+ jrRt^RIt^RIt^RIHt^RIHt^RIHt^RI H t ] !4'dK^RI H t ^RI HtHt^RIHtHtHtHtHtHtHtHtHtHtHtHtHtHtHtH t H!t!H"t"H#t#H$t$^R I%H&t&^R I'H(t(H)t)H*t*^R IH+t+H,t,H-t-H.t.H/t/H0t0H1t1H2t2]'d ^R I3H4t4H5t5H6t6H7t7]Pp!]94t:!R R]&RR7t;0Rmt<!RR4t=!RR],4t>RRlt?R#)z~ Handler for the /v1/responses endpoint (OpenAI Responses API). Supports streaming (SSE) and non-streaming (JSON) responses. N)AsyncGenerator) TYPE_CHECKING)logging)is_serve_available) HTTPException) JSONResponseStreamingResponse)ResponseResponseCompletedEventResponseContentPartAddedEventResponseContentPartDoneEventResponseCreatedEvent ResponseErrorResponseErrorEventResponseFailedEvent&ResponseFunctionCallArgumentsDoneEventResponseFunctionToolCallResponseInProgressEventResponseOutputItemAddedEventResponseOutputItemDoneEventResponseOutputMessageResponseOutputTextResponseReasoningItemResponseReasoningTextDeltaEventResponseReasoningTextDoneEventResponseTextDeltaEventResponseTextDoneEvent)ResponseCreateParamsStreaming)InputTokensDetailsOutputTokensDetails ResponseUsage)BaseGenerateManager BaseHandlerModality ReasoningTextToolCall _StreamErrorbuild_response_parserparse_assistant_message)GenerationConfigPreTrainedModelPreTrainedTokenizerFastProcessorMixinc,a]tRt^MtoV3RltRtVtR#))TransformersResponseCreateParamsStreamingc2<V^8dQh/S[;R&S[;R&#)generation_configseed)strint)format __classdict__s"b/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/cli/serving/response.py __annotate__6TransformersResponseCreateParamsStreaming.__annotate__Ms IN)__name__ __module__ __qualname____firstlineno____annotate_func____static_attributes____classdictcell__r6s@r7r.r.Msr:r.F)totalca]tRt^btoRtV3RlRltV3RlRltV3RlRltV3RlR ltV3R lR lt V3R lR lt V3RlRlt V3RlRlt V3RlRlt V3RlRltV3RlRltV3RlRltV3RlRltRtVtR#)_ResponseStreamBuilderz=Builds SSE events for one streaming Responses API generation.c&<V^8dQhRS[RS[/#)r0 request_idresponse_defaults)r3dict)r5r6s"r7r8#_ResponseStreamBuilder.__annotate__es = =c =d =r:cW nRV 2VnRV 2VnRV 2Vn^Vn^VnRVnRVnRVnRVn RVn RVn .Vn R#)resp_msg_rs_FN) _response_defaultsresp_idmsg_id reasoning_idseq output_indexcontentfull_reasoningreasoning_open message_openreasoning_item message_item tool_calls)selfrHrIs&$$r7__init___ResponseStreamBuilder.__init__esx"3zl+ ZL) !*.  #!<@:>:<r:c <V^8dQhRS[/#r0returnr3)r5r6s"r7r8rKtscr:cd\P!V4pV;P^, unV#))r" chunk_to_sserU)r^eventsses&& r7_emit_ResponseStreamBuilder._emitts$&&u- A  r:c$<V^8dQhRS[RR/#)r0statusrcr rd)r5r6s"r7r8rKysKKKKr:c 8\R/VPBRV/BVB#)rmr;)r rQ)r^rmextras&&,r7 _response _ResponseStreamBuilder._responseys J$11J&JEJJr:c0<V^8dQhRS[S[,/#rblistr3)r5r6s"r7r8rK|s  S  r:c VP\RVPVPR.R7R74VP\ RVPVPR.R7R74.#)zresponse.createdqueued)outputtypesequence_numberresponsezresponse.in_progress in_progress)rjr rUrprr^s&r7start_response%_ResponseStreamBuilder.start_response|sj JJ$+$(HH!^^HR^@  JJ'/$(HH!^^M"^E   r:c0<V^8dQhRS[S[,/#rbrs)r5r6s"r7r8rKs   c  r:cRVnVP\RVPVP\ VP R..RR7R74.#)Tresponse.output_item.added reasoningr|idrysummaryrWrmryrzrVitem)rYrjrrUrVrrTr}s&r7start_reasoning&_ResponseStreamBuilder.start_reasoningsV" JJ,5$(HH!%!2!2.,,;TV_l    r:c6<V^8dQhRS[RS[S[,/#r0textrcr3rt)r5r6s"r7r8rKs   C  DI  r:c V;PV, unVP\RVPVPVP ^VR74.#)zresponse.reasoning_text.delta)ryitem_idrzrV content_indexdelta)rXrjrrTrUrVr^rs&&r7reasoning_delta&_ResponseStreamBuilder.reasoning_deltasS t# JJ/8 --$(HH!%!2!2"#    r:c0<V^8dQhRS[S[,/#rbrs)r5r6s"r7r8rKs$s)r:c \VPR.RRRVP/.RR7VnVP \ RVPVP VP^VPR74VP \RVP VPVPR 74.pR Vn V;P^, unV#) rryreasoning_textr completedrzresponse.reasoning_text.done)ryrrzrVrrresponse.output_item.donerF) rrTrXr[rjrrUrVrrY)r^partss& r7finish_reasoning'_ResponseStreamBuilder.finish_reasonings3  .8K8KLM   JJ.7 --$(HH!%!2!2"#,,   JJ+4$(HH!%!2!2,,   ($ Q r:c0<V^8dQhRS[S[,/#rbrs)r5r6s"r7r8rKs  tCy r:c8RVnVP\RVPVP\ VP RRR.R7R74VP\RVP VPVP^\RR .R 7R 74.#) Trmessager| assistant)rryrmrolerWrzresponse.content_part.added output_textrPryr annotationsryrrzrVrpart) rZrjrrUrVrrSr rr}s&r7 start_message$_ResponseStreamBuilder.start_messages  JJ,5$(HH!%!2!2.;;Y}S^hj   JJ-6 KK$(HH!%!2!2"#+RUWX    r:c6<V^8dQhRS[RS[S[,/#rr)r5r6s"r7r8rKs  s tCy r:c V;PV, unVP\RVPVPVP ^V.R74.#)zresponse.output_text.delta)ryrrzrVrrlogprobs)rWrjrrSrUrVrs&&r7 text_delta!_ResponseStreamBuilder.text_deltasQ  JJ&5 KK$(HH!%!2!2"#   r:c0<V^8dQhRS[S[,/#rbrs)r5r6s"r7r8rKs**S *r:c  \RVP.R7p\VPRRRV..R7VnVP \ RVPVPVP^VP.R74VP \RVPVPVP^VR 74VP \R VPVPVPR 74.pR Vn V#) rrrrrrryrmrrWrzresponse.output_text.done)ryrrzrVrrrzresponse.content_part.donerrrF) rrWrrSr\rjrrUrVr rrZ)r^output_text_partrs& r7finish_message%_ResponseStreamBuilder.finish_messages-=t||acd1{{%&   JJ%4 KK$(HH!%!2!2"#  JJ,5 KK$(HH!%!2!2"#)   JJ+4$(HH!%!2!2**  - >" r:c B<V^8dQhRS[RS[RS[RS[S[,/#)r0tc_idname argumentsrcr)r5r6s"r7r8rK!s-& & s& #& #& $s)& r:c V;P^, un\VVRVVRR7pVPPV4VP \ RVP VPVR74VP \RVP VVPVVR74VP \RVP VPVR74.#) rf function_callrrcall_idryrrrmrrz%response.function_call_arguments.done)ryrzrrVrrr) rVrr]appendrjrrUrr)r^rrrrs&&&& r7 tool_call _ResponseStreamBuilder.tool_call!s Q'    t$ JJ,5$(HH!%!2!2   JJ6@$(HH!!%!2!2'   JJ+4$(HH!%!2!2  '  r:c6<V^8dQhRS[RS[S[,/#)r0msgrcr)r5r6s"r7r8rKIs     c  r:cVP\RVPVR74VP\RVPVP R.\ RVR7R7R74.#)error)ryrzrzresponse.failedfailed server_error)coder)rwrrx)rjrrUrrpr)r^rs&&r7r_ResponseStreamBuilder.errorIse JJ)wZ]^ _ JJ#*$(HH!^^ =n^a3b,   r:c0<V^8dQhRS[S[,/#rbrs)r5r6s"r7r8rKWs  $s) r:c 6.pVPeVPVP4VPVP4VPVP4VP \ RVPVPRW!R7R74.#)Nzresponse.completedr)rwusagerx) r[rr\extendr]rjr rUrp)r^r all_outputs&& r7r _ResponseStreamBuilder.completedWs    *   d11 2$++,$//* JJ&-$(HH!^^K ^X   r:) rQrWrXr\rZrSrVrTr[rYrRrUr]N)r<r=r>r?__doc__r_rjrpr~rrrrrrrrrrArBrCs@r7rFrFbsG = = KK  $      @  2  **X& & P     r:rFcaa]tRtRtoRt]t]tV3RlRlt ] V3RlRl4t ] V3RlRl4t ] V3R lR l4t V3R lR ltV3R lRltRV3RlV3RllltRtVtV;t#)ResponseHandlerihz+Handler for the ``/v1/responses`` endpoint.c<<V^8dQhRS[RS[RS[S[,/#)r0bodyrHrc)rJr3rr)r5r6s"r7r8ResponseHandler.__annotate__ns.NNN3NCTWcCcNr:c "VPV4VPV4wr4pVPPWER7pVPP WF4p\ PRV RV 24VPPW7R7pVPV4p VPW4p \;QJdRV 4F 'gK RM RM !RV 44p /p V 'd^ V R&V PVP4T PVPR 4;'g/4VPVPR 44p VP !V 3R RR T R V'dR MRRRRRRV\"P$8H;'dT /V BpV'gVP'VP(4pVP+WP,VR7pV'dVP/WO4VPRR4pV'dVP1VVVVVVVVR7#VP3VVVVVVVVR7GR jxL #L5i)aJValidate, load model, dispatch to streaming or non-streaming. Args: body (`dict`): The raw JSON request body (OpenAI Responses API format). request_id (`str`): Unique request identifier (from header or auto-generated). Returns: `StreamingResponse | JSONResponse`: SSE stream or JSON depending on ``body["stream"]``. ) processorz[Request received] Model: z, CB: use_cbc3"TFXp\VPR4\4'dVPR4M.FpVPR4R8HxK KZ R#5i)rWryvideoN) isinstancegetrt).0rcs& r7 1ResponseHandler.handle_request..sU ',6swwy7I4,P,Pcggi(VXX EE&MW $X %'sA A"TF num_frameschat_template_kwargstoolsadd_generation_promptreturn_tensorsNpt return_dicttokenizeload_audio_from_videostream) gen_manager)_validate_request_resolve_model model_managerget_model_modalitygeneration_stateuse_continuous_batchingloggerwarning get_manager_normalize_input"get_processor_inputs_from_messagesanyupdaterr_normalize_toolsapply_chat_templater# MULTIMODALtodevice_build_generation_configr1init_cb _streaming_non_streaming)r^rrHmodel_idmodelrmodalityrrmessagesprocessor_inputs has_videorrinputs gen_config streamings&&& r7handle_requestResponseHandler.handle_requestnsm t$%)%8%8%>"%%888T&&>>uO3H:VF8LM++777P ((.BB8VC ' CCC '  &( 13  .##D$=$=>##DHH-C$D$J$JK%%dhhw&78..   "&    $*4t      #+h.A.A"A"O"Oi  #  YYu||,F2249P9PY_2`     2HHXt, ??'#  ,,'-    s?CII6AI9AI;!II,A*I5I I Icb<V^8dQhRS[S[,R,RS[S[,R,/#)r0rNrcrtrJ)r5r6s"r7r8rs-   T T 1  d4j46G  r:c V'gV#VUUUu.uF=pRV9d2RRRVP4UUu/uFwr#VR8wgKW#bK upp/MTNK? uppp#uuppiuupppi)aXNormalize Responses API tool definitions for ``apply_chat_template``. The Responses API uses a flat format: ``{"type": "function", "name": ..., "parameters": ...}`` while ``apply_chat_template`` expects a nested format: ``{"type": "function", "function": {"name": ..., "parameters": ...}}``. Already-nested tools are passed through unchanged. functionry)items)rtkvs& r7r ResponseHandler._normalize_toolssqL ^hop]pVZqwwy-XytqAQWKdady-X Yvw w  -X s"A AA AAc6<V^8dQhRS[RS[S[,/#)r0rrc)rJrt)r5r6s"r7r8rs**t*T *r:c VR,pVPR4p\V\4'd RRRV/.pMW\V\4'd5V'dRV^,9d RRRV/.pM#\P V4pM \ RRR7hV'dAV'd#V^,R,R 8XdW#^,R&V#VP^RR RV/4V#) uNormalize the Responses API ``input`` field into chat messages. The Responses API accepts multiple input formats. This method converts them into a structure close to what ``apply_chat_template`` expects (messages with ``role``, ``content``, ``tool_calls``, ``tool_call_id``). Further processing is done by ``get_processor_inputs_from_messages``. NOTE: if this conversion logic grows too complex, consider having separate ``get_processor_inputs_from_messages`` implementations for chat completions and the Responses API instead of funneling both through the same path. Formats handled: - **String** → single user message. - **Flat content list** (``input_text``, ``input_image``, no ``role``) → user message. - **Multi-turn list** — messages and tool call items (``function_call``, ``function_call_output``) from a previous response, converted via :meth:`_normalize_response_items`. If ``instructions`` is present, it is prepended as a system message. input instructionsruserrWz 'input' must be a string or list status_codedetailsystem)rrr3rtr_normalize_response_itemsrinsert)rinprrs& r7r ResponseHandler._normalize_inputs,7mxx/ c3  C89H T " "vSV+#VY<=*DDSIC8Z[ [ HQK/8;)5 I&FHi#NOr:cF<V^8dQhRS[S[,RS[S[,/#)r0rrcr )r5r6s"r7r8rs#00d0T 0r:c  .pRpVEFbpVPR4pVR8Xd4RPRVPR4;'g.44pKORV9dIRVR,RVPRR4/pVeVR,R8XdW%R &RpVPV4KVR 8XdR VR ,R RVR,RVR,//pV'dAVR,R,R8Xd,VR,PR.4PV4EK VPRRRV./4EK%VR8Xd)VPRRRVR ,RVR,/4EKT\ RRV: 2R7h V#)uConvert a list of Responses API items into chat messages. Input items may be a mix of: - Messages (``EasyInputMessageParam`` with ``role``, or ``type: "message"``). - ``reasoning`` — buffered and attached as ``reasoning_content`` to the next assistant message. - ``function_call`` — merged as ``tool_calls`` onto the preceding assistant message. - ``function_call_output`` — converted to ``role: "tool"`` messages. NryrrPc32"TF qR,xK R#5i)rNr;)rrs& r7r.s+Y?X!fII?XsrWrrreasoning_contentrrrrrrr]function_call_outputtool tool_call_idrwrzUnsupported input item type: r)rjoinr setdefaultr)rrpending_reasoningr item_typertcs& r7r)ResponseHandler._normalize_response_itemssy(,D(IK'$&GG+Ytxx ?R?X?XVX?X+Y$Y!~tF|YB8OP$0T&\[5P/@+,(,%$o-$y/f{DDU V V 4 CRL++L"=DDRHOOV[,$MN44&Y!4>$>[\e[h fQ  fQ  fQfQ'fQ)fQ fQr:c  aa a aa\W2VR,4p VPVVVVSV R7wo oVR,p \V \4'd \ V 4MV P R,o RRS 2R\ P !4RVRRR .R VPR R 4R R /oRV V VVV3Rllp \V !4RR7#)zDGenerate a streaming Responses API reply (SSE) using DirectStreamer. input_ids)rHresponse_parserrrM created_atrobjectr{rparallel_tool_callsF tool_choiceautoc>V^8dQhR\\R3,/#)r0rcN)rr3)r5s"r7r80ResponseHandler._streaming..__annotate__Vs@ @ N39$=@ r:c<"\SS R7pRPVP445x^pRpV'g=SP4GRjxL .pVP SP 44K"VP'd#RPVP!445xVP"'g#RPVP3445xRPVP%445xRPVP7\9SS P:4445xR#L \ PdMi;i.pTEF1pTfRpEM)\T\4'dd\PRTP 24TPTPTP44RPT45xR#\T\4'dTP'd TPTP!44TP"'d TPTP%44TPTP'S RT 2TP(TP*44T^, pEKE\T\,4'dUTP'g TPTP/44TPTP1T44EKTP'd TPTP!44TP"'g TPTP344TPTP5T44EK4 T'gEKiRPT45xEK \<\ P>3dS PA4hi;i5i))rHrIrPFNTz"Exception in response generation: _tool_call_)!rFr,r~rr get_nowaitasyncio QueueEmptyrr&rrrrr%rYrrZrrrrr$rrrrr compute_usage total_tokens GeneratorExitCancelledErrorcancel) buildertool_call_indexdonebatchrr input_lenqueuerHrIstreamers r7 event_stream0ResponseHandler._streaming..event_streamVs, VghG> ggg44677"##(99;./E"!LL)9)9); GD  GD  GDGD'GD)GD GDr:c  "VPW#WgVR7GRjxL wrp \W2WR,V R7wrp .pV 'd(VP\RV 2R.RRR V /.R R 74VP\ R V 2R R R\ RV .R7..R74V 'dQ\ V 4FAwppV RV 2pVP\VVRVPVPR R74KC \V \V 44p\RV 2\P!4R VVRV.VPRR4RR7 p\VP!RR74#ELD5i)z;Generate a non-streaming Responses API reply (single JSON).)rHNr5)r5cleaned_contentrOrryrrrrrNrrrrrr?rrrMr{r9Fr;) rr7rmrrwr8rrr9r:T) exclude_none)generate_non_streamingr(rrrr enumeraterrrrCrSr rUrr model_dump)r^rHrrrrrrrrWrL generated_idsr'parsed_tool_calls output_itemsir0rrr{s&&&&&&&&& r7rResponseHandler._non_streamings3>2T2T fZ3U3 - )M9P mk7J\c9 5$5    %ZL)$$&6@QRS&    !*&" +WZ\]^    "#452%,k!5##, %,WW"$,,*  6i]);<zl#yy{ $)> F  H//T/BCCw- sE!EEE!c*<V^8dQhRS[RRRS[/#)r0rmodel_generation_configr)r)rJbool)r5r6s"r7r8rs$!!T!L^!hl!r:c ~<\SV`WVR7pVPR4e\VR,4VnV#)zXApply Responses API params (``max_output_tokens``) on top of the base generation config.rmax_output_tokens)superrrr4max_new_tokens)r^rrdrr1 __class__s&&&& r7r(ResponseHandler._build_generation_configsC!Gr?rr._valid_params_classUNUSED_RESPONSE_FIELDS_unused_fieldsr  staticmethodrrrrrrrArB __classcell__)rjr6s@@r7rrhs5C+NNNd    **X00hfQfQTGDGDV!!!r:rc<V^8dQhR\R\R\/#)r0 input_tokens output_tokensrc)r4r )r5s"r7r8r8s!CMr:c \TTW,\RR^/R\P9dR^/M/B\^R7R7#)zBuild a ``ResponseUsage`` object for a Responses API reply. Args: input_tokens (`int`): Number of prompt tokens. output_tokens (`int`): Number of generated tokens. Returns: `ResponseUsage`: Usage statistics with zero-filled detail fields. cached_tokenscache_write_tokens)reasoning_tokens)rrrsrDinput_tokens_detailsoutput_tokens_detailsr;)r r model_fieldsr)rrrss&&r7rCrCsV !#!1/  +?CUCbCb+b#Q'hj 21E  r:> rrstorepromptinclude background truncationr: service_tier top_logprobsmax_tool_callsprevious_response_id)@rrArUcollections.abcrtypingrutilsrutils.import_utilsrfastapirfastapi.responsesrropenai.types.responsesr r r r r rrrrrrrrrrrrrrr-openai.types.responses.response_create_paramsr%openai.types.responses.response_usagerrr r!r"r#r$r%r&r'r( transformersr)r*r+r, get_loggerr<rr.rmrFrrCr;r:r7rs  * 4%A,\ll   gg   H %0MUZ  C C LF!kF!R r: