Zi9& dZddlZddlmZmZmZddlZddlmZm Z ddl m Z ddl m Z mZmZmZmZmZmZed Zed Zeeejd fZeeeejd fZeeeejd fZd eddfdZeddgdejddfdede dededdf dZeddgddddddddddedeeejd fd eeejd! fd"eeejd# fd$eeejd% fd&eeejd' fd(eeejd) fd*eeejd+ fded,eeeejd- fded.ee ejd/ fd0ee ejd1 fd2eee ejd3 fd4eee ejd5 fd6eee!ejd7 fddf"d8Z"edd9g dYdeeejd fded"eeeejd: fdededdf d;Z# dZdeddfd<Z$ed=d>ge$ed?d@Ae$e%edBCedDg d[dedededdfdEZ&edFg d\dededeeeejd fd"eeeejd# fd$eeeejd% fd&eeeejd' fd eeeejdG fdHeeeejdI fd,eeeejd- fd.eee ejd/ fd0eee ejd1 fd2eee ejd3 fd4eee ejd5 fd6eee!ejd7 fdeddf dJZ'edKg d]dededLee(ejdMdN fdeddf dOZ)edPg d[dedededdfdQZ*edRg d]dededSee(ejdTdU fdeddf dVZ+edWg d[dedededdfdXZ,dS)^z2CLI commands for Hugging Face Inference Endpoints.N) AnnotatedAnyOptional)InferenceEndpointInferenceEndpointScalingMetric)HfHubHTTPError) FormatOpt OutputFormatQuietOptTokenOpt get_hf_apiprint_list_output typer_factoryz(Manage Hugging Face Inference Endpoints.)helpz.Interact with the Inference Endpoints catalog.zEndpoint name.z_The namespace associated with the Inference Endpoint. Defaults to the current user's namespace.endpointreturncbtjtj|jdddS)NTindent sort_keys)typerechojsondumpsraw)rs b/root/projects/butler/venv/lib/python3.11/site-packages/huggingface_hub/cli/inference_endpoints.py_print_endpointr,s+ Jtz(,qDAAABBBBBzhf endpoints lsz"hf endpoints ls --namespace my-org)examplesF namespaceformatquiettokenct|} |||}nI#t$r<}tjd|tj|jj|d}~wwxYwd|D}dtttfdttfd }t|||d gd | dS) z6Lists all Inference Endpoints for the given namespace.r%)r"r%zListing failed: codeNcg|] }|j S)r).0rs r zls..?s666x|666r itemrc |di}|di}|di}|di}t|ddtt|tr|ddndtt|tr|ddndtt|tr|d dndtt|tr|d dndtt|tr|d dndtt|tr|d dndtt|tr|d dndgS)Nstatusmodelcomputeprovidername repositorystatetask framework instanceTypevendorregion)getstr isinstancedict)r.r0r1r2r3s rrow_fnzls..row_fnAs(B''"%%((9b))88J++ $$ % % z%/F/FN ,+++B O O :fd+C+CK 7B''' L L E4)@)@H &"%%%b I I j.E.EM +r***2 N N :gt3L3LT NB///RT U U j4.H.HP Xr***b Q Q j4.H.HP Xr***b Q Q  r r4)NAMEMODELSTATUSTASK FRAMEWORKINSTANCEVENDORREGION)itemsr#r$id_keyheadersrA) rlist_inference_endpointsrrrExitresponse status_coder@r>rlistr) r"r#r$r%api endpointserrorresultsrAs rlsrV0s 5 ! ! !CE009E0RR EEE -e--...jen8999uDE76I666G T#s(^ S     ``` s* A07A++A0deployzChf endpoints deploy my-endpoint --repo gpt2 --framework pytorch ...)r4r!)r"r8r% min_replica max_replicascale_to_zero_timeoutscaling_metricscaling_thresholdr4repozeThe name of the model repository associated with the Inference Endpoint (e.g. 'openai/gpt-oss-120b').r9z@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').r8zCThe task on which to deploy the model (e.g. 'text-classification').rXzVThe minimum number of replicas (instances) to keep running for the Inference Endpoint.rYzRThe maximum number of replicas (instances) to scale to for the Inference Endpoint.rZzFThe 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.ct| }||||||||||| | | | ||| }t|dS)z3Deploy an Inference Endpoint from a Hub repository.r')r4r6r9r^r_r`r<r;r"r8r%rXrYr[r\rZN)rcreate_inference_endpointr)r4r]r9r^r_r`r<r;r"r8r%rXrYrZr[r\rRrs rrWrW[sul 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|} ||||||}nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)z4Deploy an Inference Endpoint from the Model Catalog.r')repo_idr4r^r"r%zDeployment failed: r(N) r&create_inference_endpoint_from_catalogrrrrNrOrPr)r]r4r^r"r%rRrrTs rdeploy_from_catalogrfs& 5 ! ! !C E==# >   EEE 000111jen8999uDEH- A37A..A3c>t|} ||}nI#t$r<}tjd|tj|jj|d}~wwxYwtjtj d|idddS) zList available Catalog models.r'zCatalog fetch failed: r(NmodelsrTr) rlist_inference_catalogrrrrNrOrPrr)r%rRrirTs r list_catalogrks 5 ! ! !CE++%+88 EEE 3E33444jen8999uDE Jtz8V,Q$GGGHHHHHs) A/7A**A/rVzhf endpoints catalog lsz list-catalogT)r4hiddencatalog)r4z!hf endpoints describe my-endpointc t|} ||||}nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)z+Get information about an existing endpoint.r'r4r"r%zFetch failed: r(N) rget_inference_endpointrrrrNrOrPrr4r"r%rRrrTs rdescriberrs 5 ! ! !CE--49TY-ZZ EEE +E++,,,jen8999uDEH+ A17A,,A1z/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').c"t|} ||||||||||| | | | | |}nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)zUpdate an existing endpoint.r')r4r"r6r9rtr8r^r_r`rXrYrZr[r\r%zUpdate failed: r(N) rupdate_inference_endpointrrrrNrOrPr)r4r"r]r^r_r`r9rtr8rXrYrZr[r\r%rRrrTs rupdaterws^ 5 ! ! !CE00#''##"7)/1  " EEE ,U,,---jen8999uDEHs$7 A=7A88A=zhf endpoints delete my-endpointyesz--yeszSkip confirmation prompts.c|sGtjd|d}||kr)tjdtjdt |} ||||nI#t $r<}tjd|tj|jj|d }~wwxYwtjd |d d S) z)Delete an Inference Endpoint permanently.zDelete endpoint 'z'? Type the name to confirm.zAborted.rr(r'rozDelete failed: Nz Deleted 'z'.) rpromptrrNrdelete_inference_endpointrrOrP)r4r"rxr% confirmationrRrTs rdeleter}ys %|$Z$Z$Z$Z[[ 4   Jz " " "*!$$$ $ 5 ! ! !CE %%49E%RRRR EEE ,U,,---jen8999uDE J#4###$$$$$sA44 B:>7B55B:zhf endpoints pause my-endpointc t|} ||||}nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)zPause an Inference Endpoint.r'rozPause failed: r(N) rpause_inference_endpointrrrrNrOrPrrqs rpausers 5 ! ! !CE//TYV[/\\ EEE +E++,,,jen8999uDEHrszhf endpoints resume my-endpointfail_if_already_runningz--fail-if-already-runningzWIf `True`, the method will raise an error if the Inference Endpoint is already running.ct|} ||||| }nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)zResume an Inference Endpoint.r')r4r"r% running_okzResume failed: r(N) rresume_inference_endpointrrrrNrOrPr)r4r"rr%rRrrTs rresumers 5 ! ! !C E0022 1   EEE ,U,,---jen8999uDEHrgz&hf endpoints scale-to-zero my-endpointc t|} ||||}nI#t$r<}tjd|tj|jj|d}~wwxYwt|dS)z$Scale an Inference Endpoint to zero.r'rozScale To Zero failed: r(N) r scale_to_zero_inference_endpointrrrrNrOrPrrqs r scale_to_zerors 5 ! ! !CE77TY^c7dd EEE 3E33444jen8999uDEHrs)NNNN)N)NN)NNNNNNNNNNNNNN)NFN)-__doc__rtypingrrrr$huggingface_hub._inference_endpointsrrhuggingface_hub.errorsr _cli_utilsr r r r rrrie_cli catalog_appr>ArgumentNameArgOptionNameOpt NamespaceOptrcommandtablerVintfloatrWrfrk add_typerrrrwboolr}rrrr+r rrse 88 ++++++++++ bbbbbbbb111111 F G G Gm!QRRR  EN()))+   SMEL&''')   SMEL n C/CDCCCC+-QRSS"$* ''' ' '  '  '''TS'TX)n(opp\#  eiii 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 giiiqpiX (.s-tuu "   x        c       !vuF I I  I  I I I IE )B(CDD\RRR0N400>>>9---=>??#          @?  KLMM# Wdd 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 YdddNMdN;<==# %% %%   W#?@@@ B % % %%%>=%4:;<<#          =<  ;<==#  '  'j      >=6BCDD#          ED   r