mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-16 23:41:43 +00:00
1c89dec3fe
The Caching page's cache-hit and cached-token views only reflect LiteLLM's own response cache (LiteLLM_SpendLogs.cache_hit). Provider prompt caching (e.g. Anthropic cache_read_input_tokens) never sets that flag, so those tokens were invisible and the overall numbers did not add up for anthropic_messages traffic. Add a /global/activity/cache_hits/prompt_caching endpoint that reads cache_read_input_tokens / cache_creation_input_tokens from the daily spend rollup (the only place they are persisted) and a Provider Prompt Caching section on the dashboard with read/creation token stats and a per-model chart, kept separate from the response-cache metrics. Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||