BerriAI/litellm v1.17.0

on GitHub

Major Changes

• LiteLLM Proxy now uses Gunicorn by default
  If you use LiteLLM Dockerfile or images no changes required. If you're using litellm pip package just run pip install 'litellm[proxy]'-U
• Support for litellm.ContentPolicyViolationError catch these errors from image generation models

LiteLLM Proxy Dockerfiles

• Dockerfile.Database has litellm as the entrypoint

  litellm/Dockerfile.database

  Line 59 in b103ca3

  CMD ["--port", "4000"]

  cc @Manouchehri (you can pass litellm cli args)
• Use https://github.com/BerriAI/litellm/pkgs/container/litellm for calling LLM APIs (without Virtual keys)
• Use https://github.com/BerriAI/litellm/pkgs/container/litellm-database for calling LLM APIs + Virtual Keys (this build has optimized cold boot for using Prisma (the DB Provider))

What's Changed

• [Feat] Add litellm.ContentPolicyViolationError by @ishaan-jaff in #1381
• (fix): Self-serve UI, AUTH link generation error by @speedyankur in #1385
• (fix): Self-serve UI, AUTH link generation error by @speedyankur in #1386
• (fix): Self-serve UI, AUTH link generation error by @speedyankur in #1391
• (caching) Fix incorrect usage of str, which created invalid JSON. by @Manouchehri in #1390
• Litellm dockerfile testing by @krrishdholakia in #1402
• fix(lowest_latency.py): add back tpm/rpm checks, configurable time window support, improved latency tracking by @krrishdholakia in #1403
• LiteLLM Proxy - Use Gunicorn with Uvicorn workers by @ishaan-jaff in #1399

Full Changelog: v1.16.21...v1.17.0