iO9(f dZddlmZddlZddlmZddlmZddlm Z m Z m Z m Z ddl mZmZe d Ze d Zeeej&d fZeedzej*d fZeedzej*d fZej1dddgdej2dfdede de ddfdZej1ddgddej2ddddddd dedeeej*d fdeeej*d fd eeej*d! fd"eeej*d# fd$eeej*d% fd&eeej*d' fd(eeej*d) fded*eedzej*d+ fde de d,eeej*d- fd.eeej*d/ fd0eedzej*d1 fd2eedzej*d3 fd4eedzej*d5 fddf$d6Zej1dd7gdddej2dfdeeej*d fded eedzej*d8 fdede de ddfd9Zej2dfde de ddfd:Zej1dd;geej1deejAed?@ej1dAgdej2dfdedede de ddf dBZ!ej1dCgdddddddddddddej2dfdededeedzej*d fd eedzej*d! fd"eedzej*d# fd$eedzej*d% fdeedzej*dD fdEeedzej*dF fd*eedzej*d+ fd,eedzej*d- fd.eedzej*d/ fd0eedzej*d1 fd2eedzej*d3 fd4eedzej*d5 fde de ddf"dGZ"ej1dHgddIej2dfdededJee#ej*dKdL fde de ddf dMZ$ej1dNgdej2dfdedede de ddf dOZ%ej1dPgddIej2dfdededQee#ej*dRdS fde de ddf dTZ&ej1dUgdej2dfdedede de ddf dVZ'y)Wz2CLI commands for Hugging Face Inference Endpoints.) AnnotatedN)InferenceEndpointScalingMetric)HfHubHTTPError)FormatWithAutoOptTokenOpt get_hf_api typer_factory)OutputFormatWithAutooutz(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.z list | lszhf endpoints lsz"hf endpoints ls --namespace my-org)examples namespaceformattokenreturnct|} |j||}g}|D]o}|j}|jdi} |jdi} |jdi} |jd i} |j|jd d t| tr| jd d nd t| tr| jd d nd t| tr| jdd nd t| tr| jdd nd t| tr| jdd nd t| tr| jdd nd t| tr| jdd nd drtj|d y#t$rH}tjd|t j |jj|d}~wwxYw)z6Lists all Inference Endpoints for the given namespace.r)rrzListing failed: codeNstatusmodelcomputeprovidername repositorystatetask framework instanceTypevendorregion)rrrrr instancer"r#)id_key)r list_inference_endpointsrr errortyperExitresponse status_coderawgetappend isinstancedicttable) rrrapi endpointsr'resultsendpointr,rrrrs |/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/huggingface_hub/cli/inference_endpoints.pylsr7$s 5 !CE009E0R Gll2&$'')R(77:r*+8B5$8O<4UW5?5M&**Wb1SU1;E41H &"-b;EeT;RUYY{B7XZ?I'SW?XGKK;^`8B8T8R(,,x4XZ8B8T8R(,,x4XZ  $IIgf%/ E $UG,-jjenn889uDEsF00 H9AG<<HdeployzChf endpoints deploy my-endpoint --repo gpt2 --framework pytorch ...)rr) rrrr min_replica max_replicascale_to_zero_timeoutscaling_metricscaling_thresholdrrepozeThe name of the model repository associated with the Inference Endpoint (e.g. 'openai/gpt-oss-120b').r z@The machine learning framework used for the model (e.g. 'vllm'). acceleratorz?The hardware accelerator to be used for inference (e.g. 'cpu'). instance_sizezNThe size or type of the instance to be used for hosting the model (e.g. 'x4'). instance_typezYThe cloud instance type where the Inference Endpoint will be deployed (e.g. 'intel-icl').r#zTThe cloud region in which the Inference Endpoint will be created (e.g. 'us-east-1').r"zVThe cloud provider or vendor where the Inference Endpoint will be hosted (e.g. 'aws').rzCThe task on which to deploy the model (e.g. 'text-classification').r9zVThe minimum number of replicas (instances) to keep running for the Inference Endpoint.r:zRThe maximum number of replicas (instances) to scale to for the Inference Endpoint.r;zFThe duration in minutes before an inactive endpoint is scaled to zero.r<z!The metric reference for scaling.r=zeThe scaling metric threshold used to trigger a scale up. Ignored when scaling metric is not provided.c t| }|j|||||||||| | | | |||}tj|jy)z3Deploy an Inference Endpoint from a Hub repository.r)rrr r?r@rAr#r"rrrr9r:r<r=r;N)r create_inference_endpointr r0r,)rr>r r?r@rAr#r"rrrrr9r:r;r<r=r2r5s r6r8r8Hsfn 5 !C,, ## %+3!-H$HHX\\zChf endpoints catalog deploy --repo meta-llama/Llama-3.2-1B-InstructzPThe hardware accelerator to be used for inference (e.g. 'cpu', 'gpu', 'neuron').c0t|} |j|||||}tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)z4Deploy an Inference Endpoint from the Model Catalog.r)repo_idrr?rrzDeployment failed: rN) r &create_inference_endpoint_from_catalogrr r'r(r)r*r+r0r,) r>rr?rrrr2r5r's r6deploy_from_catalogrHs( 5 !C E==# > HHX\\ E 'w/0jjenn889uDEA B ABBct|} |j|}tjd|iy#t$rH}tjd|t j |jj|d}~wwxYw)zList available Catalog models.rzCatalog fetch failed: rNmodels) r list_inference_catalogrr r'r(r)r*r+r0)rrr2rKr's r6 list_catalogrMs} 5 !CE++%+8 HHh  E *5'23jjenn889uDEs8 B ABB zhf endpoints catalog lsz list-catalogT)rhiddencatalogrz!hf endpoints describe my-endpointc,t|} |j|||}tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)z+Get information about an existing endpoint.rrrrzFetch failed: rN) r get_inference_endpointrr r'r(r)r*r+r0r,rrrrr2r5r's r6describerUs 5 !CE--49TY-Z HHX\\ E N5'*+jjenn889uDEA B ABBz/hf endpoints update my-endpoint --min-replica 2zBThe machine learning framework used for the model (e.g. 'custom').revisionzrThe specific model revision to deploy on the Inference Endpoint (e.g. '6c0e6080953db56375760c0471a8c5f2929baf11').cDt|} |j|||||||||| | | | | |}tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)zUpdate an existing endpoint.r)rrrr rWrr?r@rAr9r:r;r<r=rzUpdate failed: rN) r update_inference_endpointrr r'r(r)r*r+r0r,)rrr>r?r@rAr rWrr9r:r;r<r=rrr2r5r's r6updaterZs` 5 !CE00#''##"7)/1 (HHX\\ E OE7+,jjenn889uDEs A BABBzhf endpoints delete my-endpointFyesz--yeszSkip confirmation prompts.cZtjd|d|t|} |j|||tjd |d | y#t$rH}tj d|t j|jj|d}~wwxYw) z)Delete an Inference Endpoint permanently.zDelete endpoint 'z'?)r[rrRzDelete failed: rNz Deleted 'z'.rP) r confirmr delete_inference_endpointrr'r(r)r*r+result)rrr[rrr2r's r6deleter`jsKK#D6,#6 5 !CE %%49E%R JJ4&#$/ E OE7+,jjenn889uDEsA B*"AB%%B*zhf endpoints pause my-endpointc,t|} |j|||}tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)zPause an Inference Endpoint.rrRzPause failed: rN) r pause_inference_endpointrr r'r(r)r*r+r0r,rTs r6pausercs 5 !CE//TYV[/\ HHX\\ E N5'*+jjenn889uDErVzhf endpoints resume my-endpointfail_if_already_runningz--fail-if-already-runningzWIf `True`, the method will raise an error if the Inference Endpoint is already running.c0t|} |j|||| }tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)zResume an Inference Endpoint.r)rrr running_okzResume failed: rN) r resume_inference_endpointrr r'r(r)r*r+r0r,)rrrdrrr2r5r's r6resumerhs 5 !C E0022 1 HHX\\ E OE7+,jjenn889uDErIz&hf endpoints scale-to-zero my-endpointc,t|} |j|||}tj|jy#t$rH}tjd|t j |jj|d}~wwxYw)z$Scale an Inference Endpoint to zero.rrRzScale To Zero failed: rN) r scale_to_zero_inference_endpointrr r'r(r)r*r+r0r,rTs r6 scale_to_zerorks 5 !CE77TY^c7d HHX\\ E *5'23jjenn889uDErV)(__doc__typingrr($huggingface_hub._inference_endpointsrhuggingface_hub.errorsr _cli_utilsrrr r _outputr r ie_cli catalog_appstrArgumentNameArgOptionNameOpt NamespaceOptcommandautor7intfloatr8rHrM add_typerrUrZboolr`rcrhrkrDr6rs7 8 O1NN. F G!QR  ENN()+  $JELL&')  $JELL n  '8:^&_`" 4 9 9 & &  &  & &a &FX)n(op\# 4 9 9 gi i   x   i  S  i  R  i(  a  )i4  l  5i@   g   AiL   i   MiZ[i\  d  V   ]ih iij kil  i  mix  e  yiD% d  Y  EiP&- 4  Qi\!   x  ]ih iiqiX (.s-tu " 4 9 9!    x       d  c       ! " # v H!5 9 9 !  !  !  !L 0I/JKLY0N40>9-=>?# 4 9 9       @"KLM# 4 9 9Ye ee  d  x   e d  R  e d  a  e* d  l  +e6 d  U  7eB d  F  CeN  d  V   OeZ d  i  [ef d  e  ger% d  Y  se~&- 4  eJ!   x  KeV WeX YeZ [eNeP;<=#  4 9 90 00   W#?@ B 0 0 0 0>0.:;<# 4 9 9       =";<=#  4 9 9 '  'j     >8BCD# 4 9 9       ErD