Import a model into a workspace or a registry

Version: 0.0.39

Preview

View in Studio: https://ml.azure.com/registries/azureml/components/import_model/version/0.0.39

pipeline specific compute

Name	Description	Type	Default	Optional	Enum
compute	Common compute for model download, MLflow conversion and registration. eg. provide 'FT-Cluster' if your compute is named 'FT-Cluster'. Special characters like \ and ' are invalid in the parameter value. If compute name is provided, instance_type field will be ignored and the respective cluster will be used	string	serverless	True
instance_type	Instance type to be used for the component in case of serverless compute, eg. STANDARD_NC6s_v3. The parameter compute must be set to 'serverless' for instance_type to be used	string	STANDARD_NC6s_v3	True

Inputs for download model

Name	Description	Type	Default	Optional	Enum
model_source	Storage containers from where model will be sourced from	string	Huggingface		['AzureBlob', 'GIT', 'Huggingface']
model_id	A valid model id for the model source selected. For example you can specify bert-base-uncased for importing HuggingFace bert base uncased model. Please specify the complete URL if GIT or AzureBlob is selected in model_source	string
model_flavor	Flavor of MLFlow to which model the model is converted to.	string	HFTransformersV2	False	['HFTransformersV2', 'OSS']
model_framework	Framework from which model is imported from.	string	Huggingface	False	['Huggingface', 'MMLab', 'llava', 'AutoML']
vllm_enabled	Enable vllm in the converted model	boolean	False	False
token	If set use it to access the private models or authenticate the user. For example, user can get the token for HF private model by creating account in Huggingface, accept the condition for models that needs to be downloaded and create access token from browser. For more details please visit - https://huggingface.co/docs/hub/security-tokens	string		True

Inputs for the MlFLow conversion

Name	Description	Type	Default	Optional	Enum
license_file_path	Path to the license file	uri_file		True
task_name	A Hugging face task on which model was trained on	string		True	['chat-completion', 'fill-mask', 'token-classification', 'question-answering', 'summarization', 'text-generation', 'text2text-generation', 'text-classification', 'translation', 'image-classification', 'image-classification-multilabel', 'image-object-detection', 'image-instance-segmentation', 'image-to-text', 'text-to-image', 'text-to-image-inpainting', 'image-text-to-text', 'image-to-image', 'zero-shot-image-classification', 'mask-generation', 'video-multi-object-tracking', 'visual-question-answering']
hf_config_args	Provide args that should be used to load Huggingface model config. eg: trust_remote_code=True;	string		True
hf_tokenizer_args	Provide args that should be used to load Huggingface model tokenizer. eg: trust_remote_code=True, device_map=auto,	string		True
hf_model_args	Provide args that should be used to load Huggingface model. eg: trust_remote_code=True, device_map=auto, low_cpu_mem_usage=True	string		True
hf_pipeline_args	Provide pipeline args that should be used while loading the hugging face model. Dont use quotes. If value cannot be eval'ed it will be taken as as string. eg: trust_remote_code=True, device_map=auto	string		True
hf_config_class	AutoConfig class may not be sufficient to load config for some of the models. You can use this parameter to send Config class name as it is	string		True
hf_model_class	AutoModel classes may not be sufficient to load some of the models. You can use this parameter to send Model class name as it is	string		True
hf_tokenizer_class	AutoTokenizer class may not be sufficient to load tokenizer for some of the models. You can use this parameter to send Config class name as it is	string		True
hf_use_experimental_features	Enable experimental features for hugging face MLflow model conversion	boolean	False	True
extra_pip_requirements	Extra pip dependencies that MLflow model should capture as part of conversion. This would be used to create environment while loading the model for inference. Pip dependencies expressed as below. Do not use quotes for passing. eg: pkg1==1.0, pkg2, pkg3==1.0	string		True

Inputs for MLflow local validation

Name	Description	Type	Default	Optional	Enum
local_validation_test_data	Test data for MLflow local validation. Validation will be skipped if test data is not provided	uri_file		True
local_validation_column_rename_map	Provide mapping for local validation test data column names, that should be renamed before inferencing eg: col1:ren1; col2:ren2; col3:ren3	string		True

Inputs for Model registration

Name	Description	Type	Default	Optional	Enum
custom_model_name	Model name to use in the registration. If name already exists, the version will be auto incremented	string		True
model_version	Model version in workspace/registry. If the same model name and version exists, the version will be auto incremented	string		True
model_description	Description of the model that will be shown in AzureML registry or workspace	string		True
registry_name	Name of the AzureML asset registry where the model will be registered. Model will be registered in a workspace if this is unspecified	string		True
model_metadata	A JSON or a YAML file that contains model metadata confirming to Model V2 contract	uri_file		True
update_existing_model	If set to true, will update the existing model. If set to false, will create a new model.	boolean	False	True

Pipeline outputs

Name	Description	Type	Default	Optional	Enum

Name	Description	Type
mlflow_model_folder	Output path for the converted MLflow model	mlflow_model
model_registration_details	Output folder with a file which captures transformations applied above and registration details in JSON file	uri_folder