qi9& dZddlZddlmZmZmZddlZddlmZm Z ddl m Z ddl m Z mZmZmZmZmZmZed Zed Zeeej.d fZeeeej2d fZeeeej2d fZd eddfdZej;ddgdej<ddfdede dededdf dZej;ddgddddddddddedeeej2d fd eeej2d! fd"eeej2d# fd$eeej2d% fd&eeej2d' fd(eeej2d) fd*eeej2d+ fded,eeeej2d- fded.ee ej2d/ fd0ee ej2d1 fd2eee ej2d3 fd4eee ej2d5 fd6eee!ej2d7 fddf"d8Z"ej;dd9g dYdeeej2d fded"eeeej2d: fdededdf d;Z# dZdeddfd<Z$ej;d=d>ge$ej;d?d@Ae$ejKedBCej;dDg d[dedededdfdEZ&ej;dFg d\dededeeeej2d fd"eeeej2d# fd$eeeej2d% fd&eeeej2d' fd eeeej2dG fdHeeeej2dI fd,eeeej2d- fd.eee ej2d/ fd0eee ej2d1 fd2eee ej2d3 fd4eee ej2d5 fd6eee!ej2d7 fdeddf dJZ'ej;dKg d]dededLee(ej2dMdN fdeddf dOZ)ej;dPg d[dedededdfdQZ*ej;dRg d]dededSee(ej2dTdU fdeddf dVZ+ej;dWg d[dedededdfdXZ,y)^z2CLI commands for Hugging Face Inference Endpoints.N) AnnotatedAnyOptional)InferenceEndpointInferenceEndpointScalingMetric)HfHubHTTPError) FormatOpt OutputFormatQuietOptTokenOpt get_hf_apiprint_list_output typer_factoryz(Manage Hugging Face Inference Endpoints.)helpz.Interact with the Inference Endpoints catalog.zEndpoint name.z_The namespace associated with the Inference Endpoint. Defaults to the current user's namespace.endpointreturncntjtj|jddy)NTindent sort_keys)typerechojsondumpsraw)rs Y/opt/pipecat/venv/lib/python3.12/site-packages/huggingface_hub/cli/inference_endpoints.py_print_endpointr,s JJtzz(,,qDABzhf endpoints lsz"hf endpoints ls --namespace my-org)examplesF namespaceformatquiettokenct|} |j||}|Dcgc]}|j}}dtttfdttfd} t|||d gd | y#t$rH}tjd|tj |j j|d}~wwxYwcc}w) z6Lists all Inference Endpoints for the given namespace.r%)r"r%zListing failed: codeNitemrc <|jdi}|jdi}|jdi}|jdi}t|jddtt|tr|jddndtt|tr|jddndtt|tr|jd dndtt|tr|jd dndtt|tr|jd dndtt|tr|jd dndtt|tr|jd dgSdgS)Nstatusmodelcomputeprovidername repositorystatetask framework instanceTypevendorregion)getstr isinstancedict)r*r,r-r.r/s rrow_fnzls..row_fnAsD(B'"%((9b)88J+ $ % z%/F ,+B O :fd+C 7B' L E4)@ &"%b I j.E +r*2 N :gt3L NB/RT U j4.H Xr*b Q j4.H Xr* Q  OQ Q  r r0)NAMEMODELSTATUSTASK FRAMEWORKINSTANCEVENDORREGION)itemsr#r$id_keyheadersr=)rlist_inference_endpointsrrrExitresponse status_coderr<r:rlistr) r"r#r$r%api endpointserrorrresultsr=s rlsrR0s 5 !CE009E0R -66x||6G6 T#s(^ S  ` - E %eW-.jjenn889uDE7sA2C2 C;AB>>CdeployzChf endpoints deploy my-endpoint --repo gpt2 --framework pytorch ...)r0r!)r"r4r% min_replica max_replicascale_to_zero_timeoutscaling_metricscaling_thresholdr0repozeThe name of the model repository associated with the Inference Endpoint (e.g. 'openai/gpt-oss-120b').r5z@The machine learning framework used for the model (e.g. 'vllm'). acceleratorz?The hardware accelerator to be used for inference (e.g. 'cpu'). instance_sizezNThe size or type of the instance to be used for hosting the model (e.g. 'x4'). instance_typezYThe cloud instance type where the Inference Endpoint will be deployed (e.g. 'intel-icl').r8zTThe cloud region in which the Inference Endpoint will be created (e.g. 'us-east-1').r7zVThe cloud provider or vendor where the Inference Endpoint will be hosted (e.g. 'aws').r4zCThe task on which to deploy the model (e.g. 'text-classification').rTzVThe minimum number of replicas (instances) to keep running for the Inference Endpoint.rUzRThe maximum number of replicas (instances) to scale to for the Inference Endpoint.rVzFThe duration in minutes before an inactive endpoint is scaled to zero.rWz!The metric reference for scaling.rXzeThe scaling metric threshold used to trigger a scale up. Ignored when scaling metric is not provided.ctt| }|j|||||||||| | | | ||| }t|y)z3Deploy an Inference Endpoint from a Hub repository.r')r0r2r5rZr[r\r8r7r"r4r%rTrUrWrXrVN)rcreate_inference_endpointr)r0rYr5rZr[r\r8r7r"r4r%rTrUrVrWrXrNrs rrSrS[s^l 5 !C,, ## %+3!-H&Hr zChf endpoints catalog deploy --repo meta-llama/Llama-3.2-1B-InstructzPThe hardware accelerator to be used for inference (e.g. 'cpu', 'gpu', 'neuron').ct|} |j|||||}t|y#t$rH}tjd|tj |j j|d}~wwxYw)z4Deploy an Inference Endpoint from the Model Catalog.r')repo_idr0rZr"r%zDeployment failed: r(N) r&create_inference_endpoint_from_catalogrrrrJrKrLr)rYr0rZr"r%rNrrPs rdeploy_from_catalogrbs& 5 !C E==# > H E (01jjenn889uDE0 BAA<<BcDt|} |j|}tjtjd|iddy#t$rH}tjd|tj |j j|d}~wwxYw) zList available Catalog models.r'zCatalog fetch failed: r(NmodelsrTr) rlist_inference_catalogrrrrJrKrLrr)r%rNrerPs r list_catalogrgs 5 !CE++%+8  JJtzz8V,Q$GH E +E734jjenn889uDEsA BABBrRzhf endpoints catalog lsz list-catalogT)r0hiddencatalog)r0z!hf endpoints describe my-endpointct|} |j|||}t|y#t$rH}tjd|tj |j j|d}~wwxYw)z+Get information about an existing endpoint.r'r0r"r%zFetch failed: r(N) rget_inference_endpointrrrrJrKrLrr0r"r%rNrrPs rdescribernsx 5 !CE--49TY-Z H E ^E7+,jjenn889uDE. A?AA::A?z/hf endpoints update my-endpoint --min-replica 2zBThe machine learning framework used for the model (e.g. 'custom').revisionzrThe specific model revision to deploy on the Inference Endpoint (e.g. '6c0e6080953db56375760c0471a8c5f2929baf11').ct|} |j|||||||||| | | | | |}t|y#t$rH}tjd|tj |j j|d}~wwxYw)zUpdate an existing endpoint.r')r0r"r2r5rpr4rZr[r\rTrUrVrWrXr%zUpdate failed: r(N) rupdate_inference_endpointrrrrJrKrLr)r0r"rYrZr[r\r5rpr4rTrUrVrWrXr%rNrrPs rupdaterss^ 5 !CE00#''##"7)/1 (H E _UG,-jjenn889uDEs : B ABB zhf endpoints delete my-endpointyesz--yeszSkip confirmation prompts.c|sItjd|d}||k7r+tjdtjdt |} |j |||tjd |d y #t $rH}tjd|tj|jj|d }~wwxYw) z)Delete an Inference Endpoint permanently.zDelete endpoint 'z'? Type the name to confirm.zAborted.rr(r'rkzDelete failed: Nz Deleted 'z'.) rpromptrrJrdelete_inference_endpointrrKrL)r0r"rtr% confirmationrNrPs rdeleteryys ||&7v=Y$Z[ 4  JJz "**!$ $ 5 !CE %%49E%R  JJ4&#$ E _UG,-jjenn889uDEsB CACCzhf endpoints pause my-endpointct|} |j|||}t|y#t$rH}tjd|tj |j j|d}~wwxYw)zPause an Inference Endpoint.r'rkzPause failed: r(N) rpause_inference_endpointrrrrJrKrLrrms rpauser|sx 5 !CE//TYV[/\ H E ^E7+,jjenn889uDErozhf endpoints resume my-endpointfail_if_already_runningz--fail-if-already-runningzWIf `True`, the method will raise an error if the Inference Endpoint is already running.ct|} |j|||| }t|y#t$rH}tjd|tj |j j|d}~wwxYw)zResume an Inference Endpoint.r')r0r"r% running_okzResume failed: r(N) rresume_inference_endpointrrrrJrKrLr)r0r"r}r%rNrrPs rresumers 5 !C E0022 1 H E _UG,-jjenn889uDErcz&hf endpoints scale-to-zero my-endpointct|} |j|||}t|y#t$rH}tjd|tj |j j|d}~wwxYw)z$Scale an Inference Endpoint to zero.r'rkzScale To Zero failed: r(N) r scale_to_zero_inference_endpointrrrrJrKrLrrms r scale_to_zerorsy 5 !CE77TY^c7d H E +E734jjenn889uDEro)NNNN)N)NN)NNNNNNNNNNNNNN)NFN)-__doc__rtypingrrrr$huggingface_hub._inference_endpointsrrhuggingface_hub.errorsr _cli_utilsr r r r rrrie_cli catalog_appr:ArgumentNameArgOptionNameOpt NamespaceOptrcommandtablerRintfloatrSrbrg add_typerrnrsboolryr|rrr rrs 8 ++ b1 F G!QR  ENN()+   SMELL&')   SMELL n C/CDC+-QRS"$** '' ' '  '  'T'TX)n(op\#  ei i   x   i  S  i  R  i(  a  )i4  l  5i@   g   AiL   i   MiZ[i\   V   ]ih iij  i  kiv  e  wiB%  Y  CiN/0 4  OiZ! x  [if giqiX (.s-tu "   x      c     !vF I  I  IE )B(CD\R0N40>9-=>?#       @  KLM# Wd dd   x   d  R  d  a  d*  l  +d6  U  7dB  F  CdN   V   OdZ  i  [df  e  gdr%  Y  sd~/0 4  dJ! x  KdV WdX YdNdN;<=# % %%   W#?@ B % % %>%4:;<#       =  ;<=#  '  'j    >6BCD#       E r