* feat(proxy/utils.py): get associated litellm budget from db in combined_view for key
allows user to create rate limit tiers and associate those to keys
* feat(proxy/_types.py): update the value of key-level tpm/rpm/model max budget metrics with the associated budget table values if set
allows rate limit tiers to be easily applied to keys
* docs(rate_limit_tiers.md): add doc on setting rate limit / budget tiers
make feature discoverable
* feat(key_management_endpoints.py): return litellm_budget_table value in key generate
make it easy for user to know associated budget on key creation
* fix(key_management_endpoints.py): document 'budget_id' param in `/key/generate`
* docs(key_management_endpoints.py): document budget_id usage
* refactor(budget_management_endpoints.py): refactor budget endpoints into separate file - makes it easier to run documentation testing against it
* docs(test_api_docs.py): add budget endpoints to ci/cd doc test + add missing param info to docs
* fix(customer_endpoints.py): use new pydantic obj name
* docs(user_management_heirarchy.md): add simple doc explaining teams/keys/org/users on litellm
* Litellm dev 12 26 2024 p2 (#7432)
* (Feat) Add logging for `POST v1/fine_tuning/jobs` (#7426)
* init commit ft jobs logging
* add ft logging
* add logging for FineTuningJob
* simple FT Job create test
* (docs) - show all supported Azure OpenAI endpoints in overview (#7428)
* azure batches
* update doc
* docs azure endpoints
* docs endpoints on azure
* docs azure batches api
* docs azure batches api
* fix(key_management_endpoints.py): fix key update to actually work
* test(test_key_management.py): add e2e test asserting ui key update call works
* fix: proxy/_types - fix linting erros
* test: update test
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* fix: test
* fix(parallel_request_limiter.py): enforce tpm/rpm limits on key from tiers
* fix: fix linting errors
* test: fix test
* fix: remove unused import
* test: update test
* docs(customer_endpoints.py): document new model_max_budget param
* test: specify unique key alias
* docs(budget_management_endpoints.py): document new model_max_budget param
* test: fix test
* test: fix tests
---------
Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
* test: add new test image embedding to base llm unit tests
Addresses https://github.com/BerriAI/litellm/issues/6515
* fix(bedrock/embed/multimodal-embeddings): strip data prefix from image urls for bedrock multimodal embeddings
Fix https://github.com/BerriAI/litellm/issues/6515
* feat: initial commit for fireworks ai audio transcription support
Relevant issue: https://github.com/BerriAI/litellm/issues/7134
* test: initial fireworks ai test
* feat(fireworks_ai/): implemented fireworks ai audio transcription config
* fix(utils.py): register fireworks ai audio transcription config, in config manager
* fix(utils.py): add fireworks ai param translation to 'get_optional_params_transcription'
* refactor(fireworks_ai/): define text completion route with model name handling
moves model name handling to specific fireworks routes, as required by their api
* refactor(fireworks_ai/chat): define transform_Request - allows fixing model if accounts/ is missing
* fix: fix linting errors
* fix: fix linting errors
* fix: fix linting errors
* fix: fix linting errors
* fix(handler.py): fix linting errors
* fix(main.py): fix tgai text completion route
* refactor(together_ai/completion): refactors together ai text completion route to just use provider transform request
* refactor: move test_fine_tuning_api out of local_testing
reduces local testing ci/cd time
* fix(invoke_handler.py): fix mock response iterator to handle tool calling
returns tool call if returned by model response
* fix(prometheus.py): add new 'tokens_by_tag' metric on prometheus
allows tracking 'token usage' by task
* feat(prometheus.py): add input + output token tracking by tag
* feat(prometheus.py): add tag based deployment failure tracking
allows admin to track failure by use-case
* fix(prometheus.py): support streaming end user litellm_proxy_total_requests_metric tracking
* fix(prometheus.py): add 'requested_model' and 'end_user_id' to 'litellm_request_total_latency_metric_bucket'
enables latency tracking by end user + requested model
* fix(prometheus.py): add end user, user and requested model metrics to 'litellm_llm_api_latency_metric'
* test: update prometheus unit tests
* test(test_prometheus.py): update tests
* test(test_prometheus.py): fix test
* test: reorder test
* build(model_prices_and_context_window.json): add gemini-1.5-flash context caching
* fix(context_caching/transformation.py): just use last identified cache point
Fixes https://github.com/BerriAI/litellm/issues/6738
* fix(context_caching/transformation.py): pick first contiguous block - handles system message error from google
Fixes https://github.com/BerriAI/litellm/issues/6738
* fix(vertex_ai/gemini/): track context caching tokens
* refactor(gemini/): place transformation.py inside `chat/` folder
make it easy for user to know we support the equivalent endpoint
* fix: fix import
* refactor(vertex_ai/): move vertex_ai cost calc inside vertex_ai/ folder
make it easier to see cost calculation logic
* fix: fix linting errors
* fix: fix circular import
* feat(gemini/cost_calculator.py): support gemini context caching cost calculation
generifies anthropic's cost calculation function and uses it across anthropic + gemini
* build(model_prices_and_context_window.json): add cost tracking for gemini-1.5-flash-002 w/ context caching
Closes https://github.com/BerriAI/litellm/issues/6891
* docs(gemini.md): add gemini context caching architecture diagram
make it easier for user to understand how context caching works
* docs(gemini.md): link to relevant gemini context caching code
* docs(gemini/context_caching): add readme in github, make it easy for dev to know context caching is supported + where to go for code
* fix(llm_cost_calc/utils.py): handle gemini 128k token diff cost calc scenario
* fix(deepseek/cost_calculator.py): support deepseek context caching cost calculation
* test: fix test
* fix(main.py): support 'mock_timeout=true' param
allows mock requests on proxy to have a time delay, for testing
* fix(main.py): ensure mock timeouts raise litellm.Timeout error
triggers retry/fallbacks
* fix: fix fallback + mock timeout testing
* fix(router.py): always return remaining tpm/rpm limits, if limits are known
allows for rate limit headers to be guaranteed
* docs(timeout.md): add docs on mock timeout = true
* fix(main.py): fix linting errors
* test: fix test
* fix(proxy_server.py): enforce team id based model add only works if enterprise user
* fix(auth_checks.py): enforce common_checks can only be imported by user_api_key_auth.py
* fix(auth_checks.py): insert not premium user error message on failed common checks run
* fix(utils.py): e2e azure tts cost tracking working
moves tts response obj to include hidden params (allows for litellm call id, etc. to be sent in response headers) ; fixes spend_Tracking_utils logging payload to account for non-base model use-case
Fixes https://github.com/BerriAI/litellm/issues/7223
* fix: fix linting errors
* build(model_prices_and_context_window.json): add bedrock llama 3.3
Closes https://github.com/BerriAI/litellm/issues/7329
* fix(openai.py): fix return type for sync openai httpx response
* test: update test
* fix(spend_tracking_utils.py): fix if check
* fix(spend_tracking_utils.py): fix if check
* test: improve debugging for test
* fix: fix import
* fix(proxy_track_cost_callback.py): log to db if only end user param given
* fix: allows for jwt-auth based end user id spend tracking to work
* fix(utils.py): fix 'get_end_user_id_for_cost_tracking' to use 'user_api_key_end_user_id'
more stable - works with jwt-auth based end user tracking as well
* test(test_jwt.py): add e2e unit test to confirm end user cost tracking works for spend logs
* test: update test to use end_user api key hash param
* fix(langfuse.py): support end user cost tracking via jwt auth + langfuse
logs end user to langfuse if decoded from jwt token
* fix: fix linting errors
* test: fix test
* test: fix test
* fix: fix end user id extraction
* fix: run test earlier
* feat(router.py): support passing model-specific messages in fallbacks
* docs(routing.md): separate router timeouts into separate doc
allow for 1 fallbacks doc (across proxy/router)
* docs(routing.md): cleanup router docs
* docs(reliability.md): cleanup docs
* docs(reliability.md): cleaned up fallback doc
just have 1 doc across sdk/proxy
simplifies docs
* docs(reliability.md): add setting model-specific fallback prompts
* fix: fix linting errors
* test: skip test causing openai rate limit errros
* test: fix test
* test: run vertex test first to catch error
* fix(proxy_server.py): only update k,v pair if v is not empty/null
Fixes https://github.com/BerriAI/litellm/issues/6787
* test(test_router.py): cleanup duplicate calls
* test: add new test stream options drop params test
* test: update optional params / stream options test to test for vertex ai mistral route specifically
Addresses https://github.com/BerriAI/litellm/issues/7309
* fix(proxy_server.py): fix linting errors
* fix: fix linting errors
* fix(health.md): add rerank model health check information
* build(model_prices_and_context_window.json): add gemini 2.0 for google ai studio - pricing + commercial rate limits
* build(model_prices_and_context_window.json): add gemini-2.0 supports audio output = true
* docs(team_model_add.md): clarify allowing teams to add models is an enterprise feature
* fix(o1_transformation.py): add support for 'n', 'response_format' and 'stop' params for o1 and 'stream_options' param for o1-mini
* build(model_prices_and_context_window.json): add 'supports_system_message' to supporting openai models
needed as o1-preview, and o1-mini models don't support 'system message
* fix(o1_transformation.py): translate system message based on if o1 model supports it
* fix(o1_transformation.py): return 'stream' param support if o1-mini/o1-preview
o1 currently doesn't support streaming, but the other model versions do
Fixes https://github.com/BerriAI/litellm/issues/7292
* fix(o1_transformation.py): return tool calling/response_format in supported params if model map says so
Fixes https://github.com/BerriAI/litellm/issues/7292
* fix: fix linting errors
* fix: update '_transform_messages'
* fix(o1_transformation.py): fix provider passed for supported param checks
* test(base_llm_unit_tests.py): skip test if api takes >5s to respond
* fix(utils.py): return false in 'supports_factory' if can't find value
* fix(o1_transformation.py): always return stream + stream_options as supported params + handle stream options being passed in for azure o1
* feat(openai.py): support stream faking natively in openai handler
Allows o1 calls to be faked for just the "o1" model, allows native streaming for o1-mini, o1-preview
Fixes https://github.com/BerriAI/litellm/issues/7292
* fix(openai.py): use inference param instead of original optional param
* fix(hosted_vllm/transformation.py): return fake api key, if none give. Prevents httpx error
Fixes https://github.com/BerriAI/litellm/issues/7291
* test: fix test
* fix(main.py): add hosted_vllm/ support for embeddings endpoint
Closes https://github.com/BerriAI/litellm/issues/7290
* docs(vllm.md): add docs on vllm embeddings usage
* fix(__init__.py): fix sambanova model test
* fix(base_llm_unit_tests.py): skip pydantic obj test if model takes >5s to respond
* fix(factory.py): skip empty text blocks for bedrock user messages
Fixes https://github.com/BerriAI/litellm/issues/7169
* Add support for Gemini 2.0 GoogleSearch tool (#7257)
* Add support for google_search tool in gemini 2.0
* Add/modify tests
* Fix grounding check
* Remove 2.0 grounding test; exclude experimental model in VERTEX_MODELS_TO_NOT_TEST
* Swap order of tools
* DFix formatting
* fix(get_api_base.py): return api base in streaming response
Fixes https://github.com/BerriAI/litellm/issues/7249
Closes https://github.com/BerriAI/litellm/pull/7250
* fix(cost_calculator.py): only set base model to model if not none
Fixes https://github.com/BerriAI/litellm/issues/7223
* fix(cost_calculator.py): enforce stricter order when picking model for cost calculation
* fix(cost_calculator.py): fix '_select_model_name_for_cost_calc' to return model name with region name prefix if provided
* fix(utils.py): fix 'get_model_info()' to handle edge case where model name starts with custom llm provider AND custom llm provider is given
* fix(cost_calculator.py): handle `custom_llm_provider-` scenario
* fix(cost_calculator.py): e2e working tts cost tracking
ensures initial message is passed in, to cost calculator
* fix(factory.py): suppress linting errors
* fix(cost_calculator.py): strip llm provider from model name after selecting cost calc model
* fix(litellm_logging.py): store initial request in 'input' field + accept base_model to be passed in litellm_params directly
* test: handle none env var value in flaky test
* fix(litellm_logging.py): fix linting errors
---------
Co-authored-by: Sam B <samlingx@gmail.com>
* fix(utils.py): fix openai-like api response format parsing
Fixes issue passing structured output to litellm_proxy/ route
* fix(cost_calculator.py): fix whisper transcription cost calc to use file duration, not response time
'
* test: skip test if credentials not found
* fix(router.py): fix reading + using deployment-specific num retries on router
Fixes https://github.com/BerriAI/litellm/issues/7001
* fix(router.py): ensure 'timeout' in litellm_params overrides any value in router settings
Refactors all routes to use common '_update_kwargs_with_deployment' which has the timeout handling
* fix(router.py): fix timeout check
* fix(main.py): fix retries being multiplied when using openai sdk
Closes https://github.com/BerriAI/litellm/pull/7130
* docs(prompt_management.md): add langfuse prompt management doc
* feat(team_endpoints.py): allow teams to add their own models
Enables teams to call their own finetuned models via the proxy
* test: add better enforcement check testing for `/model/new` now that teams can add their own models
* docs(team_model_add.md): tutorial for allowing teams to add their own models
* test: fix test
* fix test_deployment_budget_limits_e2e_test
* refactor async_log_success_event to track spend for provider + deployment
* fix format
* rename class to RouterBudgetLimiting
* rename func
* rename types used for budgets
* add new types for deployment budgets
* add budget limits for deployments
* fix checking budgets set for provider
* update file names
* fix linting error
* _track_provider_remaining_budget_prometheus
* async_filter_deployments
* fix model list passed to router
* update error
* test_deployment_budgets_e2e_test_expect_to_fail
* fix test case
* run deployment budget limits
* fix(litellm_logging.py): pass user metadata to langsmith on sdk calls
* fix(litellm_logging.py): pass nested user metadata to logging integration - e.g. langsmith
* fix(exception_mapping_utils.py): catch and clarify watsonx `/text/chat` endpoint not supported error message.
Closes https://github.com/BerriAI/litellm/issues/7213
* fix(watsonx/common_utils.py): accept new 'WATSONX_IAM_URL' env var
allows user to use local watsonx
Fixes https://github.com/BerriAI/litellm/issues/4991
* fix(litellm_logging.py): cleanup unused function
* test: skip bad ibm test
* feat(bedrock/): add bedrock converse top k param
Closes https://github.com/BerriAI/litellm/issues/7087
* Fix bedrock empty content error (#7177)
* add resolver
* handle empty content on bedrock with default content
* use existing default message, tests
* Update tests/llm_translation/test_bedrock_completion.py
* fix tests
* Revert "add resolver"
This reverts commit c717e376ee09a7547cc2c7770405f9aebca64d62.
* fallback to empty
---------
Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
* fix(factory.py): handle empty content blocks in messages
Fixes https://github.com/BerriAI/litellm/issues/7169
* feat(router.py): add stripped model check to model fallback search
if model_name="openai/gpt-3.5-turbo" and fallback=[{"gpt-3.5-turbo"..}] the fallback should just work as expected
* fix: fix linting error
* fix(factory.py): fix linting error
* fix(factory.py): in base case still support skip empty text blocks
---------
Co-authored-by: Engel Nyst <enyst@users.noreply.github.com>
* fix(azure/): support passing headers to azure openai endpoints
Fixes https://github.com/BerriAI/litellm/issues/6217
* fix(utils.py): move default tokenizer to just openai
hf tokenizer makes network calls when trying to get the tokenizer - this slows down execution time calls
* fix(router.py): fix pattern matching router - add generic "*" to it as well
Fixes issue where generic "*" model access group wouldn't show up
* fix(pattern_match_deployments.py): match to more specific pattern
match to more specific pattern
allows setting generic wildcard model access group and excluding specific models more easily
* fix(proxy_server.py): fix _delete_deployment to handle base case where db_model list is empty
don't delete all router models b/c of empty list
Fixes https://github.com/BerriAI/litellm/issues/7196
* fix(anthropic/): fix handling response_format for anthropic messages with anthropic api
* fix(fireworks_ai/): support passing response_format + tool call in same message
Addresses https://github.com/BerriAI/litellm/issues/7135
* Revert "fix(fireworks_ai/): support passing response_format + tool call in same message"
This reverts commit 6a30dc692986a513cfb99c7a10c7cd34d8b93a4f.
* test: fix test
* fix(replicate/): fix replicate default retry/polling logic
* test: add unit testing for router pattern matching
* test: update test to use default oai tokenizer
* test: mark flaky test
* test: skip flaky test
* fix(acompletion): support fallbacks on acompletion
allows health checks for wildcard routes to use fallback models
* test: update cohere generate api testing
* add max tokens to health check (#7000)
* fix: fix health check test
* test: update testing
---------
Co-authored-by: Cameron <561860+wallies@users.noreply.github.com>
Krish Dholakia