+
    QV-jØ6  ã                   ó  € R t ^ RIHt ^ RIt^ RIHt ^ RIHt ^RIH	t	H
t
Ht ^RIHt ]! RR7      t]! R	R7      t]]]P"                  ! R
R7      3,          t]]R,          ]P&                  ! R
R7      3,          t]]R,          ]P&                  ! RR7      3,          t]P-                  RRR.R7      R<R R ll4       t]P-                  RR.R7      RRRRRRR^R^RRRRRR/R R ll4       t]P-                  RR.R7      R=R  R! ll4       tR>R" R# llt]P-                  RR$.R7      ! ]4       ]P-                  R%R&R'7      ! ]4       ]P7                  ]R(R)7       ]P-                  R*.R7      R<R+ R, ll4       t]P-                  R-.R7      R?R. R/ ll4       t]P-                  R0.R7      R@R1 R2 ll4       t]P-                  R3.R7      R<R4 R5 ll4       t]P-                  R6.R7      R@R7 R8 ll4       t ]P-                  R9.R7      R<R: R; ll4       t!R# )Az2CLI commands for Hugging Face Inference Endpoints.)Ú	AnnotatedN)ÚInferenceEndpointScalingMetric)ÚHfHubHTTPError)ÚTokenOptÚ
get_hf_apiÚtyper_factory)Úoutz(Manage Hugging Face Inference Endpoints.©Úhelpz.Interact with the Inference Endpoints catalog.zEndpoint name.z_The namespace associated with the Inference Endpoint. Defaults to the current user's namespace.z	list | lszhf endpoints lsz"hf endpoints ls --namespace my-org)Úexamplesc                ó4   € V ^8„  d   QhR\         R\        RR/# )é   Ú	namespaceÚtokenÚreturnN)ÚNamespaceOptr   )Úformats   "Úx/Volumes/fast/ai/experiments/ui-tars-smoke/.venv/lib/python3.14/site-packages/huggingface_hub/cli/inference_endpoints.pyÚ__annotate__r   %   s&   € ÷ &ñ &Üð&äð&ð 
ñ&ó    c                óx  € \        VR7      p VP                  WR7      p. pT EF¡  pTP                  pTP                  R/ 4      pTP                  R/ 4      p	TP                  R/ 4      p
TP                  R	/ 4      pTP                  R
TP                  R
R4      R\        T	\        4      '       d   T	P                  RR4      MRR\        T\        4      '       d   TP                  RR4      MRR\        T	\        4      '       d   T	P                  RR4      MRR\        T	\        4      '       d   T	P                  RR4      MRR\        T
\        4      '       d   T
P                  RR4      MRR\        T\        4      '       d   TP                  RR4      MRR\        T\        4      '       d   TP                  RR4      MR/4       EK¤  	  \        P                  ! TR
R7       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)z6Lists all Inference Endpoints for the given namespace.©r   )r   r   zListing failed: ©ÚcodeNÚstatusÚmodelÚcomputeÚproviderÚnameÚ Ú
repositoryÚstateÚtaskÚ	frameworkÚinstanceÚinstanceTypeÚvendorÚregion)Úid_key)r   Úlist_inference_endpointsr   r   ÚerrorÚtyperÚExitÚresponseÚstatus_codeÚrawÚgetÚappendÚ
isinstanceÚdictÚtable)r   r   ÚapiÚ	endpointsr*   ÚresultsÚendpointr/   r   r   r   r   s   &&          r   Úlsr9   $   sÎ  € ô ˜5Ô
!€CðEØ×0Ñ0¸9Ð0ÓRˆ	ð
 €GÜˆØ�l‰lˆØ—‘˜ 2Ó&ˆØ—‘˜ Ó$ˆØ—'‘'˜) RÓ(ˆØ—7‘7˜: rÓ*ˆØ�‰à˜Ÿ™ ¨Ó+Ø¼
À5Ì$×8OÒ8O˜Ÿ™ <°Ô4ÐUWØ´ZÀÌ×5MÒ5M˜&Ÿ*™* W¨bÔ1ÐSUØ´¸EÄ4×1HÒ1H˜Ÿ	™	 &¨"Ô-ÈbØ¼:ÀeÌT×;RÒ;R˜UŸY™Y {°BÔ7ÐXZØ¼zÈ'ÔSW×?XÒ?X˜GŸK™K¨¸Ô;Ð^`Ø¼
À8ÌT×8RÒ8R˜(Ÿ,™, x°Ô4ÐXZØ¼
À8ÌT×8RÒ8R˜(Ÿ,™, x°Ô4ÐXZð	÷	
ñ ô$ ‡I‚Iˆg˜f×%øô/ ô EÜ�	Š	Ð$ U GÐ,Ô-Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   ŽG$ Ç$H9Ç/AH4È4H9ÚdeployzChf endpoints deploy my-endpoint --repo gpt2 --framework pytorch ...)r   r   r   r"   r   Úmin_replicaÚmax_replicaÚscale_to_zero_timeoutÚscaling_metricÚscaling_thresholdc          %      óˆ  € V ^8„  d   QhR\         R\        \        \        P                  ! RR7      3,          R\        \        \        P                  ! RR7      3,          R\        \        \        P                  ! RR7      3,          R	\        \        \        P                  ! R
R7      3,          R\        \        \        P                  ! RR7      3,          R\        \        \        P                  ! RR7      3,          R\        \        \        P                  ! RR7      3,          R\
        R\        \        R,          \        P                  ! RR7      3,          R\        R\        \        \        P                  ! RR7      3,          R\        \        \        P                  ! RR7      3,          R\        \        R,          \        P                  ! RR7      3,          R\        \        R,          \        P                  ! RR7      3,          R\        \        R,          \        P                  ! RR7      3,          R R/# )!r   r   ÚrepoúeThe name of the model repository associated with the Inference Endpoint (e.g. 'openai/gpt-oss-120b').r	   r#   z@The machine learning framework used for the model (e.g. 'vllm').Úacceleratorú?The hardware accelerator to be used for inference (e.g. 'cpu').Úinstance_sizeúNThe size or type of the instance to be used for hosting the model (e.g. 'x4').Úinstance_typeúYThe 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').r   r"   NúCThe task on which to deploy the model (e.g. 'text-classification').r   r;   úVThe minimum number of replicas (instances) to keep running for the Inference Endpoint.r<   úRThe maximum number of replicas (instances) to scale to for the Inference Endpoint.r=   úFThe duration in minutes before an inactive endpoint is scaled to zero.r>   ú!The metric reference for scaling.r?   úeThe scaling metric threshold used to trigger a scale up. Ignored when scaling metric is not provided.r   )
ÚNameArgr   Ústrr+   ÚOptionr   r   Úintr   Úfloat)r   s   "r   r   r   H   sX  € ÷ hñ hÜ
ðhä
ÜÜ�ŠØxô	
ð	õðhô ÜÜ�ŠØSô	
ð	õðhô ÜÜ�ŠØRô	
ð	õðhô( ÜÜ�ŠØaô	
ð	õð)hô4 ÜÜ�ŠØlô	
ð	õð5hô@ ÜÜ�ŠØgô	
ð	õðAhôL ÜÜ�ŠØiô	
ð	õðMhôZ ð[hô\ Üˆd�
Ü�ŠØVô	
ð	õð]hôh ðihôj ÜÜ�ŠØiô	
ð	õðkhôv ÜÜ�ŠØeô	
ð	õðwhôB %Üˆd�
Ü�ŠØYô	
ð	õðChôN Ü&¨Õ-Ü�ŠØ4ô	
ð	õðOhôZ !Ü��Ü�ŠØxô	
ð	õð[hðf 
ñghr   c               ó    € \        V
R7      pVP                  V VVVVVVVVV	V
VVVVVR7      p\        P                  ! VP                  4       R# )z3Deploy an Inference Endpoint from a Hub repository.r   )r   r    r#   rC   rE   rG   r'   r&   r   r"   r   r;   r<   r>   r?   r=   N)r   Úcreate_inference_endpointr   r3   r/   )r   rA   r#   rC   rE   rG   r'   r&   r   r"   r   r;   r<   r=   r>   r?   r5   r8   s   &&&&&&&&$$$$$$$$  r   r:   r:   G   sf   € ôl ˜5Ô
!€CØ×,Ñ,ØØØØØ#Ø#ØØØØØØØØ%Ø+Ø3ð! -ó €Hô$ ‡H‚HˆX�\‰\Ör   zChf endpoints catalog deploy --repo meta-llama/Llama-3.2-1B-Instructc                óî   € V ^8„  d   QhR\         \        \        P                  ! RR7      3,          R\        R\         \        R,          \        P                  ! RR7      3,          R\
        R	\        R
R/# )r   rA   rB   r	   r   rC   NzPThe hardware accelerator to be used for inference (e.g. 'cpu', 'gpu', 'neuron').r   r   r   )r   rP   r+   rQ   ÚNameOptr   r   )r   s   "r   r   r   ´   s‚   € ÷ ñ Ü
ÜÜ�ŠØxô	
ð	õðô ðô Üˆd�
Ü�ŠØcô	
ð	õðô ðô ðð  
ñ!r   c                ó<  € \        VR7      p VP                  V VVVVR7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)z4Deploy an Inference Endpoint from the Model Catalog.r   )Úrepo_idr   rC   r   r   zDeployment failed: r   N)r   Ú&create_inference_endpoint_from_catalogr   r   r*   r+   r,   r-   r.   r3   r/   )rA   r   rC   r   r   r5   r8   r*   s   &&&&&   r   Údeploy_from_catalogr[   ³   s�   € ô& ˜5Ô
!€Cð
EØ×=Ñ=ØØØ#ØØð >ó 
ˆô ‡H‚HˆX�\‰\Öøô	 ô EÜ�	Š	Ð'¨ wÐ/Ô0Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   ŽA ÁBÁABÂBc                ó(   € V ^8„  d   QhR\         RR/# )r   r   r   N)r   )r   s   "r   r   r   Ö   s   € ÷ !ñ !Üð!à	ñ!r   c                ó$  € \        V R7      p VP                  V R7      p\        P                  ! RT/4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)zList available Catalog models.r   zCatalog fetch failed: r   NÚmodels)
r   Úlist_inference_catalogr   r   r*   r+   r,   r-   r.   r3   )r   r5   r^   r*   s   &   r   Úlist_catalogr`   Ö   s}   € ô ˜5Ô
!€CðEØ×+Ñ+°%Ð+Ó8ˆô
 ‡H‚Hˆh˜ÐÖ øô	 ô EÜ�	Š	Ð*¨5¨'Ð2Ô3Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   Ž: ºBÁAB
Â
Bzhf endpoints catalog lszlist-catalogT)r   ÚhiddenÚcatalog©r   z!hf endpoints describe my-endpointc                ó@   € V ^8„  d   QhR\         R\        R\        RR/# ©r   r   r   r   r   N©rO   r   r   )r   s   "r   r   r   ì   ó0   € ÷ ñ Ü
ðäðô ðð 
ñ	r   c                ó6  € \        VR7      p VP                  WVR7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)z+Get information about an existing endpoint.r   ©r   r   r   zFetch failed: r   N)r   Úget_inference_endpointr   r   r*   r+   r,   r-   r.   r3   r/   ©r   r   r   r5   r8   r*   s   &&&   r   Údescriberl   ë   s~   € ô ˜5Ô
!€CðEØ×-Ñ-°4ÐTYÐ-ÓZˆô
 ‡H‚HˆX�\‰\Öøô	 ô EÜ�	Š	�N 5 'Ð*Ô+Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEúó   ŽA ÁBÁABÂBz/hf endpoints update my-endpoint --min-replica 2c          !      ó¨  € V ^8„  d   QhR\         R\        R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R	\        \        R,          \        P
                  ! R
R7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        \        R,          \        P
                  ! RR7      3,          R\        RR/# )r   r   r   rA   NrB   r	   rC   rD   rE   rF   rG   rH   r#   zBThe machine learning framework used for the model (e.g. 'custom').ÚrevisionzrThe specific model revision to deploy on the Inference Endpoint (e.g. '6c0e6080953db56375760c0471a8c5f2929baf11').r"   rI   r;   rJ   r<   rK   r=   rL   r>   rM   r?   rN   r   r   )
rO   r   r   rP   r+   rQ   rR   r   rS   r   )r   s   "r   r   r   ý   sP  € ÷ dñ dÜ
ðdäðdô Üˆd�
Ü�ŠØxô	
ð	õðdô Üˆd�
Ü�ŠØRô	
ð	õðdô Üˆd�
Ü�ŠØaô	
ð	õðdô* Üˆd�
Ü�ŠØlô	
ð	õð+dô6 Üˆd�
Ü�ŠØUô	
ð	õð7dôB Üˆd�
Ü�Šð Fô	
ð	õðCdôN Üˆd�
Ü�ŠØVô	
ð	õðOdôZ Üˆd�
Ü�ŠØiô	
ð	õð[dôf Üˆd�
Ü�ŠØeô	
ð	õðgdôr %Üˆd�
Ü�ŠØYô	
ð	õðsdô~ Ü&¨Õ-Ü�ŠØ4ô	
ð	õðdôJ !Ü��Ü�ŠØxô	
ð	õðKdôV ðWdðX 
ñYdr   c                óP  € \        VR7      p VP                  V VVVVVVVVV	V
VVVVR7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)zUpdate an existing endpoint.r   )r   r   r    r#   ro   r"   rC   rE   rG   r;   r<   r=   r>   r?   r   zUpdate failed: r   N)r   Úupdate_inference_endpointr   r   r*   r+   r,   r-   r.   r3   r/   )r   r   rA   rC   rE   rG   r#   ro   r"   r;   r<   r=   r>   r?   r   r5   r8   r*   s   &&&&&&&&&&&&&&&   r   Úupdaterr   ü   s«   € ô^ ˜5Ô
!€CðEØ×0Ñ0ØØØØØØØ#Ø'Ø'Ø#Ø#Ø"7Ø)Ø/Øð 1ó 
ˆô( ‡H‚HˆX�\‰\Öøô ô EÜ�	Š	�O E 7Ð+Ô,Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   Ž A ÁB%ÁAB Â B%zhf endpoints delete my-endpointc                ó’   € V ^8„  d   QhR\         R\        R\        \        \        P
                  ! RRR7      3,          R\        RR	/# )
r   r   r   Úyesz--yeszSkip confirmation prompts.r	   r   r   N©rO   r   r   Úboolr+   rQ   r   )r   s   "r   r   r   e  sW   € ÷ 0ñ 0Ü
ð0äð0ô 
ÜÜ�Š�WÐ#?Ô@ð	Bõ
ð0ô ð0ð 
ñ0r   c                óf  € \         P                  ! RV  R2VR7       \        VR7      p VP                  WVR7       \         P                  ! R	T  R
2T R7       R#   \         dK   p\         P
                  ! RT 24       \        P                  ! TP                  P                  R7      ThRp?ii ; i)z)Delete an Inference Endpoint permanently.zDelete endpoint 'z'?)rt   r   ri   zDelete failed: r   Nz	Deleted 'z'.rc   )r   Úconfirmr   Údelete_inference_endpointr   r*   r+   r,   r-   r.   Úresult)r   r   rt   r   r5   r*   s   &&&&  r   Údeleter{   d  sš   € ô ‡K‚KÐ# D 6¨Ð,°#Õ6ä
˜5Ô
!€CðEØ×%Ñ%¨4ÈEÐ%ÔRô
 ‡J‚J�˜4˜& Ð#¨$×/øô	 ô EÜ�	Š	�O E 7Ð+Ô,Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   ªA ÁB0Á&AB+Â+B0zhf endpoints pause my-endpointc                ó@   € V ^8„  d   QhR\         R\        R\        RR/# re   rf   )r   s   "r   r   r   |  rg   r   c                ó6  € \        VR7      p VP                  WVR7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)zPause an Inference Endpoint.r   ri   zPause failed: r   N)r   Úpause_inference_endpointr   r   r*   r+   r,   r-   r.   r3   r/   rk   s   &&&   r   Úpauser   {  s~   € ô ˜5Ô
!€CðEØ×/Ñ/°TÐV[Ð/Ó\ˆô
 ‡H‚HˆX�\‰\Öøô	 ô EÜ�	Š	�N 5 'Ð*Ô+Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEúrm   zhf endpoints resume my-endpointc                ó’   € V ^8„  d   QhR\         R\        R\        \        \        P
                  ! RRR7      3,          R\        RR	/# )
r   r   r   Úfail_if_already_runningz--fail-if-already-runningzWIf `True`, the method will raise an error if the Inference Endpoint is already running.r	   r   r   Nru   )r   s   "r   r   r   �  sY   € ÷ ñ Ü
ðäðô 'ÜÜ�ŠØ'Øjô	
ð	õðô ðð 
ñr   c                óD  € \        VR7      p VP                  V VVV'       * R7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)zResume an Inference Endpoint.r   )r   r   r   Ú
running_okzResume failed: r   N)r   Úresume_inference_endpointr   r   r*   r+   r,   r-   r.   r3   r/   )r   r   r�   r   r5   r8   r*   s   &&&&   r   Úresumer…   Œ  sŒ   € ô ˜5Ô
!€Cð	EØ×0Ñ0ØØØØ2Ô2ð	 1ó 
ˆô ‡H‚HˆX�\‰\Öøô ô EÜ�	Š	�O E 7Ð+Ô,Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEús   ŽA
 Á
BÁABÂBz&hf endpoints scale-to-zero my-endpointc                ó@   € V ^8„  d   QhR\         R\        R\        RR/# re   rf   )r   s   "r   r   r   ©  rg   r   c                ó6  € \        VR7      p VP                  WVR7      p\        P                  ! TP                  4       R#   \         dK   p\        P                  ! RT 24       \
        P                  ! TP                  P                  R7      ThRp?ii ; i)z$Scale an Inference Endpoint to zero.r   ri   zScale To Zero failed: r   N)r   Ú scale_to_zero_inference_endpointr   r   r*   r+   r,   r-   r.   r3   r/   rk   s   &&&   r   Úscale_to_zeror‰   ¨  s   € ô ˜5Ô
!€CðEØ×7Ñ7¸TÐ^cÐ7Ódˆô
 ‡H‚HˆX�\‰\Öøô	 ô EÜ�	Š	Ð*¨5¨'Ð2Ô3Ü�jŠj˜eŸn™n×8Ñ8Ô9¸uÐDûðEúrm   )NN)NNNN)N)NNNNNNNNNNNNNN)NFN)"Ú__doc__Útypingr   r+   Ú$huggingface_hub._inference_endpointsr   Úhuggingface_hub.errorsr   Ú
_cli_utilsr   r   r   Ú_outputr   Úie_cliÚcatalog_apprP   ÚArgumentrO   rQ   rW   r   Úcommandr9   r:   r[   r`   Ú	add_typerrl   rr   r{   r   r…   r‰   © r   r   Ú<module>r–      s¬  ðÙ 8å ã å OÝ 1ç ;Ñ ;Ý ñ 
ÐFÔ	G€áÐ!QÔR€ð ØØ	‡N‚NÐ(Ô)ð+õ€ð Øˆ$…JØ	‡L‚LÐ&Ô'ð)õ€ð
 Øˆ$…JØ	‡L‚LØnôðõ€ð ‡��Ð'8Ð:^Ð&_€Ó`ö&ó að&ðD ‡��XÐ)nÐ(o€ÓpðhðZ #ð[hðf 	ðghðh ðihðt 	
ðuhð@ 	
ðAhðL 	ðMhðX 	ðYhðd 	öehó qðhðV ×Ñ˜(Ð.sÐ-tÐÓuöó vð÷D!ð × Ñ ˜Ð0IÐ/JÐ Ô KÈLÔ YØ ‡��N¨4€Ô 0°Ô >ð × Ñ � 9Ð Ô -ð ‡�Ð=Ð>€Ó?öó @ðð  ‡�ÐKÐL€ÓMödó NðdðN ‡�Ð;Ð<€Ó=ö0ó >ð0ð, ‡�Ð:Ð;€Ó<öó =ðð  ‡�Ð;Ð<€Ó=öó >ðð6 ‡�ÐBÐC€ÓDöó Eòr   