1.9.2

1.9.2

Bug Fixes

• DataConnector: Fix self._session related errors inside Container Runtime.
• Registry: Fix a bug when trying to pass None to array (pd.dtype('O')) in signature and pandas data handler.

New Features

• Experiment Tracking (PrPr): Automatically log the model, metrics, and parameters while training
  XGBoost and LightGBM models.

from snowflake.ml.experiment import ExperimentTracking
from snowflake.ml.experiment.callback import SnowflakeXgboostCallback, SnowflakeLightgbmCallback

exp = ExperimentTracking(session=sp_session, database_name="ML", schema_name="PUBLIC")

exp.set_experiment("MY_EXPERIMENT")

# XGBoost
callback = SnowflakeXgboostCallback(
  exp, log_model=True, log_metrics=True, log_params=True, model_name="model_name", model_signature=sig
)
model = XGBClassifier(callbacks=[callback])
with exp.start_run():
  model.fit(X, y, eval_set=[(X_test, y_test)])

# LightGBM
callback = SnowflakeLightgbmCallback(
  exp, log_model=True, log_metrics=True, log_params=True, model_name="model_name", model_signature=sig
)
model = LGBMClassifier()
with exp.start_run():
  model.fit(X, y, eval_set=[(X_test, y_test)], callbacks=[callback])

1.9.1

1.9.1

Bug Fixes

• Registry: Fix a bug when trying to set the PAD token the HuggingFace text-generation model had multiple EOS tokens.
  The handler picks the first EOS token as PAD token now.

New Features

• DataConnector: DataConnector objects can now be pickled
• Dataset: Dataset objects can now be pickled
• Registry (PrPr): Introducing create_service function in snowflake/ml/model/models/huggingface_pipeline.py
  which creates a service to log a HF model and upon successful logging, an inference service is created.

from snowflake.ml.model.models import huggingface_pipeline

hf_model_ref = huggingface_pipeline.HuggingFacePipelineModel(
  model="gpt2",
  task="text-generation", # Optional
)


hf_model_ref.create_service(
    session=session,
    service_name="test_service",
    service_compute_pool="test_compute_pool",
    image_repo="test_repo",
    ...
)

• Experiment Tracking (PrPr): New module for managing and tracking ML experiments in Snowflake.

from snowflake.ml.experiment import ExperimentTracking

exp = ExperimentTracking(session=sp_session, database_name="ML", schema_name="PUBLIC")

exp.set_experiment("MY_EXPERIMENT")

with exp.start_run():
  exp.log_param("batch_size", 32)
  exp.log_metrics("accuracy", 0.98, step=10)
  exp.log_model(my_model, model_name="MY_MODEL")

• Registry: Added support for wide input (500+ features) for inference done using SPCS

1.9.0

1.9.0

Bug Fixes

• Registry: Fixed bug causing snowpark to pandas dataframe conversion to fail when QUOTED_IDENTIFIERS_IGNORE_CASE
  parameter is enabled
• Registry: Fixed duplicate UserWarning logs during model packaging

Behavior Changes

• ML Job: The list_jobs() API has been modified. The scope parameter has been removed,
  optional database and schema parameters have been added, the return type has changed
  from snowpark.DataFrame to pandas.DataFrame, and the returned columns have been updated
  to name, status, message, database_name, schema_name, owner, compute_pool,
  target_instances, created_time, and completed_time.
• Registry: Set relax_version to false when pip_requirements are specified while logging model
• Registry: UserWarning will now be raised based on specified target_platforms (addresses spurious warnings)

New Features

• Registry: target_platforms supports TargetPlatformMode: WAREHOUSE_ONLY, SNOWPARK_CONTAINER_SERVICES_ONLY,
  or BOTH_WAREHOUSE_AND_SNOWPARK_CONTAINER_SERVICES.
• Registry: Introduce snowflake.ml.model.target_platform.TargetPlatform, target platform constants, and
  snowflake.ml.model.task.Task.
• ML Job: Single-node ML Jobs are now in GA. Multi-node support is now in PuPr
  • Moved less frequently used job submission parameters to **kwargs
  • Platform metrics are now enabled by default
  • list_jobs() behavior changed, see Behavior Changes for more info

1.8.6

1.8.6

Bug Fixes

New Features

• Registry: Add service container info to logs.

1.8.5

1.8.5

Bug Fixes

• Registry: Fixed a bug when listing and deleting container services.
• Registry: Fixed explainability issue with scikit-learn pipelines, skipping explain function creation.
• Explainability: bump minimum streamlit version down to 1.30

1.8.4

1.8.4

Bug Fixes

• Registry: Default enable_explainability to True when the model can be deployed to Warehouse.
• Registry: Add custom_model.partitioned_api decorator and deprecate partitioned_inference_api.
• Registry: Fixed a bug when logging pytroch and tensorflow models that caused
  UnboundLocalError: local variable 'multiple_inputs' referenced before assignment.

Breaking change

• ML Job: Updated property id to be fully qualified name; Introduced new property name to represent the ML Job name
• ML Job: Modified list_jobs() to return ML Job name instead of id
• Registry: Error in log_model if enable_explainability is True and model is only deployed to
  Snowpark Container Services, instead of just user warning.

New Features

• ML Job: Extend @remote function decorator, submit_file() and submit_directory() to accept database and
  schema parameters
• ML Job: Support querying by fully qualified name in get_job()
• Explainability: Added visualization functions to snowflake.ml.monitoring to plot explanations in notebooks.
• Explainability: Support explain for categorical transforms for sklearn pipeline
• Support categorical type for xgboost.DMatrix inputs.

1.8.3

1.8.3

Bug Fixes

Behavior Change

New Features

• Registry: Default to the runtime cuda version if available when logging a GPU model in Container Runtime.
• ML Job: Added as_list argument to MLJob.get_logs() to enable retrieving logs
  as a list of strings
• Registry: Support ModelVersion.run_job to run inference with a single-node Snowpark Container Services job.
• DataConnector: Removed PrPr decorators

1.8.2

1.8.2

Bug Fixes

Behavior Change

New Features

• ML Job now available as a PuPr feature
• ML Job: Add ability to retrieve results for @remote decorated functions using
  new MLJobWithResult.result() API, which will return the unpickled result
  or raise an exception if the job execution failed.
• ML Job: Pre-created Snowpark Session is now available inside job payloads using
  snowflake.snowpark.context.get_active_session()
• Registry: Introducing save_location to log_model using the options argument.
  User's can provide the path to write the model version's files that get stored in Snowflake's stage.
• Registry: Include model dependencies in pip requirements by default when logging in Container Runtime.

reg.log_model(
    model=...,
    model_name=...,
    version_name=...,
    ...,
    options={"save_location": "./model_directory"},
)

• ML Job (PrPr): Add instance_id argument to get_logs and show_logs method to support multi node log retrieval
• ML Job (PrPr): Add job.get_instance_status(instance_id=...) API to support multi node status retrieval

1.8.1

1.8.1

Bug Fixes

• Registry: Fix a bug that caused unsupported model type error while logging a sklearn model with score_samples
  inference method.
• Registry: Fix a bug that model inference service creation fails on an existing and suspended service.

Behavior Change

New Features

• ML Job (PrPr): Update Container Runtime image version to 1.0.1
• ML Job (PrPr): Add enable_metrics argument to job submission APIs to enable publishing service metrics to Event Table.
  See Accessing Event Table service metrics
  for retrieving published metrics
  and Costs of telemetry data collection
  for cost implications.
• Registry: When creating a copy of a ModelVersion with log_model, raise an exception if unsupported arguments are provided.

1.8.0

1.8.0

Bug Fixes

• Modeling: Fix a bug in some metrics that allowed an unsupported version of numpy to be installed
  automatically in the stored procedure, resulting in a numpy error on execution
• Registry: Fix a bug that leads to incorrect Model is does not have _is_inference_api error message when assigning
  a supported model as a property of a CustomModel.
• Registry: Fix a bug that inference is not working when models with more than 500 input features
  are deployed to SPCS.

Behavior Change

• Registry: With FeatureGroupSpec support, auto inferred model signature for transformers.Pipeline models have been
  updated, including:

  • Signature for fill-mask task has been changed from

    ModelSignature(
        inputs=[
            FeatureSpec(name="inputs", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="outputs", dtype=DataType.STRING),
        ],
    )

    to

    ModelSignature(
        inputs=[
            FeatureSpec(name="inputs", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureGroupSpec(
                name="outputs",
                specs=[
                    FeatureSpec(name="sequence", dtype=DataType.STRING),
                    FeatureSpec(name="score", dtype=DataType.DOUBLE),
                    FeatureSpec(name="token", dtype=DataType.INT64),
                    FeatureSpec(name="token_str", dtype=DataType.STRING),
                ],
                shape=(-1,),
            ),
        ],
    )

  • Signature for token-classification task has been changed from

    ModelSignature(
        inputs=[
            FeatureSpec(name="inputs", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="outputs", dtype=DataType.STRING),
        ],
    )

    to

    ModelSignature(
        inputs=[FeatureSpec(name="inputs", dtype=DataType.STRING)],
        outputs=[
            FeatureGroupSpec(
                name="outputs",
                specs=[
                    FeatureSpec(name="word", dtype=DataType.STRING),
                    FeatureSpec(name="score", dtype=DataType.DOUBLE),
                    FeatureSpec(name="entity", dtype=DataType.STRING),
                    FeatureSpec(name="index", dtype=DataType.INT64),
                    FeatureSpec(name="start", dtype=DataType.INT64),
                    FeatureSpec(name="end", dtype=DataType.INT64),
                ],
                shape=(-1,),
            ),
        ],
    )

  • Signature for question-answering task when top_k is larger than 1 has been changed from

    ModelSignature(
        inputs=[
            FeatureSpec(name="question", dtype=DataType.STRING),
            FeatureSpec(name="context", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="outputs", dtype=DataType.STRING),
        ],
    )

    to

    ModelSignature(
        inputs=[
            FeatureSpec(name="question", dtype=DataType.STRING),
            FeatureSpec(name="context", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureGroupSpec(
                name="answers",
                specs=[
                    FeatureSpec(name="score", dtype=DataType.DOUBLE),
                    FeatureSpec(name="start", dtype=DataType.INT64),
                    FeatureSpec(name="end", dtype=DataType.INT64),
                    FeatureSpec(name="answer", dtype=DataType.STRING),
                ],
                shape=(-1,),
            ),
        ],
    )

  • Signature for text-classification task when top_k is None has been changed from

    ModelSignature(
        inputs=[
            FeatureSpec(name="text", dtype=DataType.STRING),
            FeatureSpec(name="text_pair", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="label", dtype=DataType.STRING),
            FeatureSpec(name="score", dtype=DataType.DOUBLE),
        ],
    )

    to

    ModelSignature(
        inputs=[
            FeatureSpec(name="text", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="label", dtype=DataType.STRING),
            FeatureSpec(name="score", dtype=DataType.DOUBLE),
        ],
    )

  • Signature for text-classification task when top_k is not None has been changed from

    ModelSignature(
        inputs=[
            FeatureSpec(name="text", dtype=DataType.STRING),
            FeatureSpec(name="text_pair", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureSpec(name="outputs", dtype=DataType.STRING),
        ],
    )

    to

    ModelSignature(
        inputs=[
            FeatureSpec(name="text", dtype=DataType.STRING),
        ],
        outputs=[
            FeatureGroupSpec(
                name="labels",
                specs=[
                    FeatureSpec(name="label", dtype=DataType.STRING),
                    FeatureSpec(name="score", dtype=DataType.DOUBLE),
                ],
                shape=(-1,),
            ),
        ],
    )

  • Signature for text-generation task has been changed from

    ModelSignature(
        inputs=[FeatureSpec(name="inputs", dtype=DataType.STRING)],
        outputs=[
            FeatureSpec(name="outputs", dtype=DataType.STRING),
        ],
    )

    to

    ModelSignature(
        inputs=[
            FeatureGroupSpec(
                name="inputs",
                specs=[
                    FeatureSpec(name="role", dtype=DataType.STRING),
                    FeatureSpec(name="content", dtype=DataType.STRING),
                ],
                shape=(-1,),
            ),
        ],
        outputs=[
            FeatureGroupSpec(
                name="outputs",
                specs=[
                    FeatureSpec(name="generated_text", dtype=DataType.STRING),
                ],
                shape=(-1,),
            )
        ],
    )

• Registry: PyTorch and TensorFlow models now expect a single tensor input/output by default when logging to Model
  Registry. To use multiple tensors (previous behavior), set options={"multiple_inputs": True}.

  Example with single tensor input:

  import torch
  
  class TorchModel(torch.nn.Module):
      def __init__(self, n_input: int, n_hidden: int, n_out: int, dtype: torch.dtype = torch.float32) -> None:
          super().__init__()
          self.model = torch.nn.Sequential(
              torch.nn.Linear(n_input, n_hidden, dtype=dtype),
              torch.nn.ReLU(),
              torch.nn.Linear(n_hidden, n_out, dtype=dtype),
              torch.nn.Sigmoid(),
          )
  
      def forward(self, tensor: torch.Tensor) -> torch.Tensor:
          return cast(torch.Tensor, self.model(tensor))
  
  # Sample usage:
  data_x = torch.rand(size=(batch_size, n_input))
  
  # Log model with single tensor
  reg.log_model(
      model=model,
      ...,
      sample_input_data=data_x
  )
  
  # Run inference with single tensor
  mv.run(data_x)

  For multiple tensor inputs/outputs, use:

  reg.log_model(
      model=model,
      ...,
      sample_input_data=[data_x_1, data_x_2],
      options={"multiple_inputs": True}
  )

• Registry: Default enable_explainability to False when the model can be deployed to Snowpark Container Services.

New Features

• Registry: Added support to single torch.Tensor, tensorflow.Tensor and tensorflow.Variable as input or output
  data.
• Registry: Support xgboost.DMatrix
  datatype for XGBoost models.