* fix(lowest_latency.py): Handle ZeroDivisionError with zero completion tokens (#12641)
Fixes ZeroDivisionError when LLM responses have zero completion tokens, which can
occur with Gemini models on very long contexts that only use tool calls.
Changes:
- Add check for completion_tokens > 0 before division in log_success_event
- Handle both timedelta and float types for response times (supporting both time.time() and datetime)
- Apply fix to both occurrences of the division operation in the file
- Add comprehensive tests for zero completion token scenarios
This ensures the lowest latency routing continues to work properly even when
models return responses with no completion tokens.
* test: Move lowest_latency zero tokens test to test_litellm for CI execution
* refactor: Use safe_divide helper to eliminate code duplication
- Added safe_divide utility function to litellm_core_utils.core_helpers
- Handles both timedelta and float types for numerator
- Prevents ZeroDivisionError and negative denominator issues
- Replaced duplicated division logic in lowest_latency.py
- Added comprehensive unit tests for safe_divide function
This improves code quality by reducing duplication and centralizing the
division safety logic in a reusable helper function.
* refactor: Rename to safe_divide_seconds for clarity
- Renamed safe_divide to safe_divide_seconds to better indicate it handles time durations
- Simplified implementation with single-line ternary for seconds conversion
- Updated all references and tests accordingly
The function name now clearly indicates it's specifically for dividing
time durations (in seconds) by a denominator.
* refactor: Simplify safe_divide_seconds to only accept float arguments
- Changed safe_divide_seconds to accept only float parameters for consistency
- Callers now handle timedelta to seconds conversion explicitly
- Removed timedelta-specific test cases
* fix: Remove unused timedelta import
* fix: Remove Union[timedelta, float] type annotations
Since safe_divide_seconds now only accepts floats, we handle the
timedelta conversion explicitly in the code. The type annotations
are no longer needed and can be simplified.
* Revert "fix: Remove Union[timedelta, float] type annotations"
This reverts commit 19c6c62154.
* fix: Clean up implementation
- Remove unnecessary type annotations
- Use safe_divide_seconds utility for zero-division protection
- Handle both timedelta and float types explicitly
- Maintain compatibility with both datetime and time.time() usage
* feat: integrate Google Cloud Model Armor guardrails (LIT-298)
- Add ModelArmorGuardrail class that extends CustomGuardrail and VertexBase
- Support for both pre-call (sanitizeUserPrompt) and post-call (sanitizeModelResponse) sanitization
- Integrate with existing Vertex AI authentication using VertexBase
- Add configuration model for Model Armor in guardrail types
- Register Model Armor in guardrail initializers and registry
- Include comprehensive test suite for Model Armor functionality
- Support content masking for both requests and responses
- Handle streaming responses with content sanitization
This integration allows LiteLLM to use Google Cloud Model Armor API for
content moderation and sanitization, providing similar functionality to
Bedrock Guards but using Google Cloud's security infrastructure.
* fix: remove unused imports flagged by ruff linter
- Remove unused asyncio import at top level (moved to local import where needed)
- Remove unused TextCompletionResponse import
* fix: remove additional unused imports
- Remove unused TextChoices import from line 34
- Remove duplicate asyncio import from line 384
- Replace asyncio.iscoroutine() with hasattr check for __await__
* fix: remove final unused imports
- Remove unused 'import sys' from line 9
- Remove unused 'StreamingChoices' from imports
* fix: remove unused import from model_armor.py
- Remove unused 'import os' from the top of the file
* fix: remove commented-out header from model_armor.py
- Eliminate unnecessary comments at the top of the file to improve code clarity.
* feat(guardrails): Add Model Armor UI support
- Convert model_armor.py to a directory structure with __init__.py for dynamic discovery
- Add get_config_model() method to ModelArmorGuardrail class for UI integration
- Add ui_friendly_name() to ModelArmorConfigModel returning "Google Cloud Model Armor"
- Remove manual registration from guardrail_registry.py to use dynamic discovery
- Model Armor now appears in the UI guardrails dropdown with proper configuration fields
This enables users to configure Model Armor guardrails through the LiteLLM UI interface.
* fix(guardrails): Fix undefined name 'GuardrailConfigModel' in Model Armor
- Import TYPE_CHECKING and GuardrailConfigModel from base module
- Fixes F821 linting error for undefined name in type annotation
- Follows same pattern as other guardrail implementations
* fix(guardrails): Fix Model Armor type errors and config model inheritance
- Create ModelArmorGuardrailConfigModel that properly inherits from GuardrailConfigModel base class
- Move config model to litellm/types/proxy/guardrails/guardrail_hooks/model_armor.py following convention
- Update get_config_model() to return the properly typed config model
- Remove ModelArmorConfigModel from LitellmParams inheritance chain
- Add template_id field to BaseLitellmParams instead
This fixes the mypy type errors and follows the same pattern as other guardrail implementations.
* fix(guardrails): Add missing Model Armor fields to BaseLitellmParams
- Add location, credentials, api_endpoint, and fail_on_error fields
- Fixes mypy errors about missing attributes in LitellmParams
- All Model Armor configuration parameters are now properly defined
* fix: Apply PR review feedback for Model Armor guardrail
- Move test file to tests/test_litellm/ for GitHub Actions
- Extract only last consecutive user messages to avoid context limits
- Use get_content_from_model_response helper for response extraction
- Handle non-ModelResponse types (e.g., TTS) gracefully
- Maintain newline separation for multi-part content
* refactor: Simplify message content extraction in ModelArmorGuardrail
- Removed the custom _extract_content_from_messages method.
- Integrated get_last_user_message helper for improved content extraction.
- Updated test to reflect changes in content formatting.
* refactor: Remove unused import in model_armor.py
- Deleted the unused import of AllMessageValues to clean up the codebase.
* Add unit tests for Model Armor guardrail functionality
- Implement tests for pre-call and post-call hooks, including content sanitization and blocking behavior.
- Validate error handling for API responses and credential management.
- Test streaming responses and handling of list content in user messages.
- Ensure proper assertions for API interactions and response sanitization.
* Add comprehensive test coverage for Model Armor guardrail
- Add test for requests with no messages field
- Add test for empty message content handling
- Add test for system/assistant-only messages
- Add test for fail_on_error=False behavior
- Add test for custom API endpoint configuration
- Add test for dictionary credentials (non-file path)
- Add test for action=NONE response handling
- Add test for missing sanitized_text field fallback
- Add test for non-text response types (TTS/image)
- Add test for auth token refresh behavior
All tests ensure robust edge case handling and proper error management.
* feat: improve Model Armor handling of non-ModelResponse types
- Add debug logging when skipping non-text responses (TTS, images, etc.)
- Improve docstring to clarify behavior for non-text responses
- Add test coverage for non-ModelResponse handling
- Ensure guardrail gracefully skips processing for response types it cannot handle
* build: move build_and_test to use prisma migrate
* fix(guardrails_ai.py): default to guardrail accepting 'llmOutput' as the input param
enables same guardrail to work for pre call and post call
* fix(__init__.py): set default value
* fix(guardrails_ai.py): updates
* fix: fix linting error
* fix(prompt_templates/factory.py): handle anthropic cache control on individual tool results
Fixes issue where cache control on individual tool result was being ignored
* test(test_vertex_And_google_ai_studio_gemini.py): initial unit test covering translation for grounding metadata on streaming chunk
* Vllm rerank (#12737)
* Add Hosted VLLM rerank provider integration
This commit implements the Hosted VLLM rerank provider integration for LiteLLM. The integration includes:
Adding Hosted VLLM as a supported rerank provider in the main rerank function
Implementing the HostedVLLMRerank handler class for making API requests
Creating a transformation class to convert Hosted VLLM responses to LiteLLM's standardized format
The integration supports both synchronous and asynchronous rerank operations. API credentials can be provided directly or through environment variables (HOSTED_VLLM_API_KEY and HOSTED_VLLM_API_BASE).
Notable features:
Proper error handling for missing credentials
Standard response transformation
Support for common rerank parameters (top_n, return_documents, etc.)
Proper token usage tracking
This expands LiteLLM's rerank provider ecosystem to include Hosted VLLM alongside existing providers like Cohere, Together AI, Azure AI, and Bedrock.
* refactor(rerank): use base_llm_http_handler for hosted_vllm rerank
- Replace custom HostedVLLMRerank handler with base_llm_http_handler
- Implement proper HostedVLLMRerankConfig inheriting from BaseRerankConfig
- Follow Cohere-compatible implementation pattern
- Clean up unnecessary comments
* Fix lint errors in hosted_vllm rerank transformer: remove unused imports
* Fix linting errors in rerank transformation modules
* fix: resolve type errors in Hosted VLLM rerank module
---------
Co-authored-by: Philip D'Souza <philip.dsouza@macro4.com>
Co-authored-by: Philip D'Souza <philip.a.dsouza@gmail.com>
* added a few tests
---------
Co-authored-by: Philip D'Souza <philip.dsouza@macro4.com>
Co-authored-by: Philip D'Souza <philip.a.dsouza@gmail.com>
Replace MagicMock with AsyncMock for litellm_teamtable.update to fix:
TypeError: object MagicMock can't be used in 'await' expression
The test was failing because it tried to await a MagicMock object.
Added AsyncMock for the update method to properly handle async operations.
* fix(team_endpoints.py): ensure user id correctly added when new team created with user email as member
Fixes issue where user not correctly added to team on /team/new
* fix(internal_user_endpoints.py): make user email validation check case insensitive
Fixes issue where uppercase email was added even when lowercase email existed
* test: update test
* build: move build_and_test to use prisma migrate
* feat(proxy_setting_endpoints.py): encrypt env var before storing in db
Ensures env var can be read when loaded in from DB
Fixes issue when trying to add SSO from admin UI
* test: update tests
* fix: Handle circular references in spend tracking metadata JSON serialization
- Fixes issue #12634 where circular references in metadata caused
ValueError: Circular reference detected when logging spend data
- Adds _safe_json_dumps() function that detects and handles circular
references by replacing them with placeholder strings
- Maintains full functionality for normal objects while preventing
crashes from circular references
- Adds comprehensive tests for circular reference handling
- Critical fix for v1.74.3 stable release
* fix: Replace bare except clauses with specific Exception handling
- Fixes E722 linting errors in _safe_json_dumps function
- Maintains same error handling behavior while following best practices
- All tests continue to pass
* refactor: Use existing safe_dumps utility instead of custom implementation
- Replace custom _safe_json_dumps() with existing safe_dumps() from litellm_core_utils
- Remove duplicate code and leverage existing circular reference handling
- Update tests to use safe_dumps function
- Maintains same functionality while reducing code duplication
- All tests continue to pass
* fix(router.py): use more descriptive error message
* fix(proxy/_types.py): note `/team/member_update` is a self-managed route
route has it's own logic for rbac - enables team admins to update member permissions
Fixes issue where team admins on UI could not update member permissions
* fix(token_counter.py): move log line to being '.debug' instead of '.error'
Fixes https://github.com/BerriAI/litellm/issues/12269
* fix(bedrock): auto-generate session name when only aws_role_name is provided
Fixes#12583 - AWS role assumption not working correctly when aws_role_name
is provided without aws_session_name.
Previously, if only aws_role_name was provided in the config without
aws_session_name, the code would fall back to using environment credentials
instead of assuming the specified role. This was problematic in EKS/IRSA
environments where users want to assume a different role.
The fix:
- When aws_role_name is provided without aws_session_name, we now
auto-generate a session name with format 'litellm-session-{timestamp}'
- This ensures role assumption happens as expected
- Added comprehensive test coverage for this scenario
* style: format test file with black
---------
Co-authored-by: openhands <openhands@all-hands.dev>
* fix(anthropic): fix streaming + response_format + tools bug
- Fix _handle_json_mode_chunk to only convert response_format tools to content
- Regular user tools now remain as proper tool_calls in streaming mode
- Add comprehensive test for the fix
- Resolves issue where all tools were incorrectly converted to content chunks
Before: All tools converted to content with different indices
After: Only response_format tool converted, regular tools remain as tool_calls
* fix(anthropic): improve streaming + response_format + tools handling
* fix: lint error (too many statements)
* fix(anthropic): correct finish_reason for streaming response_format tools
- Add ttl parameter to cache_control for Gemini models
- Support Google's TTL format (e.g., '3600s', '7200s')
- Implement robust TTL extraction and validation
- Extract TTL before system message transformation to handle all cases
- Add comprehensive test suite with 17 test cases in tests/test_litellm/
- Update documentation with TTL usage examples
- Maintain backward compatibility with existing cache_control usage
Fixes#9810
* feat(route_checks.py): allow admin to disable proxy management endpoints on instance
useful for preventing multiple instances from doing admin actions
* docs(scaling_multiple_instances.md): add architecture doc on scaling multiple litellm instances
provide guidance on scaling proxy
* docs(scaling_multiple_instances.md): add doc on scaling across multiple regions for litellm
* fix(route_checks.py): allow disabling llm api endpoints on an instance
allows pure admin instance to exist
* refactor(enterprise/route_checks.py): refactor env var checks
* refactor: finish refactoring
* docs(control_plane_and_data_plane.md): refactor docs
* test: update tests
Added cache_read_input_token_cost (25% of regular input cost) to all 39 Gemini 2.x models
to properly support implicit context caching cost calculations. Previously, cached tokens
were being charged at full price instead of the discounted rate.
Fixes#11156
* [Feat] New LLM API Integration - Add Moonshot API (Kimi) (#12551)
* Add Moonshot AI provider support to LiteLLM
Co-authored-by: ishaan <ishaan@berri.ai>
* Refactor Moonshot provider params handling and transformation logic
Co-authored-by: ishaan <ishaan@berri.ai>
* fix constants
* add Moonshot AI
* fix get_supported_openai_params
* handle kimi temp
* add tool choice handling
* test moonshot unit tests
* fix kimi
* fix linting
* Add pricing information for Moonshot AI's kimi-k2 model (#12566)
* Add pricing information for Moonshot AI's kimi-k2 model
* Update model name to kimi-k2-0711-preview
- Changed model name from moonshot/kimi-k2 to moonshot/kimi-k2-0711-preview
- This reflects the specific model version as requested
* Update moonshot_models list to match model_context JSON
---------
Co-authored-by: openhands <openhands@all-hands.dev>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: ishaan <ishaan@berri.ai>
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
Co-authored-by: openhands <openhands@all-hands.dev>
* update docs
* docs moonshot
* fixes model cost map
* fix map_openai_params
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: ishaan <ishaan@berri.ai>
Co-authored-by: Xingyao Wang <xingyao@all-hands.dev>
Co-authored-by: openhands <openhands@all-hands.dev>
* just use 1 param for mcp groups
* fix just use 1 param for access groups
* test_get_tools_from_mcp_servers
* docs access groups
* group MCPs
* test fix
* fix screenshots on docs
* TestMCPAccessGroupsE2E
* update img
* fix MCP connect
* Get the basics of the integration working.
* Cleanup bytez integration.
* Update user agent for Bytez integration.
* Use the config class directly. Create the start of the docs.
* Finish up bytez documentation. Include a provider integration guide.
* Fix typing bug in custom_logger_utils. Add tests for bytez integration.
* Add token tracking for model usage for Bytez integration.
* Create a units test for the Bytez config.
* Make changes to Bytez transformation code per PR feedback.
* Cleanup coment in Bytez transformation test.
* Remove LRU usage for bytez integration.
* Consolidate Bytez tests into a single file. Conform to project structure for tests.
* Fix linting error with Bytez impl.
* added mcp tools on internal user and divide it by teams
* add support for server api call
* Added frontend for test key
* added tools used output
* fix ui for servers
* All servers to personal
* change columns format
* revert ui logic
* Added vertical align
* fix mapped tests
* fix lint
* fix lint
* remove extra file
* fix ui test
* comments fixes
* change query type
* change query type
* mcp acces group init
* add ability to change server display on ui through access groups
* Mcp access group names UI (#12486)
* Added ui changes to reflect mcp_access_groups
* fix edit mcp page
* change to string array (#12491)
* change to string array
* Remove print
* add ability to change server display on ui through access groups
* Litellm mcp access groups accesses (#12498)
* added mcp access groups for keys and teams
* added access groups above servers
* fixed ruff
* fixed mypy
* revert couple changes
* fix test
* fixed double asterisks
* Litellm mcp groups UI (#12522)
* add ui for teams
* fix object permissions
* fix mcp servers test object permission
* remove print
* add helper method
* add tests + remove logs
* add mcp access group servers to test key
* add mcp access group support for headers
* lint fix
* add tests and helper function
* fixed test
* change list -> List
* tests
* fix(proxy_setting_endpoints.py): require store model in db is enabled for setting user default settings
* test(test_proxy_server.py): update test
* fix(reset_budget_job.py): initial commit adding reset budget logic for team members
* test: update unit testing
* test(test_proxy_budget_reset.py): validate team member budget was reset
* test(test_reset_budget_job.py): update unit tests
* test: update tests
* fix(internal_user_endpoints.py): initial commit removing logic to create new budget for new user if default max budget in team set
* feat(proxy_setting_endpoints.py): update team member budget when set via default internal user endpoint
removes need to create a unique budget per user
* feat(proxy_server.py): set team max member budget on startup, if set on config.yaml
* fix(prometheus.py): support custom tags for tracking on prometheus
Allows tracking user agent values on prometheus metrics
* test(test_internal_user_endpoints.py): fix test
* update types
* add STDIO to client
* allow using STDIO with MCP manager
* add Stdio-specific fields to schema.prisma
* fixes for MCP mgmt
* fix for adding stdio MCP to DB
* ui - allow adding stdio MCPs
* fix MCP server manager
* docs stdio MCP
* add_stdio_mcp.png
* new stdio tests
* allow adding MCPs through config.yaml
* fix tool test panel
* use TestMCPClient
* ui fixes for testing circle ci mcp
* added mcp tools on internal user and divide it by teams
* add support for server api call
* Added frontend for test key
* added tools used output
* fix ui for servers
* All servers to personal
* change columns format
* revert ui logic
* Added vertical align
* fix mapped tests
* fix lint
* fix lint
* remove extra file
* fix ui test
* comments fixes
* change query type
* change query type
* mcp acces group init
* add ability to change server display on ui through access groups
* Mcp access group names UI (#12486)
* Added ui changes to reflect mcp_access_groups
* fix edit mcp page
* change to string array (#12491)
* change to string array
* Remove print
* add ability to change server display on ui through access groups
* Litellm mcp access groups accesses (#12498)
* added mcp access groups for keys and teams
* added access groups above servers
* fixed ruff
* fixed mypy
* revert couple changes
* fix test
* fixed double asterisks
* Litellm mcp groups UI (#12522)
* add ui for teams
* fix object permissions
* fix mcp servers test object permission
* remove print
* add helper method
* add tests + remove logs
* added mcp tools on internal user and divide it by teams
* add support for server api call
* Added frontend for test key
* added tools used output
* fix ui for servers
* All servers to personal
* change columns format
* revert ui logic
* Added vertical align
* fix mapped tests
* fix lint
* fix lint
* remove extra file
* fix ui test
* comments fixes
* change query type
* change query type
* mcp acces group init
* add ability to change server display on ui through access groups
* Mcp access group names UI (#12486)
* Added ui changes to reflect mcp_access_groups
* fix edit mcp page
* change to string array (#12491)
* change to string array
* Remove print
* add ability to change server display on ui through access groups
* Litellm mcp access groups accesses (#12498)
* added mcp access groups for keys and teams
* added access groups above servers
* fixed ruff
* fixed mypy
* revert couple changes
* fix test
* fixed double asterisks
* fix(guardrails_ai.py): initial commit adding pre-call hook support for guardrails ai
enables running user input through guardrails ai - if set
* feat(guardrails_ai.py): working pre call guardrail
enables pii detection to work via guardrails ai
* feat(guardrails_ai.py): support logging hook
enables masking input via guardrails ai on logging integrations
* test(test_guardrails_ai.py): add unit test for new input processing function
* Add unit tests
* Fix tool call handling in Anthropic pass-through adapter
- Move tool message placement before user messages to maintain proper order
- Add safety check for empty function arguments in tool calls
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
