex and complex_reasoning defaults so the heuristic scorer stops parking technical, multi-part prompts in the cheapest tiers. On the in-repo labelled eval set this moves 5 of 29 cases up a tier and takes accuracy from 28/29 to 29/29. Also points _effective_tier_boundaries() at DEFAULT_TIER_BOUNDARIES instead of restating the three numbers, so a config that overrides only one boundary no longer fills the other two from a second, now stale copy of the defaults. |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||