ibm_watson_machine_learning🐛 Bug Fixes:
Improved writing to file while downloading data_assets withbinary=True.
🚀 Features:
Improved loading all documents from subfolders.
Load AI service generated in AutoAI RAG experiment in get_pattern method.
TabularDocumentsIterableDataset added.
🚀 Features:
Changed AutoAI RAG experiment test_data_references as optional field.
Aded support forFolderAssets.
🐛 Bug Fixes:
Decreased number of requests to datasource_types API.
Removed delay inclient.spaces.store() method.
🚀 Features:
Added support fordocumentation.init anddocumentation.functions in AI services
🐛 Bug Fixes:
Fixresults_reference.location.path for Fine Tuning
🚀 Features:
Addfuture_data as a param inforecast() method instead of kwargs
Deprecateinference_function andindexing_function
🐛 Bug Fixes:
Improve elastic strategy
Added support for initializing Volumes only on CPD
Add checking for models supporting function chat only
Handle custom datasource types in list method
Fixed downloading AutoAI’s artifacts on AWS
Fix for storing package extensions on AWS
Lock token generation
🚀 Features:
FixedDataConnection methods for AWS IBM Cloud
Add Semantic Chunker
Model Gateway - support for beta release
🐛 Bug Fixes:
Double token generation fix
🚀 Features:
Addto_langchain_tool method toRetriever class
Support loadingxml files
Add methods to runtime context
Add suffix path ai service deployment
Added support for Vector Index assets
🐛 Bug Fixes:
Add missing docstring example for TS BYOM
Map all private AWS wx.ai endpoints to public platform
🐛 Bug Fixes:
Improvements for private AWS endpoints
🚀 Features:
Added support for Granite Guardian ingenerate,generate_stream methods
get_copy method added toAPIClient
Add async methods forEmbeddings
🐛 Bug Fixes:
Add fallback for adding documents to Elasticsearch vector store
Add error when externalVectorStore used
🐛 Bug Fixes:
TSModelInference not working when client has project or space is provided
🚀 Features:
Add support fortime_series BYOM on Cloud
🚀 Features:
Support forcontainer data reference and folder download forFS andContainer location on CPD
🐛 Bug Fixes:
Changed Flight service connection closing time
Add text field to hybrid strategy
🚀 Features:
Support for authorization in AWS environments
🐛 Bug Fixes:
Added retries forclient.connections.get_datasource_type_details_by_id()
Auth method has timedelta set automatically when not provided
🐛 Bug Fixes:
Fix passing vector field names for Elasticsearch
🐛 Bug Fixes:
Updated CPD version requirements forTextExtractionsV2 class
Disable inference function inRagPattern
🚀 Features:
Support for hybrid search in AutoAI RAG experiments
🚀 Features:
Deprecation warning for Prompt Tuning with CPD 5.2
Added support for Python 3.13
Added support for Guardian (Text Detection)
🐛 Bug Fixes:
Patch payload generation for CPD 5.2 fixed
🐛 Bug Fixes:
Usepymilvus<2.5.7 inrag dependencies
🚀 Features:
WDU 2.0 adaption for TextExtractions
AutoAI RAG support for BYOM
Allowed passing connection args withoutconnection_id and langchain VS
Enable store base Foundation Models on Cloud
🐛 Bug Fixes:
Removed settingfirst_line_header Flight interaction property when reading data in binary mode and removedquote_character when readingcsv files
Added support for deleting Deployment Jobs on CPD 5.1 version and lower
🚀 Features:
AutoAI RAG adjustments for the preview
Added support for folder download
🐛 Bug Fixes:
Updated url inload_file_from_file_system
Fixed logic to check if the deployment update was successful
Fixed AutoAI-Generated Notebook
Fixedsummary method, added support BYOM forget_pattern
Modify message for missing cryptography package error intext_loader.py
🚀 Features:
Rework Toolkit class
🐛 Bug Fixes:
ImprovedDeployments.update() to wait for the deployment to be ready
Improve model store response handling for CPD 5.2
🚀 Features:
Add converters for utility agent tools
Improve flight retry logic
.yaml documents supported in Document Loader
🐛 Bug Fixes:
ModelInference should set project/space whendeployment_id andvalidate is provided
Disable TokenBucket for CPD
🚀 Features:
Implementation of hybrid search - RAG
.json documents supported in document loader
Added initial support for CP4D 5.2
Addget_space_id() toRuntimeContext
.pptx documents supported in document loader
🐛 Bug Fixes:
Validation of training data reference type moved to back-end repository Service
Updated error message for AutoAI RAG run
🚀 Features:
Addedget_headers method toAPIClient class
🚀 Features:
Add default params to RAGPattern AI service
Added support forretrieval parameter of AutoAI RAG
Added support forgeneration parameter of AutoAI RAG
Add async httpx client
Exposegradient_checkpointing parameter to users for fine-tuning
Assetsduplicate_action parameter support added
Support for tooling inRepository.store_ai_service method
Trashed assets module added to SDK
🐛 Bug Fixes:
Fix removing tmp file
Removed requirement for default preset for AI service run
Fix for hanging of job when the passed benchmark data path is non existing
Added deprecation for SentenceTransformerEmbeddings class
Fixed building docs
Replacedproject_lib package removed in RT 25.1 withibm_watson_studio_lib
Added support for number columns in CSV-s and Excel files
Fixedsummary method when Fine-Tuning experiment has 1 epoch
🚀 Features:
Introduce httpx client in deployments, foundation models
🐛 Bug Fixes:
support data fields to detached prompt template
🚀 Features:
Introduce httpx client in deployments, foundation models
🐛 Bug Fixes:
support data fields to detached prompt template
🚀 Features:
Handle retry more robustly
add reference documents to ragpattern service stream
Support forUtility Agent Tools
🐛 Bug Fixes:
Verify is set correctly for httpx client
Support for deployment of ai service generated in experiments with chroma vector store
Fixed validation version parameter only for CPD
Fixed extra-space fortime_limit parameter inGenTextParamsMetaNames
🚀 Features:
Change authentication method in flight and refactor to allow private RT
Added support for window search with Elastisearch
🐛 Bug Fixes:
Change approach to listing limit of resources
Proxies setting for httpx fixed
Raise error when adding not chunked documents to vector store
🚀 Features:
Addget_all param toAutoPipelinesRuns.list() and set200 as default listed records number
🐛 Bug Fixes:
Allow inference deployments without providing project_id or space_id
APIClient.set_token is correctly changing authentication method
Updated concurrency_limit default value to 8
Improved CP4D version value handling
Convert strings with regex causingSyntaxWarning to raw strings
Added missing parameters toTextChatParameters dataclass
🚀 Features:
Addedget_all param toRAGEngine.get_details()
🐛 Bug Fixes:
Allow passingconcurrency_limit in init Embeddings
Raise error that AutoAI RAG is not available only if response is 404
Returnusage metadata when runningModelInference.chat_stream() method
🚀 Features:
Support for reading documents from NFS folder
🚀 Features:
Support for LoRA/QLoRA Tuning in watsonx.ai Software 5.1.1 and latest
🐛 Bug Fixes:
Replace only the first occurrence in the response text of stream mode
Added error to handle method not allowed in run AI Services
Disable warnings frompypdf.PdfReader inTextLoader
🚀 Features:
Added support to generate token fromzen-service-broker-secret
Added support for AI Services on CPD 5.1.1 and latest
Added indexing step to autogenerated inference AI service in ChromaDB scenario in RAGPattern
Added support for native async stream methods for text generation and chat
🐛 Bug Fixes:
Fixed authentication error during text generation with usingmeta-llama/llama-3-1-8b-instruct
Check file extension with lowercase filename
Raise error when error event occurs during generating stream
Removed default list length limit for WML Resources
Improved error message for creating connections with invalid payload
🚀 Features:
Added integration with ILAB/wx.ai
🐛 Bug Fixes:
Import pyarrow only when FlightConnection is used
Closing httpx.Response in retry mechanism
🚀 Features:
Introduced support for AI-Service within RAGPattern
Implemented validation for CPD url to ensure correctness
Added support for newstatus object in get AutoAI RAG run response
Removed potential scheme inflight_location
🐛 Bug Fixes:
Rename parameter nameword_to_token_factor toword_to_token_ratio in prompt builder
EnhancedAutoPipelinesRuns.get_run_details() to retrieve the latest run across all endpoints and raise an error when resources are unavailable
Added custom timeout to close persistent connection
Task Credentials support improvements
Discover all available assets instead of limited to 100
Fixed endpoints for API DataPlatform
Disable warnings logger forLangChainChunker.split_documents()
🐛 Bug Fixes:
Get the latest run including both endpoints and raise an error if empty resources -TuneRuns.get_run_details()
Added validation of storage details in project/ space details for Container DataConnection
Addedfs as model supported training data references type
Updated the cancel_run method in RAGOptimizer to avoid returning cached training details
🚀 Features:
Add cancel job to the text extractions
Support for Deploy on Demand
Addedword_to_token_factor parameter to RAGPattern constructor
Add batch size param to embeddings
Updated add_documents method in milvus_adapter
🐛 Bug Fixes:
add batch size to dict only if customized
Added training data references type validation in repository.store_model
Disable loggingNotS3Connection in one case
Pagination support added to lookup datasources types
Pagination added for datasources
🚀 Features:
Add support for IBM watsonx.data Milvus connection
Added method to get chat model specs with function calling support
Add support for achat method
🐛 Bug Fixes:
Improve storing AI services content
🚀 Features:
Support for Sydney watsonx.ai API endpoint
Add ‘Large’ option for HardwareRequestSizes
🐛 Bug Fixes:
Update connections to enable AutoAI RAG experiment
🚀 Features:
Adjustedai_service support to align with recent changes in the AI services backend
Added deprecation warning on Model class
Support for writing to FSLocation
Add get id by name for spaces
🐛 Bug Fixes:
Downgraded thegrpcio version to>=1.54.3
Remove flag from RagPattern constructor
Updatedshow schema method
Update text extraction delete job
SSL error fixed on FIPS cluster when downloading documents from COS folder
save Milvus’s SSL certificate file in temporary file
Improve error message for creating connections with invalid payload
Update set_api_client method
🚀 Features:
Added documentation for Fine-Tuning:Working with TuneExperiment and FineTuner
🐛 Bug Fixes:
Improved error reporting forDataConnection.read() when attempting to read from a database with the flight service disabled
Added a generic error message for failed attempts to download datasets due to flight service issues
Restricted Python version compatibility to versions below 3.13
Migrated dependencies to LangChain version 0.3.x
🚀 Features:
Introduced support for loading text from Markdown files in the TextLoader class
Added an enum for rerank models:client.foundation_models.RerankModels
Enhanced theget_details method with an optional name parameter
Implemented theset_api_client method in the ModelInference class
🐛 Bug Fixes:
Added 429 error handling for theset.default_project method’s project_id component
Improved error handling during data loading operations
Fixed an issue where theget_id_by_name method raised an error when no ID was found for the specified name
🚀 Features:
Introduced support for AI services
Added support for Rerank functionality:Rerank Documentation
Added support for Text Extraction results in Markdown format.
🐛 Bug Fixes:
Updated the route for CPD token generation for CPD versions 5.1 and above
Disabled AutoAI TSAD in the SDK for CPD versions starting from 5.1
🚀 Features:
Added support for the Chat API
🐛 Bug Fixes:
Automatically added thehttps:// prefix to the COS connectionendpoint_url if missing.
🚀 Features:
Introduced theget_evaluation_results method in theRAGOptimizer class
Added support for CPD version 5.1
🐛 Bug Fixes:
Corrected the example value forTextExtractionsMetaNames
Relaxed version constraints for LangChain packages
🚀 Features:
Added the option to installlangchain_community with the[rag] option
🐛 Bug Fixes:
Fixed an issue withDataConnection as a container folder.
Updated theload() method docstring in the documentation.
Added support for platform URLs in credentials.
🚀 Features:
Added integration for AutoAI RAG SDK:Working with AutoAI RAG class and rag_optimizer
🐛 Bug Fixes:
Increasedbatch_size reduction in AutoAI RAG experiments
🐛 Bug Fixes:
Added a timeout toDocumentsIterableDataset iteration
Included a private development URL in the platform map
🚀 Features:
Introduced support for Fine-Tuning
Added support for native asynchronous methods for model inference
🚀 Features:
Added support for MCSP
🐛 Bug Fixes:
Fixed an issue with reading JSON files without binary mode, resolving problems with reading Prompt Tuning files
🚀 Features:
Added a timeout for persistent connections
🐛 Bug Fixes:
Set a defaultmax_sequence_length to avoid model limits errors
🚀 Features:
Added support forConnectionAsset without a specified location
🐛 Bug Fixes:
Resolved an issue where model specification filters were not available in version 4.8
Fixed a bug related to unsupported connection types
Aligned logging for improved consistency
🚀 Features:
Implemented Keep-Alive and persistent connections for tokenization
Added support for the text extraction endpoint:Text Extractions Documentation
🐛 Bug Fixes:
Resolved the issue with missingpython-docx
🚀 Features:
Introduced support for Hardware Request field instead of the Hardware Specification
Added an asynchronousgenerate method
Implemented support for theapi_client parameter inDocumentIterableDataset
🐛 Bug Fixes:
Fixed an issue with storingTaskCredential
Improved de-duplication in theadd_documents method of the vector store to include both text and its document ID
🚀 Features:
Introduced support for Bring Your Own Model (BYOM)
🐛 Bug Fixes:
Setinput_shape during training, addressing issues with dataset training data
Removed the timeout for reading documents
Implemented improvements to the Documents Dataset
🐛 Bug Fixes:
Resolved an issue caused by a breaking change insetuptools version 71.0.1
🐛 Bug Fixes:
Added a setter for theServiceInstance.details property
🚀 Features:
Added Keep-Alive functionality for FoundationModels requests
Introduced support for the privatewml-fvt endpoint
Added installation support for RAG group packages
🐛 Bug Fixes:
Improved performance of theAPIClient initialization
Disabled lifecycle checks when validation is turned off
Updated Scikit-learn for federated learning support
Resolved issues with reading data from NFS connections in Git-based projects
Changed the default behavior to read files on the cloud with Flight Service enabled
🚀 Features:
Added support for different input modes of Prompt Templates(PromptTemplate,FreeformPromptTemplate,DetachedPromptTemplate).
🐛 Bug Fixes:
Fixed validation for tech preview models.
Fixed set project for git-based-project in new scenario.
Fixed used filters forget_model_specs method.
🚀 Features:
Added support for tech preview models
Added additional information in request headers.
Extend VectorStore functionalities via concrete wrappers(For Vector Index notebooks)
🚀 Features:
Addedvalidate_prompt_variables parameter to generate method inModel andModelInference class.
Addedhardware_spec support in Batch class.
🐛 Bug Fixes:
Fixed correct schema for timeseries-anomaly-prediction prediction type.
🐛 Bug Fixes:
Added more clear Error message when user pass invalid credentials.
Fixed “Invalid authorization token” error when initializing the client with the “USER_ACCESS_TOKEN” environment variable on a cluster
🚀 Features:
Added auto-generated Enum classes (TextModels, EmbeddingModels, PromptTunableModels) with available models
🐛 Bug Fixes:
Better filtering of supported runtimes for r-scripts
Fixed downloading model content to file
Improved scaling Prompt Tuning charts
Provided a clearer error message when a user passes an incorrect URL to the cluster
🚀 Features:
Added forecast_window parameter support for online deployment scoring
🚀 Features:
Milvus support for RAG extension
Autogenerated changelog
Travis is tagging the version during push into production
🐛 Bug Fixes:
Reading data assets as binary when flight is unavailable
next resource generator type fixed, other internal type issues fixed
Reading tabular dataset with non-unique columns
Deprecation warnings removed when using fm_model_inference
🚀 Features:
Added get and get_item methods for better compatibility with SDK v0
🐛 Bug Fixes:
Relaxed package version checking
Fixed AutoAI initialization without version in credentials
Fixed calls to wx endpoints in git base project
Fixed backward compatibility of WebService and Batch class for credentials in dictionary
🐛 Bug Fixes:
Hotfix for imports
🚀 Features:
RAGutils module added
Getting foundation models specs moved under foundation models object
Credentials as object + proxies supported
WCA service support
🐛 Bug Fixes:
Minor corrections/improvements to Foundation Models module