o i8&@sndZddlZddlmZmZmZddlZddlmZm Z ddl m Z ddl m Z mZmZmZmZmZmZedd Zed d Zeeejd d fZeeeejd d fZeeeejd d fZd eddfddZejddgddejddfdede dededdf ddZejddgddddddddddded eeejd!d fd"eeejd#d fd$eeejd%d fd&eeejd'd fd(eeejd)d fd*eeejd+d fd,eeejd-d fded.eeeejd/d fded0ee ejd1d fd2ee ejd3d fd4eee ejd5d fd6eee ejd7d fd8eee!ejd9d fddf"d:dZ"ejdd;gd   dad eeejd!d fdedededdf dd?Z$ejdd@gde$ejdAdBdCe$ej%edDdEejdFgd  dcdedededdfdGdHZ&ejdIgd              dddeded eeeejd!d fd$eeeejd%d fd&eeeejd'd fd(eeeejd)d fd"eeeejdJd fdKeeeejdLd fd.eeeejd/d fd0eee ejd1d fd2eee ejd3d fd4eee ejd5d fd6eee ejd7d fd8eee!ejd9d fdeddf dMdNZ'ejdOgd   dedededPee(ejdQdRd fdeddf dSdTZ)ejdUgd  dcdedededdfdVdWZ*ejdXgd   dedededYee(ejdZd[d fdeddf d\d]Z+ejd^gd  dcdedededdfd_d`Z,dS)fz2CLI commands for Hugging Face Inference Endpoints.N) AnnotatedAnyOptional)InferenceEndpointInferenceEndpointScalingMetric)HfHubHTTPError) FormatOpt OutputFormatQuietOptTokenOpt get_hf_apiprint_list_output typer_factoryz(Manage Hugging Face Inference Endpoints.)helpz.Interact with the Inference Endpoints catalog.zEndpoint name.z_The namespace associated with the Inference Endpoint. Defaults to the current user's namespace.endpointreturncCsttj|jddddS)NTindent sort_keys)typerechojsondumpsraw)rri/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/huggingface_hub/cli/inference_endpoints.py_print_endpoint,srzhf endpoints lsz"hf endpoints ls --namespace my-org)examplesF namespaceformatquiettokenc Cst|d}z |j||d}Wnty+}ztd|tj|jjd|d}~wwdd|D}dtt t fd t t fd d }t |||d gd |ddS)z6Lists all Inference Endpoints for the given namespace.r#)r r#zListing failed: codeNcSsg|]}|jqSr)r).0rrrr ?szls..itemrc Ss |di}|di}|di}|di}t|ddtt|tr+|ddndtt|tr9|ddndtt|trG|d dndtt|trU|d dndtt|trc|d dndtt|trq|d dndtt|tr|d dgSdgS)Nstatusmodelcomputeprovidername repositorystatetask framework instanceTypevendorregion)getstr isinstancedict)r)r*r+r,r-rrrrow_fnAs    zls..row_fnr.)NAMEMODELSTATUSTASK FRAMEWORKINSTANCEVENDORREGION)itemsr!r"id_keyheadersr;) r list_inference_endpointsrrrExitresponse status_coder:r8rlistr) r r!r"r#api endpointserrorresultsr;rrrls0s$  rPdeployzChf endpoints deploy my-endpoint --repo gpt2 --framework pytorch ...)r.r)r r2r# min_replica max_replicascale_to_zero_timeoutscaling_metricscaling_thresholdr.repozeThe name of the model repository associated with the Inference Endpoint (e.g. 'openai/gpt-oss-120b').r3z@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').r6zTThe cloud region in which the Inference Endpoint will be created (e.g. 'us-east-1').r5zVThe cloud provider or vendor where the Inference Endpoint will be hosted (e.g. 'aws').r2zCThe task on which to deploy the model (e.g. 'text-classification').rRzVThe minimum number of replicas (instances) to keep running for the Inference Endpoint.rSzRThe maximum number of replicas (instances) to scale to for the Inference Endpoint.rTzFThe duration in minutes before an inactive endpoint is scaled to zero.rUz!The metric reference for scaling.rVzeThe scaling metric threshold used to trigger a scale up. Ignored when scaling metric is not provided.cCst| d}|jdid|d|d|d|d|d|d|d |d |d | d | d | d| d|d|d| }t|dS)z3Deploy an Inference Endpoint from a Hub repository.r$r.r0r3rXrYrZr6r5r r2r#rRrSrUrVrTNr)r create_inference_endpointr)r.rWr3rXrYrZr6r5r r2r#rRrSrTrUrVrLrrrrrQ[sF V       zChf endpoints catalog deploy --repo meta-llama/Llama-3.2-1B-Instructc Csht|d}z |j||||d}Wnty-}ztd|tj|jjd|d}~wwt|dS)z4Deploy an Inference Endpoint from the Model Catalog.r$)repo_idr.r r#zDeployment failed: r%N) r &create_inference_endpoint_from_catalogrrrrHrIrJr)rWr.r r#rLrrNrrrdeploy_from_catalogs   r^c Cstt|d}z|j|d}Wnty*}ztd|tj|jjd|d}~wwttj d|iddddS) zList available Catalog models.r$zCatalog fetch failed: r%NmodelsrTr) r list_inference_catalogrrrrHrIrJrr)r#rLr_rNrrr list_catalogs razhf endpoints catalog lsz list-catalogT)r.hiddencatalog)r.z!hf endpoints describe my-endpointc Cft|d}z |j|||d}Wnty,}ztd|tj|jjd|d}~wwt|dS)z+Get information about an existing endpoint.r$r.r r#zFetch failed: r%N) r get_inference_endpointrrrrHrIrJrr.r r#rLrrNrrrdescribe  rhz/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').cCs~t|d}z|j|||||||||| | | | | |d}Wnty8}ztd|tj|jjd|d}~wwt|dS)zUpdate an existing endpoint.r$)r.r r0r3rjr2rXrYrZrRrSrTrUrVr#zUpdate failed: r%N) r update_inference_endpointrrrrHrIrJr)r.r rWrXrYrZr3rjr2rRrSrTrUrVr#rLrrNrrrupdate s2 O  rlzhf endpoints delete my-endpointyesz--yeszSkip confirmation prompts.c Cs|std|d}||krtdtjddt|d}z |j|||dWntyF}ztd|tj|jjd|d }~wwtd |d d S) z)Delete an Inference Endpoint permanently.zDelete endpoint 'z'? Type the name to confirm.zAborted.rr%r$rezDelete failed: Nz Deleted 'z'.) rpromptrrHr delete_inference_endpointrrIrJ)r.r rmr# confirmationrLrNrrrdeleters    rqzhf endpoints pause my-endpointc Crd)zPause an Inference Endpoint.r$rezPause failed: r%N) r pause_inference_endpointrrrrHrIrJrrgrrrpauserirszhf endpoints resume my-endpointfail_if_already_runningz--fail-if-already-runningzWIf `True`, the method will raise an error if the Inference Endpoint is already running.c Csjt|d}z |j|||| d}Wnty.}ztd|tj|jjd|d}~wwt|dS)zResume an Inference Endpoint.r$)r.r r# running_okzResume failed: r%N) r resume_inference_endpointrrrrHrIrJr)r.r rtr#rLrrNrrrresumes   rwz&hf endpoints scale-to-zero my-endpointc Crd)z$Scale an Inference Endpoint to zero.r$rezScale To Zero failed: r%N) r scale_to_zero_inference_endpointrrrrHrIrJrrgrrr scale_to_zeroriry)NNN)N)NN)NNNNNNNNNNNNNN)NFN)-__doc__rtypingrrrr$huggingface_hub._inference_endpointsrrhuggingface_hub.errorsr _cli_utilsr r r r r rrie_cli catalog_appr8ArgumentNameArgOptionNameOpt NamespaceOptrcommandtablerPintfloatrQr^ra add_typerrhrlboolrqrsrwryrrrrs $    *. &-.45;AGMS l      !'-39?EKL g