* fix(router.py): add acompletion_streaming_iterator inside router
allows router to catch errors mid-stream for fallbacks
Work for https://github.com/BerriAI/litellm/issues/6532
* fix(router.py): working mid-stream fallbacks
* fix(router.py): more iterations
* fix(router.py): working mid-stream fallbacks with fallbacks set on router
* fix(router.py): pass prior content back in new request as assistant prefix message
* fix(router.py): add a system prompt to help guide non-prefix supporting models to use the continued text correctly
* fix(common_utils.py): support converting `prefix: true` for non-prefix supporting models
* fix: reduce LOC in function
* test(test_router.py): add unit tests for new function
* test: add basic unit test
* fix(router.py): ensure return type of fallback stream is compatible with CustomStreamWrapper
prevent client code from breaking
* fix: cleanup
* test: update test
* 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
* fix(vertex_and_google_ai_studio.py): ensure grounding metadata is preserved on streaming
Closes https://github.com/BerriAI/litellm/issues/10237
* fix(core_helpers.py): include usage in expected openai keys
* 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 19c6c62154fceb9c431fea3e448753495dec35d2.
* 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
* fix(streaming_handler.py): don't return stream options in clientside response
* test: add unit test
* fix(custom_guardrail.py): fix start time
don't start until function actually called
leading to incorrect high latency numbers
* fix(custom_guardrail.py): fix start time
* build(model_prices_and_context_window.json): remove 'supports_tool_choice' for specific mistral models
Closes https://github.com/BerriAI/litellm/issues/11750
* feat: initial commit adding cleaner ui for azure text moderation guardrails
* feat(guardrail_endpoints.py): add discoverable guardrail configs and improve converting base model to dict with types
* fix(guardrail_provider_fields.tsx): render from api endpoint correctly
* fix(guardrail_provider_fields.tsx): cleanup
* refactor(guardrail_endpoints.py): refactor to handle dictionaries with literal - allows multiselect
* feat(ui/): render dictionary with known keys correctly
* feat(ui/): render optional params on separate page
* style(ui/): style improvements to rendering optional params on the UI
* feat(azure/prompt_shield.py): add azure prompt shield back on UI
* fix(add_guardrail_form.tsx): fix form to handle updated api
* fix(guardrail_optional_params.tsx): ensure values are nested correctly for writing to api
* fix: fix linting error
* feat(text_moderation.py): handle str to int conversion
* fix(guardrail_info.tsx): only render pii settings if guardrail is presidio
* fix(guardrail_info.tsx): add guardrail specific fields to update settings
allows updating guardrail fields (e.g. severity threshold) post-create
* fix(guardrail_endpoints.py): set guardrail_id in guardrail object
ensures duplicate objects not created on guardrail update
* fix(guardrail_endpoints.py): allow provider specific fields to be updated on patch update
* refactor(guardrail_endpoints.py): remove duplicate info endpoint
* fix(guardrail_endpoints.py): mask sensitive keys on returning via guardrail `/info`
Prevent leaking keys
* fix(guardrail_optional_params.tsx): return numerical input when numerical component used
fixes issue where output was a str
* fix(guardrail_optional_params.tsx): render dict keys correctly
* fix(text_moderation.py): fix severity by category check
* fix(proxy/utils.py): check if guardrail should run for post call streaming hook
Prevents invalid guardrails from running if not requested
* test: fix import
* fix: fix linting error
* test: update test
* fix: fix tests
* fix: fix code qa errors
* fix(guardrail_endpoints.py): set max depth for function
* test: update recursive_detector.py
* test: update list
* build: merge main
* fix: fix ruff check errors
* Litellm dev 03 05 2025 contributor prs (#9079)
* feat: add support for copilot provider
* test: add tests for github copilot
* chore: clean up github copilot authenticator
* test: add test for github copilot authenticator
* test: add test for github copilot for sonnet 3.7 thought model
* Fix#7629 - Add tzdata package to Dockerfile (#8915)
* Add tzdata package to Dockerfile
* Move tzdata to python requirement.txt
* feat: add support for copilot provider (#8577)
* feat: add support for copilot provider
* test: add tests for github copilot
* chore: clean up github copilot authenticator
* test: add test for github copilot authenticator
* test: add test for github copilot for sonnet 3.7 thought model
---------
Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
* feat: add model information for copilot models
* fix: fix linting errors
* test: remove integration test for github_copilot + fix misisng mock
* fix: use print to make sure the logger message shown
* test: remove debug print
* fix lint (#11112)
* Add init files to make test directories Python packages and update import paths in test_token_counter.py (#11119)
* Update litellm/model_prices_and_context_window_backup.json
Co-authored-by: மனோஜ்குமார் பழனிச்சாமி <smartmanoj42857@gmail.com>
---------
Co-authored-by: Son H. Nguyen <nhs.000.dev@gmail.com>
Co-authored-by: subnet.dev <50828879+subnet-dev@users.noreply.github.com>
Co-authored-by: Son H. Nguyen <33925625+nhs000@users.noreply.github.com>
Co-authored-by: மனோஜ்குமார் பழனிச்சாமி <smartmanoj42857@gmail.com>
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* refactor github copilot
* test_github_copilot_transformation.py
* test_github_copilot_authenticator.py
* add GitHub Copilot
* fix order
* doc fix
---------
Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
Co-authored-by: Son H. Nguyen <nhs.000.dev@gmail.com>
Co-authored-by: subnet.dev <50828879+subnet-dev@users.noreply.github.com>
Co-authored-by: Son H. Nguyen <33925625+nhs000@users.noreply.github.com>
Co-authored-by: மனோஜ்குமார் பழனிச்சாமி <smartmanoj42857@gmail.com>
When JSON_LOGS=True is set, error logs were not being formatted as JSON despite
the configuration. This was because the logging initialization code configured
individual loggers but failed to properly initialize all loggers with the JSON
formatter.
This fix ensures that when json_logs is enabled, the _initialize_loggers_with_handler()
function is called to:
- Configure all loggers (root, LiteLLM, Router, Proxy) with JSON formatter
- Disable logger propagation to prevent duplicate entries
- Set up exception handlers for JSON formatting
Fixes LIT-267
* Allow strings in calculate cost
Sometimes the cost per unit is a string (e.g.: If a value like "3e-7" was read from the config.yaml)
* Add comprehensive tests for string cost value handling
- Added test_string_cost_values() to test basic string cost conversion functionality
- Added test_calculate_cost_component_with_string_values() to test the calculate_cost_component function directly
- Added test_string_cost_values_edge_cases() to test mixed string/float costs and error handling
- Added test_string_cost_values_with_threshold() to test string costs with threshold pricing
- Enhanced _get_token_base_cost() to handle string-to-float conversion for base costs and threshold costs
- Enhanced generic_cost_per_token() to handle string-to-float conversion for audio and reasoning token costs
- All tests cover scientific notation (e.g., '3e-7'), decimal notation (e.g., '0.000001'), and error handling for invalid strings
- Maintains backward compatibility with existing float cost values
* Dry up code
* Fixed case where number was an integer
* Allowing None
---------
Co-authored-by: openhands <openhands@all-hands.dev>
* fix(rebuild-usage-object---ensure-cache_tokens-is-set): Ensures cache tokens is correctly set
Fixes https://github.com/BerriAI/litellm/issues/12149
* test(test_stream_chunk_builder_utils.py): add unit test to ensure cached tokens is part of stream chunk builder
Ensures standardized values are used
* fix(anthropic/experimental_pass_through): use given model name when returning streaming chunks
don't harcode model name on streaming
confusing for user
* fix(anthropic/streaming_iterator.py): remove scope of import
* feat(litellm_logging.py): allow admin to specify additional headers for using as spend tags
Closes https://github.com/BerriAI/litellm/issues/12129
* test(test_litellm_logging.py): add unit tests
* feat(openweb_ui.md): add custom tag tutorial to docs
* docs(cost_tracking.md): add tag based usage UI screenshot
* test: update test
* fix: fix import
* Add Azure OpenAI assistant features cost tracking
Implements cost tracking for Azure's new assistant features:
- File Search: $0.1 USD per 1 GB/Day (storage-based pricing)
- Code Interpreter: $0.03 USD per session
- Computer Use: $0.003 input + $0.012 output per 1K tokens
Features:
- Provider-specific pricing (Azure vs OpenAI)
- Model-specific pricing overrides via JSON config
- Environment variable configuration
- Backwards compatible with existing OpenAI pricing
* Add comprehensive tests for Azure assistant features cost tracking
- Unit tests for file search, code interpreter, computer use, vector store
- Integration tests for combined cost calculation
- Provider-specific pricing tests (Azure vs OpenAI)
- Model-specific pricing override tests
- Edge case handling (None inputs, zero values)
- All 17 tests passing
* Fix test and ensure all Azure assistant cost tracking tests pass
- Fixed integration test approach
- All 17 tests now passing
- Comprehensive coverage of Azure assistant features cost tracking
* Enhance cost tracking for Azure assistant features
- Safely convert and extract parameters for file search, computer use, and code interpreter sessions.
- Ensure model_info is consistently converted to a dictionary format.
- Improve error handling for input values to prevent type-related issues.
- Maintain compatibility with existing cost calculation methods.
* Refactor cost tracking for Azure assistant features
- Introduced separate methods for handling costs related to web search, file search, vector store, computer use, and code interpreter.
- Enhanced parameter extraction and conversion for file search and computer use.
- Improved error handling and type safety throughout the cost calculation process.
- Maintained compatibility with existing cost calculation methods while streamlining the overall structure.
* refactor(unpack_defs): enhance handling of schema properties and anyOf structures
- Improved the unpack_defs function to handle top-level properties and nested structures more effectively.
- Added recursion for items in schemas and refined the handling of anyOf branches to ensure proper unpacking of references.
- Streamlined the logic for resolving $ref keys and managing nested schemas.
* test(unpack_defs): add test for resolving nested $ref in anyOf items
- Introduced a new test to verify that unpack_defs correctly resolves references within items of anyOf structures, addressing a specific bug scenario (Issue #11372).
- The test includes a minimal schema to ensure proper unpacking and validation of the resolved items schema.
* refactor(unpack_defs): implement a generic resolver for $ref entries
- Redesigned the unpack_defs function to provide a more robust and dependency-free implementation for resolving all $ref entries in JSON schemas.
- Introduced a depth-first traversal method that efficiently handles nested structures, including anyOf, allOf, and items, while avoiding infinite recursion.
- Enhanced memory management by resolving nodes in-place without creating a full dereferenced copy, improving performance and reducing overhead.
* Remove test for unpack_defs resolving nested references in anyOf items from test_utils.py
* Add test for unpack_defs resolving nested references in anyOf items
This commit introduces a new test to ensure that the unpack_defs function correctly resolves $ref references within items of anyOf schemas, addressing issue #11372. The test verifies that the unpacked schema contains the expected properties and structure.
* fix(litellm_logging.py): fix using router model id for logging calls
Fixes https://github.com/BerriAI/litellm/issues/11975#issuecomment-2995882238
* test(test_litellm_logging.py): add unit test for custom price tracking
* fix(vertex_ai/): don't send invalid format parameter to vertex
causes calls to fail
* fix(vertex_ai_context_caching.py): if cached content present and tools in message, cache tools as well
gemini throws errors if tools passed in alongside cached content
* test: add unit tests
* fix: fix linting errors
* test: test_vertex_ai_common_utils.py
update test
* fix(streaming_handler.py): unset response cost when creating model response
* fix(factory.py): handle qs:.. in mime type
Fixes https://github.com/BerriAI/litellm/issues/11839
* feat(litellm_proxy/): don't transform messages client-side
leave litellm proxy messages untouched - allow proxy to handle transformation
prevents double transformation
* feat(tag_management_endpoints.py): support adding models to tag by adding model_name
Closes https://github.com/BerriAI/litellm/issues/11884
* test(test_tag_management_endpoints.py): add unit tests for adding new model by public model name
* test: update test
* fix(spend_tracking_utils.py): add user agent tags from standard logging payload, in spend logs payload
* feat(litellm_logging.py): identify user agent tags as `User-Agent: ..` and allow admin to disable storing user agent as tag
* fix(azure_ai/): pass content type header in azure ai request
Fixes https://github.com/BerriAI/litellm/issues/11227
* test: add unit test
* fix(router.py): fix passing dynamic credentials to retrieve batch
Fixes batch retrieval when using router
* test: add more unit tests
* fix(prometheus.py): remove request increment from inside the log success event
it's only done on post-call success/failure
* fix(litellm_logging.py): add additional validation step for checking if 'stream' is true
prevent double counting on non-stream requests
* test: add unit testing to ensure stream is not incorrectly set to true
* feat(litellm_logging.py): emit request route in standard logging payload
used by prometheus streaming metrics for route
* fix: fix otel test
* fix: fix linting errors
* test: update test
* fix: fix linting error
* feat(anthropic/passthrough): pass dynamic api key/api base params to litellm.completion
allows calls to work with config.yaml
* fix(responses_api/transformation): fix passing dynamic params to responses api from .completion()
Allows responses api to work with config.yaml
* fix(langfuse.py): fix responses api usage logging to langfuse
* refactor(litellm_logging.py): add more generic solution for responses api usage logging
ensures it works across all logging integrations
* fix(litellm_logging.py): patch for anthropic messages not returning a pydantic object
it should ideally return a pydantic object, which would simplify checks and reduce errors
* fix(handler.py): correctly bubble up empty choices errors to litellm.completion
causes downstream errors as it is expected there is at least one choice set
* feat(litellm_logging.py): prevent double logging litellm responses
ensures accurate spend tracking for calls when bridges are used
* fix(litellm_logging.py): ensure logging is consistently enforced across all call types
* fix: patch - set calltype before entering bridge api
ensures logging object is applying the correct logic on the event hooks
* fix(types/router.py): loosen type hint for mock response
* change space_key header to space_id for Arize (#11595)
* feat(schema): add additional indexes to LiteLLM_SpendLogs for improved query performance (#11675)
* Revert "feat(schema): add additional indexes to LiteLLM_SpendLogs for improve…" (#11683)
This reverts commit 2a7f113fde.
* [Feat] Use dedicated Rest endpoints for list, calling MCP tools (#11684)
* fix: (fix) use specific rest endpoints for MCP
* ui - use rest mcp endpoints
* fix imports
* docs DISABLE_AIOHTTP_TRUST_ENV
* docs(caching.md): remove batch redis get recommendation - old code path, no longer necessary
* fix(vertex_and_google_ai_studio_gemini.py): handle gemini not passing audio token usage data
* Chat Completions <-> Responses API Bridge Improvements (#11685)
* feat(anthropic/passthrough): pass dynamic api key/api base params to litellm.completion
allows calls to work with config.yaml
* fix(responses_api/transformation): fix passing dynamic params to responses api from .completion()
Allows responses api to work with config.yaml
* fix(langfuse.py): fix responses api usage logging to langfuse
* refactor(litellm_logging.py): add more generic solution for responses api usage logging
ensures it works across all logging integrations
* fix(litellm_logging.py): patch for anthropic messages not returning a pydantic object
it should ideally return a pydantic object, which would simplify checks and reduce errors
* fix(handler.py): correctly bubble up empty choices errors to litellm.completion
causes downstream errors as it is expected there is at least one choice set
* fix(response_metadata.py): allow model_info to be none
* fix(litellm_logging.py): copy object before mutating
* fix: fix lint check
* fix: fix linting error
* fix: fix linting error
---------
Co-authored-by: vanities <mischkeaa@gmail.com>
Co-authored-by: Cole McIntosh <82463175+colesmcintosh@users.noreply.github.com>
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* fix(internal_user_endpoints.py): support user with `+` in email on user info
ensures user is correctly parsed from input
* fix(factory.py): support vertex function call args as None
handles empty string in args for vertex gemini calls
* docs(langfuse_integration.md): pin langfuse sdk version on docs
* fix(vertex_ai/): return empty dict, instead of none when empty string given
* refactor: reduce function size
* fix: fix linting errors
* fix: revert check
* fix(internal_user_endpoints.py): fix check
* test: update tests
* test: update tests
* fix(streaming_handler.py): maintain same 'created' across all chunks
Fixes https://github.com/BerriAI/litellm/issues/11437
* test: add unit test to ensure created is always the same across all chunks
* fix(types/utils.py): set a tool call id, if missing in delta tool call
Ensures stream chunk builder can reconstruct tool calls correctly
Fixes https://github.com/BerriAI/litellm/issues/11262
* fix(responses/transformation.py): support passing mcp server tool call to anthropic
allows switching between openai and anthropic for mcp tool calling
* fix(ollama/chat/transformation.py): set tool call id's when missing
* fix(vertex_and_google_ai_studio_gemini.py): add web search request tracking
Enables cost calculation for google web search
* fix(vertex_and_gemini): use common processing logic across stream / non-stream calls
* fix(vertex_And_google_ai_studio_Gemini.py): fix initial choice
* fix: fix linting error
* fix: add initial support for google search cost tracking
* fix(tool_call_cost_tracking.py): working tool cost tracking for gemini
* fix(vertex_ai/gemini/cost_calculator.py): add google web search tool cost tracking for vertex ai
Closes LIT-210
* fix: fix check
* build(model_prices_and_context_window.json): fix amazon nova max output tokens
Closes https://github.com/BerriAI/litellm/issues/11441
* fix: fix ruff check
* feat: add video support for bedrock converse api (#11043)
* fixes: bedrock add video support
* fixes: bedrock add video support
---------
Co-authored-by: yytdfc <fuchen@foxmail.com>
Krrish Dholakia