V1 Migration Guide¶
The new ibm_watsonx_ai Python SDK 1.0 is backward compatible in most common scenarios,
but a few minor adjustments in code might be needed.
What’s new?¶
Refactor and Cleanup¶
This major release marks a thorough refactoring and cleanup of the codebase. We’ve diligently worked to improve code readability and enhance overall performance. As a result, you can expect a more efficient and maintainable package.
Easier project and space setup¶
Starting from ibm_watsonx_ai 1.0, you can set the default space
or project while initializing the APIClient object.
✅ New approach
client_project = APIClient(credentials, project_id = project_id)
client_space = APIClient(credentials, space_id = space_id)
❌ Old approach
client_project = APIClient(credentials)
client_project.set.default_project(project_id)
client_space = APIClient(credentials)
client_space.set.default_space(space_id)
Foundation model functions moved under APIClient¶
Starting from ibm_watsonx_ai 1.0, foundation model functions (get_model_specs, get_custom_model_specs,
get_model_lifecycle, get_model_specs_with_prompt_tuning_support, get_embedding_model_specs) are moved into the client.foundation_models object.
Old versions of the functions are not removed but will no longer work because they lack authentication.
✅ New approach - Using functions from client.foundation_models.
client.foundation_models.get_model_specs()
client.foundation_models.get_custom_model_specs()
client.foundation_models.get_custom_model_specs(
model_id='mistralai/Mistral-7B-Instruct-v0.2'
)
client.foundation_models.get_custom_model_specs(limit=20)
client.foundation_models.get_custom_model_specs(limit=20, get_all=True)
client.foundation_models.get_model_lifecycle(
model_id="ibm/granite-13b-instruct-v2"
)
client.foundation_models.get_model_specs_with_prompt_tuning_support()
client.foundation_models.get_embeddings_model_specs()
❌ Old approach - Using functions from the ibm_watsonx_ai.foundation_models module.
from ibm_watsonx_ai.foundation_models import (
get_model_specs,
get_custom_models_spec,
get_model_lifecycle,
get_model_specs_with_prompt_tuning_support,
get_embedding_model_specs
)
get_model_specs(
url="https://us-south.ml.cloud.ibm.com"
)
get_custom_models_spec(api_client=client)
get_custom_models_spec(credentials=credentials)
get_custom_models_spec(api_client=client, model_id='mistralai/Mistral-7B-Instruct-v0.2')
get_custom_models_spec(api_client=client, limit=20)
get_custom_models_spec(api_client=client, limit=20, get_all=True)
get_model_lifecycle(
url="https://us-south.ml.cloud.ibm.com",
model_id="ibm/granite-13b-instruct-v2"
)
get_model_specs_with_prompt_tuning_support(
url="https://us-south.ml.cloud.ibm.com"
)
get_embedding_model_specs(
url="https://us-south.ml.cloud.ibm.com"
)
Breaking changes¶
The
client.<resource>.list()methods don’t print the table with listed assets. They return the table aspandas.DataFrame. The optional parameter for thereturn_as_dfmethods was removed.
✅ New approach - Access to credential fields as class attributes.
conn_list = client.connections.list() ## table not printed
❌ Old approach - Table with listed resources printed.
conn_list = client.connections.list()
### Table returned as pandas.DataFrame `conn_list` object and printed output:
--------------------------- ------------------------------------ -------------------- ------------------------------------
NAME ID CREATED DATASOURCE_TYPE_ID
Connection to COS 71738a79-6585-4f33-bf4a-18907abcf06a 2024-04-25T10:42:23Z 193a97c1-4475-4a19-b90c-295c4fdc6517
--------------------------- ------------------------------------ -------------------- ------------------------------------
Methods and parameters that were marked as deprecated in
ibm_watsonx_ai v0were removed. For example:
✅ New approach - Use the method that replaced the deprecated one, for get_uid it’s get_id
asset_id = client.data_assets.get_id(asset_details)
❌ Old approach - Deprecated get_uid method called
client.data_assets.get_uid(asset_details)
client.credentialsandclient.wml_credentialsreturns aCredentialsobject instead of a dictionary.
✅ New approach - Access to credential fields as class attributes.
from ibm_watsonx_ai import APIClient
from ibm_watsonx_ai import Credentials
credentials = Credentials(
url = "https://us-south.ml.cloud.ibm.com",
token = "***********",
)
client = APIClient(credentials)
url = client.credentials.url
# OR
url = credentials.url
❌ Old approach - Access to credential fields as keys in a dictionary.
from ibm_watsonx_ai import APIClient
credentials = {
"url": "https://us-south.ml.cloud.ibm.com",
"token":"***********",
}
client = APIClient(credentials)
url = client.wml_credentials.get('url')
Parameter changes in
service.score(...)for online deployment
service.score(...) has an additional forecast_window parameter added before the transaction_id parameter.
From ibm_watsonx_ai 1.0.0, in the service.score(...) function, you need to pass all named parameters except for the payload parameter.
✅ New approach - named parameters and a new parameter.
predictions = service.score({
"observations": AbstractTestTSAsync.observations,
"supporting_features": AbstractTestTSAsync.supporting_features,
},
forecast_window=1,
transaction_id=transaction_id,
)
❌ Old approach - all parameters are not named, the forecast_window parameter is absent
predictions = service.score({
"observations": AbstractTestTSAsync.observations,
"supporting_features": AbstractTestTSAsync.supporting_features,
},
transaction_id,
)
Deprecations¶
Initializing
APIClientwithcredentialsas a dictionary is deprecated.- All parameters with
wmlin the name are deprecated. They were renamed to: wml_credentials->credentialswml_client->api_client.
- All parameters with
idnaming was aligned in parameters and methods and alluidwere replaced byid. Methods that were usinguidwere removed or marked as deprecated.