mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-03 07:23:33 +00:00
The cooldown store no longer takes the v2 selector's DualCache and raw keys. The selector prefetch moves next to its caller in router_utils, and the store surface is model ids only, so a Rust-backed store never mixes a redis-py read into its own call The Rust tier policy now comes from the host. main reads the cooldown refresh interval and the batch throttle size from the environment and the usage refresh from litellm.default_redis_batch_cache_expiry, so a new router_tiers accessor in litellm/rust_bridge/host/settings.py reports them and both pyclasses build their store with those values. POLICY in redis-store stays as main's defaults Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|---|---|---|
| .. | ||
| .agents/skills | ||
| crates | ||
| docs/adrs | ||
| .gitignore | ||
| AGENTS.md | ||
| Cargo.lock | ||
| Cargo.toml | ||
| clippy.toml | ||