h:dZddlmZmZmZmZmZmZmZm Z m Z ddl m Z ddl mZddlmZmZddlmZmZddlmZerdd lmZdd lmZmZdd lmZdd lmZm Z Gd deeZ!y)z5 OpenAI Moderation Guardrail Integration for LiteLLM ) TYPE_CHECKINGAnyAsyncGeneratorDictListLiteralOptionalTypeUnion) HTTPException)verbose_proxy_logger)CustomGuardraillog_guardrail_information)get_async_httpx_clienthttpxSpecialProvider)OpenAIGuardrailBase)UserAPIKeyAuth)AllMessageValuesOpenAIModerationResponse)GuardrailConfigModel) ModelResponseModelResponseStreamceZdZdZ d dedeedeedeedffd Zdeefd Zd edd fd Z d!d Z e ddde de ee fdeddee ee ff dZe de ee fdddeddee ee ffdZe de ee fddddddfdZe ddde de ee fdedfdZdddeefdZedeedfdZxZS)"OpenAIModerationGuardraila LiteLLM Built-in Guardrail for OpenAI Content Moderation. This guardrail scans prompts and responses using the OpenAI Moderation API to detect harmful content, including violence, hate, harassment, self-harm, sexual content, etc. Configuration: guardrail_name: Name of the guardrail instance api_key: OpenAI API key api_base: OpenAI API endpoint model: OpenAI moderation model to use default_on: Whether to enable by default guardrail_nameapi_keyapi_basemodel)omni-moderation-latestztext-moderation-latestc ddlm}|j|j|jg}t |d ||d|ttj|_ |xs|j|_ |xsd|_ |xsd|_|js tdt!j"d|d |jy ) z/Initialize OpenAI Moderation guardrail handler.r)GuardrailEventHooks)rsupported_event_hooks) llm_providerzhttps://api.openai.com/v1r zlOpenAI Moderation: api_key is required. Set OPENAI_API_KEY environment variable or pass it in configuration.z)Initialized OpenAI Moderation Guardrail: z with model: N)litellm.types.guardrailsr"pre_call during_call post_callsuper__init__rrGuardrailCallback async_handler _get_api_keyrrr ValueErrorr info) selfrrrrkwargsr"r# __class__s w/var/www/Befach/backend/env/lib/python3.12/site-packages/litellm/proxy/guardrails/guardrail_hooks/openai/moderations.pyr+z"OpenAIModerationGuardrail.__init__6s A ( (  + +  ) )!   )"7  4-??  5$"3"3"5  ?$? RWRs[s ||LM M!!77G}UYU_U_T` a returncddl}ddl}ddlm}|jj dxs$|j xs|jxs|dS)z?Get API key from environment variables or litellm configurationrN)get_secret_strOPENAI_API_KEY)oslitellmlitellm.secret_managers.mainr8environgetr openai_key)r1r:r;r8s r4r.z&OpenAIModerationGuardrail._get_api_key]sM? JJNN+ , 0 0!! 0./  r5 input_textrcK|j|d}tjd||jj |j dd|j dd|d{}tjd |j|jd k7r$t|jd |jd  ddl m }|di|jS7ww)z> Make a request to the OpenAI Moderation API. )rinputz#OpenAI Moderation guard request: %sz /moderationszBearer zapplication/json) Authorizationz Content-Type)urlheadersjsonNz$OpenAI Moderation guard response: %sz$OpenAI Moderation API request failed)errordetails status_codedetailr)rr%) rr debugr-postrrrFrKr textlitellm.types.llms.openair)r1r@ request_bodyresponsers r4async_make_requestz,OpenAIModerationGuardrail.async_make_requestksZZ "" 1< ++00==/.#*4<<.!9 2 1   "" 2HMMO    3 &$00C'}}  G':(--/::/ sA$C &C'A8C cN|jsy|jd}|jr}g}|jr6|jjD]\}}|s |j |||j xsid}t jd|tdd|dy) zk Check if the moderation response indicates harmful content and raise exception if needed. Nr)violated_categoriescategory_scoresz5OpenAI Moderation: Content flagged for violations: %siz!Violated OpenAI moderation policy)rHmoderation_resultrJ) resultsflagged categoriesitemsappendrVr warningr )r1moderation_responseresultrUcategory is_violatedviolation_detailss r4_check_moderation_resultz2OpenAIModerationGuardrail._check_moderation_results#** $,,Q/ >>"$   -3->->-D-D-F)Hk"+228<.G (;#)#9#9#?R!  ! ( (G!   @): % r5user_api_key_dictrcachedata call_type) completiontext_completion embeddingsimage_generation moderationaudio_transcriptionpass_through_endpointrerankcKtjd||dk(r|S|jd}|tjd|S|j |}|rItjd|ddd|j | d{}|j ||Stjd |S7.w) z Pre-call hook to scan user prompts before sending to LLM. Raises HTTPException if content should be blocked. zAOpenAI Moderation: Running pre-call prompt scan, on call_type: %srlmessagesN=OpenAI Moderation: not running guardrail. No messages in dataz OpenAI Moderation: User prompt: d...r@z'OpenAI Moderation: No user prompt foundr r0r>r]get_user_promptrSrc)r1rdrerfrg new_messages user_promptr^s r4async_pre_call_hookz-OpenAIModerationGuardrail.async_pre_call_hooks, !! O   $K;?88J;O   ( (O K**<8  % %2;t3D2ESI )-(?(?&)@)#   ) )*= >  ! ( (9  #sB B?B=/B?)rhrjrkrlrm responsescKtjd||dk(r|S|jd}|tjd|S|j |}|r+|j |d{}|j ||S7w)z Moderation hook to scan user prompts during call processing. Raises HTTPException if content should be blocked. z #sA0B 2B 3B rRrcKtjd|j|}|rGtjd|ddd|j|d{}|j ||S7w)z Post-call hook to scan LLM responses before returning to user. Raises HTTPException if response should be blocked. z2OpenAI Moderation: Running post-call response scanz"OpenAI Moderation: Response text: Nrsrtru)r r0_extract_response_textrSrc)r1rfrdrR response_textr^s r4async_post_call_hookz.OpenAIModerationGuardrail.async_post_call_hooks !! @ 33H=  % %4]4C5H4IM )-(?(?()@)#   ) )*= >#sAA6A4A6 request_data)rNcKddlm}ddlm}ddlm}t jdg}|23d{}|j|76||} t| td|fr"t jd|D]}| y|j| } | rHt jd | dd d |j| d{7} |j| || } | 23d{7}|6yw)a Process streaming response chunks for OpenAI moderation. Collects all chunks from the stream, assembles them into a complete response, and applies moderation check. If content violates moderation policy, raises HTTPException. r)MockResponseIterator)stream_chunk_builder)TextCompletionResponsez2OpenAI Moderation: Running streaming response scanN)chunkszTOpenAI Moderation: Could not assemble ModelResponse from chunks, skipping moderationz,OpenAI Moderation: Streaming response text: rsrtru)model_response))litellm.llms.base_llm.base_model_iteratorr litellm.mainrlitellm.types.utilsrr r0r\ isinstancetyper]rrSrc) r1rdrRrrrr all_chunkschunkassembled_model_responserr^ mock_responses r4'async_post_call_streaming_iterator_hookzAOpenAIModerationGuardrail.async_post_call_streaming_iterator_hook<s9 S5>!! @ 35 # % %%   e $ %8 !  ! .d=S0T U ! ( (f $ $ 334LM  % %>}Tc?R>SSVW  )-(?(?()@)##   ) )*= >-3 )  %K)sJ,DA A A D A  BDC!D4D8C;9D= Dct|dr |jsyg}|jD]} t|dd}|r1t|dd}|r"t|tr|j |Dt|dd}|r"t|tr|j |ut|dd}|r1t|dd}|r"t|tr|j ||rdj|SdS#t tf$rYwxYw)zN Extract text content from the model response for moderation. choicesNmessagecontentrOdelta ) hasattrrgetattrrstrr\AttributeError TypeErrorjoin)r1rRresponse_textschoicerrrOrs r4rz0OpenAIModerationGuardrail._extract_response_texts x+83C3C&&F !&)T:%gy$?G:gs#;&--g6 vvt4JtS1"))$/ 6%eY=G:gs#;&--g6 -'8-;tyy(DD #I.  s?C4,0C4?C44DDrcddlm}|S)zK Get the config model for the OpenAI Moderation guardrail. r$OpenAIModerationGuardrailConfigModel)Glitellm.types.proxy.guardrails.guardrail_hooks.openai.openai_moderationrrs r4get_config_modelz*OpenAIModerationGuardrail.get_config_models  43r5)NNN)r^rr6N)__name__ __module__ __qualname____doc__rr rr+r.rSrcrrrrzr}rrrr staticmethodr r __classcell__)r3s@r4rr's ""&"&W[ % % #% 3- %  RST % N  hsm  &;&; #&;P D6+6638n 6    6 $sCx. !66p+38n+,+  + $sCx. !++Z38n,"   >A+AA38n A 3 4 AAF$E$E8C=$EL4ht,B'CD44r5rN)"rtypingrrrrrrr r r fastapir litellm._loggingr %litellm.integrations.custom_guardrailrr&litellm.llms.custom_httpx.http_handlerrrbaserlitellm.proxy._typesrrPrr3litellm.types.proxy.guardrails.guardrail_hooks.baserrrrrr%r5r4rsP   "1 &3TXFH4 3_H4r5