v1.87.4 - Guardrails & Logging Fixes
Deploy this version​
- Docker
- Pip
docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:1.87.4
pip install litellm==1.87.4
v1.87.4 is a patch release on top of v1.87.3. It backports guardrail correctness fixes (a single pre-call hook for model-level guardrails, no DB re-init on every poll, 400 instead of 500 when AIM blocks a request), caps Anthropic cache-control injection, and resolves several logging issues across Datadog batching, passthrough duplication, and Claude Code traces.
What's Changed​
- fix(datadog): split oversized batches on 413 instead of re-queueing forever - PR #29444
- fix: stop use_chat_completions_api flag from leaking into provider request body - PR #29447
- fix: duplicate Claude Code traces - PR #29311
- fix: passthrough endpoints duplicate logs - PR #29598
- fix(integrations): cap Anthropic cache_control injection at 4 blocks - PR #30480
- fix(guardrails): run pre_call hook once for model-level guardrails - PR #30543
- fix(guardrails): stop re-initializing DB guardrails on every poll - PR #30542
- fix(guardrails): return 400 not 500 when AIM blocks a request - PR #30573
Full Changelog​
https://github.com/BerriAI/litellm/compare/v1.87.3...v1.87.4