* fix(utils.py): add 'disallowed_special' for token counting on .encode()
Fixes error when '<
endoftext
>' in string
* Revert "(fix) standard logging metadata + add unit testing (#6366)" (#6381)
This reverts commit 8359cb6fa9.
* add new 35 mode lcard (#6378)
* Add claude 3 5 sonnet 20241022 models for all provides (#6380)
* Add Claude 3.5 v2 on Amazon Bedrock and Vertex AI.
* added anthropic/claude-3-5-sonnet-20241022
* add new 35 mode lcard
---------
Co-authored-by: Paul Gauthier <paul@paulg.com>
Co-authored-by: lowjiansheng <15527690+lowjiansheng@users.noreply.github.com>
* test(skip-flaky-google-context-caching-test): google is not reliable. their sample code is also not working
* Fix metadata being overwritten in speech() (#6295)
* fix: adding missing redis cluster kwargs (#6318)
Co-authored-by: Ali Arian <ali.arian@breadfinancial.com>
* Add support for `max_completion_tokens` in Azure OpenAI (#6376)
Now that Azure supports `max_completion_tokens`, no need for special handling for this param and let it pass thru. More details: https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/models?tabs=python-secure#api-support
* build(model_prices_and_context_window.json): add voyage-finance-2 pricing
Closes https://github.com/BerriAI/litellm/issues/6371
* build(model_prices_and_context_window.json): fix llama3.1 pricing model name on map
Closes https://github.com/BerriAI/litellm/issues/6310
* feat(realtime_streaming.py): just log specific events
Closes https://github.com/BerriAI/litellm/issues/6267
* fix(utils.py): more robust checking if unmapped vertex anthropic model belongs to that family of models
Fixes https://github.com/BerriAI/litellm/issues/6383
* Fix Ollama stream handling for tool calls with None content (#6155)
* test(test_max_completions): update test now that azure supports 'max_completion_tokens'
* fix(handler.py): fix linting error
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
Co-authored-by: Low Jian Sheng <15527690+lowjiansheng@users.noreply.github.com>
Co-authored-by: David Manouchehri <david.manouchehri@ai.moda>
Co-authored-by: Paul Gauthier <paul@paulg.com>
Co-authored-by: John HU <hszqqq12@gmail.com>
Co-authored-by: Ali Arian <113945203+ali-arian@users.noreply.github.com>
Co-authored-by: Ali Arian <ali.arian@breadfinancial.com>
Co-authored-by: Anand Taralika <46954145+taralika@users.noreply.github.com>
Co-authored-by: Nolan Tremelling <34580718+NolanTrem@users.noreply.github.com>
* feat(proxy_cli.py): add new 'log_config' cli param
Allows passing logging.conf to uvicorn on startup
* docs(cli.md): add logging conf to uvicorn cli docs
* fix(get_llm_provider_logic.py): fix default api base for litellm_proxy
Fixes https://github.com/BerriAI/litellm/issues/6332
* feat(openai_like/embedding): Add support for jina ai embeddings
Closes https://github.com/BerriAI/litellm/issues/6337
* docs(deploy.md): update entrypoint.sh filepath post-refactor
Fixes outdated docs
* feat(prometheus.py): emit time_to_first_token metric on prometheus
Closes https://github.com/BerriAI/litellm/issues/6334
* fix(prometheus.py): only emit time to first token metric if stream is True
enables more accurate ttft usage
* test: handle vertex api instability
* fix(get_llm_provider_logic.py): fix import
* fix(openai.py): fix deepinfra default api base
* fix(anthropic/transformation.py): remove anthropic beta header (#6361)
* refactor(main.py): streaming_chunk_builder
use <100 lines of code
refactor each component into a separate function - easier to maintain + test
* fix(utils.py): handle choices being None
openai pydantic schema updated
* fix(main.py): fix linting error
* feat(streaming_chunk_builder_utils.py): update stream chunk builder to support rebuilding audio chunks from openai
* test(test_custom_callback_input.py): test message redaction works for audio output
* fix(streaming_chunk_builder_utils.py): return anthropic token usage info directly
* fix(stream_chunk_builder_utils.py): run validation check before entering chunk processor
* fix(main.py): fix import
* nvidia nim support embedding config
* add nvidia config in init
* nvidia nim embeddings
* docs nvidia nim embeddings
* docs embeddings on nvidia nim
* fix llm translation test
* feat(together_ai/completion): handle together ai completion calls
* fix: handle list of int / list of list of int for text completion calls
* fix(utils.py): check if base model in bedrock converse model list
Fixes https://github.com/BerriAI/litellm/issues/6003
* test(test_optional_params.py): add unit tests for bedrock optional param mapping
Fixes https://github.com/BerriAI/litellm/issues/6003
* feat(utils.py): enable passing dummy tool call for anthropic/bedrock calls if tool_use blocks exist
Fixes https://github.com/BerriAI/litellm/issues/5388
* fixed an issue with tool use of claude models with anthropic and bedrock (#6013)
* fix(utils.py): handle empty schema for anthropic/bedrock
Fixes https://github.com/BerriAI/litellm/issues/6012
* fix: fix linting errors
* fix: fix linting errors
* fix: fix linting errors
* fix(proxy_cli.py): fix import route for app + health checks path (#6026)
* (testing): Enable testing us.anthropic.claude-3-haiku-20240307-v1:0. (#6018)
* fix(proxy_cli.py): fix import route for app + health checks gettsburg.wav
Fixes https://github.com/BerriAI/litellm/issues/5999
---------
Co-authored-by: David Manouchehri <david.manouchehri@ai.moda>
---------
Co-authored-by: Ved Patwardhan <54766411+vedpatwardhan@users.noreply.github.com>
Co-authored-by: David Manouchehri <david.manouchehri@ai.moda>
* use vertex llm as base class for embeddings
* use correct vertex class in main.py
* set_headers in vertex llm base
* add types for vertex embedding requests
* add embedding handler for vertex
* use async mode for vertex embedding tests
* use vertexAI textEmbeddingConfig
* fix linting
* add sync and async mode testing for vertex ai embeddings
* LiteLLM Minor Fixes & Improvements (09/26/2024) (#5925)
* fix(litellm_logging.py): don't initialize prometheus_logger if non premium user
Prevents bad error messages in logs
Fixes https://github.com/BerriAI/litellm/issues/5897
* Add Support for Custom Providers in Vision and Function Call Utils (#5688)
* Add Support for Custom Providers in Vision and Function Call Utils Lookup
* Remove parallel function call due to missing model info param
* Add Unit Tests for Vision and Function Call Changes
* fix-#5920: set header value to string to fix "'int' object has no att… (#5922)
* LiteLLM Minor Fixes & Improvements (09/24/2024) (#5880)
* LiteLLM Minor Fixes & Improvements (09/23/2024) (#5842)
* feat(auth_utils.py): enable admin to allow client-side credentials to be passed
Makes it easier for devs to experiment with finetuned fireworks ai models
* feat(router.py): allow setting configurable_clientside_auth_params for a model
Closes https://github.com/BerriAI/litellm/issues/5843
* build(model_prices_and_context_window.json): fix anthropic claude-3-5-sonnet max output token limit
Fixes https://github.com/BerriAI/litellm/issues/5850
* fix(azure_ai/): support content list for azure ai
Fixes https://github.com/BerriAI/litellm/issues/4237
* fix(litellm_logging.py): always set saved_cache_cost
Set to 0 by default
* fix(fireworks_ai/cost_calculator.py): add fireworks ai default pricing
handles calling 405b+ size models
* fix(slack_alerting.py): fix error alerting for failed spend tracking
Fixes regression with slack alerting error monitoring
* fix(vertex_and_google_ai_studio_gemini.py): handle gemini no candidates in streaming chunk error
* docs(bedrock.md): add llama3-1 models
* test: fix tests
* fix(azure_ai/chat): fix transformation for azure ai calls
* feat(azure_ai/embed): Add azure ai embeddings support
Closes https://github.com/BerriAI/litellm/issues/5861
* fix(azure_ai/embed): enable async embedding
* feat(azure_ai/embed): support azure ai multimodal embeddings
* fix(azure_ai/embed): support async multi modal embeddings
* feat(together_ai/embed): support together ai embedding calls
* feat(rerank/main.py): log source documents for rerank endpoints to langfuse
improves rerank endpoint logging
* fix(langfuse.py): support logging `/audio/speech` input to langfuse
* test(test_embedding.py): fix test
* test(test_completion_cost.py): fix helper util
* fix-#5920: set header value to string to fix "'int' object has no attribute 'encode'"
---------
Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
* Revert "fix-#5920: set header value to string to fix "'int' object has no att…" (#5926)
This reverts commit a554ae2695.
* build(model_prices_and_context_window.json): add azure ai cohere rerank model pricing
Enables cost tracking for azure ai cohere rerank models
* fix(litellm_logging.py): fix debug log to be clearer
Closes https://github.com/BerriAI/litellm/issues/5909
* test(test_utils.py): fix test name
* fix(azure_ai/cost_calculator.py): support cost tracking for azure ai rerank models
* fix(azure_ai): fix azure ai base model cost tracking for rerank endpoints
* fix(converse_handler.py): support new llama 3-2 models
Fixes https://github.com/BerriAI/litellm/issues/5901
* fix(litellm_logging.py): ensure response is redacted for standard message logging
Fixes https://github.com/BerriAI/litellm/issues/5890#issuecomment-2378242360
* fix(cost_calculator.py): use 'get_model_info' for cohere rerank cost calculation
allows user to set custom cost for model
* fix(config.yml): fix docker hub auht
* build(config.yml): add docker auth to all tests
* fix(db/create_views.py): fix linting error
* fix(main.py): fix circular import
* fix(azure_ai/__init__.py): fix circular import
* fix(main.py): fix import
* fix: fix linting errors
* test: fix test
* fix(proxy_server.py): pass premium user value on startup
used for prometheus init
---------
Co-authored-by: Cole Murray <colemurray.cs@gmail.com>
Co-authored-by: bravomark <62681807+bravomark@users.noreply.github.com>
* handle streaming for azure ai studio error
* [Perf Proxy] parallel request limiter - use one cache update call (#5932)
* fix parallel request limiter - use one cache update call
* ci/cd run again
* run ci/cd again
* use docker username password
* fix config.yml
* fix config
* fix config
* fix config.yml
* ci/cd run again
* use correct typing for batch set cache
* fix async_set_cache_pipeline
* fix only check user id tpm / rpm limits when limits set
* fix test_openai_azure_embedding_with_oidc_and_cf
* test: fix test
* test(test_rerank.py): fix test
---------
Co-authored-by: Cole Murray <colemurray.cs@gmail.com>
Co-authored-by: bravomark <62681807+bravomark@users.noreply.github.com>
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* LiteLLM Minor Fixes & Improvements (09/23/2024) (#5842)
* feat(auth_utils.py): enable admin to allow client-side credentials to be passed
Makes it easier for devs to experiment with finetuned fireworks ai models
* feat(router.py): allow setting configurable_clientside_auth_params for a model
Closes https://github.com/BerriAI/litellm/issues/5843
* build(model_prices_and_context_window.json): fix anthropic claude-3-5-sonnet max output token limit
Fixes https://github.com/BerriAI/litellm/issues/5850
* fix(azure_ai/): support content list for azure ai
Fixes https://github.com/BerriAI/litellm/issues/4237
* fix(litellm_logging.py): always set saved_cache_cost
Set to 0 by default
* fix(fireworks_ai/cost_calculator.py): add fireworks ai default pricing
handles calling 405b+ size models
* fix(slack_alerting.py): fix error alerting for failed spend tracking
Fixes regression with slack alerting error monitoring
* fix(vertex_and_google_ai_studio_gemini.py): handle gemini no candidates in streaming chunk error
* docs(bedrock.md): add llama3-1 models
* test: fix tests
* fix(azure_ai/chat): fix transformation for azure ai calls
* feat(azure_ai/embed): Add azure ai embeddings support
Closes https://github.com/BerriAI/litellm/issues/5861
* fix(azure_ai/embed): enable async embedding
* feat(azure_ai/embed): support azure ai multimodal embeddings
* fix(azure_ai/embed): support async multi modal embeddings
* feat(together_ai/embed): support together ai embedding calls
* feat(rerank/main.py): log source documents for rerank endpoints to langfuse
improves rerank endpoint logging
* fix(langfuse.py): support logging `/audio/speech` input to langfuse
* test(test_embedding.py): fix test
* test(test_completion_cost.py): fix helper util
* LiteLLM Minor Fixes & Improvements (09/23/2024) (#5842)
* feat(auth_utils.py): enable admin to allow client-side credentials to be passed
Makes it easier for devs to experiment with finetuned fireworks ai models
* feat(router.py): allow setting configurable_clientside_auth_params for a model
Closes https://github.com/BerriAI/litellm/issues/5843
* build(model_prices_and_context_window.json): fix anthropic claude-3-5-sonnet max output token limit
Fixes https://github.com/BerriAI/litellm/issues/5850
* fix(azure_ai/): support content list for azure ai
Fixes https://github.com/BerriAI/litellm/issues/4237
* fix(litellm_logging.py): always set saved_cache_cost
Set to 0 by default
* fix(fireworks_ai/cost_calculator.py): add fireworks ai default pricing
handles calling 405b+ size models
* fix(slack_alerting.py): fix error alerting for failed spend tracking
Fixes regression with slack alerting error monitoring
* fix(vertex_and_google_ai_studio_gemini.py): handle gemini no candidates in streaming chunk error
* docs(bedrock.md): add llama3-1 models
* test: fix tests
* fix(azure_ai/chat): fix transformation for azure ai calls
* feat(aws_base_llm.py): prevents recreating boto3 credentials during high traffic
Leads to 100ms perf boost in local testing
* fix(base_aws_llm.py): fix credential caching check to see if token is set
* refactor(bedrock/chat): separate converse api and invoke api + isolate converse api transformation logic
Make it easier to see how requests are transformed for /converse
* fix: fix imports
* fix(bedrock/embed): fix reordering of headers
* fix(base_aws_llm.py): fix get credential logic
* fix(converse_handler.py): fix ai21 streaming response
* add max_completion_tokens
* add max_completion_tokens
* add max_completion_tokens support for OpenAI models
* add max_completion_tokens param
* add max_completion_tokens for bedrock converse models
* add test for converse maxTokens
* fix openai o1 param mapping test
* move test optional params
* add max_completion_tokens for anthropic api
* fix conftest
* add max_completion tokens for vertex ai partner models
* add max_completion_tokens for fireworks ai
* add max_completion_tokens for hf rest api
* add test for param mapping
* add param mapping for vertex, gemini + testing
* predibase is the most unstable and unusable llm api in prod, can't handle our ci/cd
* add max_completion_tokens to openai supported params
* fix fireworks ai param mapping
* fix(cost_calculator.py): move to debug for noisy warning message on cost calculation error
Fixes https://github.com/BerriAI/litellm/issues/5610
* fix(databricks/cost_calculator.py): Handles model name issues for databricks models
* fix(main.py): fix stream chunk builder for multiple tool calls
Fixes https://github.com/BerriAI/litellm/issues/5591
* fix: correctly set user_alias when passed in
Fixes https://github.com/BerriAI/litellm/issues/5612
* fix(types/utils.py): allow passing role for message object
https://github.com/BerriAI/litellm/issues/5621
* fix(litellm_logging.py): Fix langfuse logging across multiple projects
Fixes issue where langfuse logger was re-using the old logging object
* feat(proxy/_types.py): support adding key-based tags for tag-based routing
Enable tag based routing at key-level
* fix(proxy/_types.py): fix inheritance
* test(test_key_generate_prisma.py): fix test
* test: fix test
* fix(litellm_logging.py): return used callback object
* fix(main.py): pass default azure api version as alternative in completion call
Fixes api error caused due to api version
Closes https://github.com/BerriAI/litellm/issues/5584
* Fixed gemini-1.5-flash pricing (#5590)
* add /key/list endpoint
* bump: version 1.44.21 → 1.44.22
* docs architecture
* Fixed gemini-1.5-flash pricing
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* fix(bedrock/chat.py): fix converse api stop sequence param mapping
Fixes https://github.com/BerriAI/litellm/issues/5592
* fix(databricks/cost_calculator.py): handle databricks model name changes
Fixes https://github.com/BerriAI/litellm/issues/5597
* fix(azure.py): support azure api version 2024-08-01-preview
Closes https://github.com/BerriAI/litellm/issues/5377
* fix(proxy/_types.py): allow dev keys to call cohere /rerank endpoint
Fixes issue where only admin could call rerank endpoint
* fix(azure.py): check if model is gpt-4o
* fix(proxy/_types.py): support /v1/rerank on non-admin routes as well
* fix(cost_calculator.py): fix split on `/` logic in cost calculator
---------
Co-authored-by: F1bos <44951186+F1bos@users.noreply.github.com>
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* fix(utils.py): return citations for perplexity streaming
Fixes https://github.com/BerriAI/litellm/issues/5535
* fix(anthropic/chat.py): support fallbacks for anthropic streaming (#5542)
* fix(anthropic/chat.py): support fallbacks for anthropic streaming
Fixes https://github.com/BerriAI/litellm/issues/5512
* fix(anthropic/chat.py): use module level http client if none given (prevents early client closure)
* fix: fix linting errors
* fix(http_handler.py): fix raise_for_status error handling
* test: retry flaky test
* fix otel type
* fix(bedrock/embed): fix error raising
* test(test_openai_batches_and_files.py): skip azure batches test (for now) quota exceeded
* fix(test_router.py): skip azure batch route test (for now) - hit batch quota limits
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* All `model_group_alias` should show up in `/models`, `/model/info` , `/model_group/info` (#5539)
* fix(router.py): support returning model_alias model names in `/v1/models`
* fix(proxy_server.py): support returning model alias'es on `/model/info`
* feat(router.py): support returning model group alias for `/model_group/info`
* fix(proxy_server.py): fix linting errors
* fix(proxy_server.py): fix linting errors
* build(model_prices_and_context_window.json): add amazon titan text premier pricing information
Closes https://github.com/BerriAI/litellm/issues/5560
* feat(litellm_logging.py): log standard logging response object for pass through endpoints. Allows bedrock /invoke agent calls to be correctly logged to langfuse + s3
* fix(success_handler.py): fix linting error
* fix(success_handler.py): fix linting errors
* fix(team_endpoints.py): Allows admin to update team member budgets
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* Minor IAM AWS OIDC Improvements (#5246)
* AWS IAM: Temporary tokens are valid across all regions after being issued, so it is wasteful to request one for each region.
* AWS IAM: Include an inline policy, to help reduce misuse of overly permissive IAM roles.
* (test_bedrock_completion.py): Ensure we are testing cross AWS region OIDC flow.
* fix(router.py): log rejected requests
Fixes https://github.com/BerriAI/litellm/issues/5498
* refactor: don't use verbose_logger.exception, if exception is raised
User might already have handling for this. But alerting systems in prod will raise this as an unhandled error.
* fix(datadog.py): support setting datadog source as an env var
Fixes https://github.com/BerriAI/litellm/issues/5508
* docs(logging.md): add dd_source to datadog docs
* fix(proxy_server.py): expose `/customer/list` endpoint for showing all customers
* (bedrock): Fix usage with Cloudflare AI Gateway, and proxies in general. (#5509)
* feat(anthropic.py): support 'cache_control' param for content when it is a string
* Revert "(bedrock): Fix usage with Cloudflare AI Gateway, and proxies in gener…" (#5519)
This reverts commit 3fac0349c2.
* refactor: ci/cd run again
---------
Co-authored-by: David Manouchehri <david.manouchehri@ai.moda>
* feat(proxy/_types.py): add lago billing to callbacks ui
Closes https://github.com/BerriAI/litellm/issues/5472
* fix(anthropic.py): return anthropic prompt caching information
Fixes https://github.com/BerriAI/litellm/issues/5364
* feat(bedrock/chat.py): support 'json_schema' for bedrock models
Closes https://github.com/BerriAI/litellm/issues/5434
* fix(bedrock/embed/embeddings.py): support async embeddings for amazon titan models
* fix: linting fixes
* fix: handle key errors
* fix(bedrock/chat.py): fix bedrock ai21 streaming object
* feat(bedrock/embed): support bedrock embedding optional params
* fix(databricks.py): fix usage chunk
* fix(internal_user_endpoints.py): apply internal user defaults, if user role updated
Fixes issue where user update wouldn't apply defaults
* feat(slack_alerting.py): provide multiple slack channels for a given alert type
multiple channels might be interested in receiving an alert for a given type
* docs(alerting.md): add multiple channel alerting to docs
* refactor(bedrock): initial commit to refactor bedrock to a folder
Improve code readability + maintainability
* refactor: more refactor work
* fix: fix imports
* feat(bedrock/embeddings.py): support translating embedding into amazon embedding formats
* fix: fix linting errors
* test: skip test on end of life model
* fix(cohere/embed.py): fix linting error
* fix(cohere/embed.py): fix typing
* fix(cohere/embed.py): fix post-call logging for cohere embedding call
* test(test_embeddings.py): fix error message assertion in test
* fix(utils.py): support 'drop_params' for embedding requests
Fixes https://github.com/BerriAI/litellm/issues/5444
* feat(anthropic/cost_calculation.py): Support calculating cost for prompt caching on anthropic
* feat(types/utils.py): allows us to migrate to openai's equivalent, once that comes out
* fix: fix linting errors
* test: mark flaky test
