Releases: ExtensityAI/symbolicai
Releases · ExtensityAI/symbolicai
v0.6.5
Release Note
Updates
- Removed unused imports for optimized performance.
- Improved argument and keyword argument passing on engine creation.
- Introduced preview and representation improvements.
- Added beam search keyword to the huggingface engine.
- Introduced a custom property exception on reserved assignment.
- Added support for GPT-4o-mini.
- Optimized performance with the removal of the blip to avoid M3 incompatibility, thus it is now only optional.
- Added huggingface support through a local OpenAI-compatible FastAPI server.
- Refactored llama.cpp engine and added symserver.
- Revised async tests in llama.cpp.
- Implemented minor tweaks in symrun and removed OpenAI cost tracker and all related elements.
- Added anthropic dependency and support for Anthropic API, with minor refactoring.
- Added a check to disable symai warnings and marked cost tracking for deprecation.
- Switched to pytest for easier and more efficient testing.
Bug Fixes
- Fixed a major bug in the type system with operators, ensuring operator type safety.
- Fixed a race condition bug on symrun exit.
- Fixed tests-related issues.
- Fixed a warning if local is not used.
Removals
- Removed deprecated services.
- Removed print statements.
Security
- Addressed security issues noted by Dependabot.
Please note that this update also includes minor refactoring and code improvements and is part of our ongoing efforts to maintain and improve our platform's performance.
v0.6.0
Release Notes
New Features:
- Support for similarity normalization and extended support for models and a cost tracker.
- Implementation of SAT Solver, added a linker for results and circular reference checks.
- Added a shape fix, torch type check for embedding, and generalized radial basis function (RBF) implementation for the distance function.
- Introduced an improved aggregator class with better error handling.
- Added parallel component support.
- Implemented a custom vector database (DB) and added defaults for non-API settings to favor local embedding models and vector DB.
- Introduced LLaVA support for in-memory images and format adaptation.
- Extended similarities and added a metric component to enhance tracking and comparison capabilities.
Enhancements:
- Added various kernel functions to expand computational capabilities.
- Expanded tensor and array support for typing, improving the handling of different data structures.
- Improved distance function implementations to enhance calculations.
- Improved type checking and conversion mechanisms, allowing for more robust data handling.
- Added an improved graph construction and visualization process.
- Improved memory and tagging management within Conversations.
- Added a new tutorial video to enhance user learning resources.
- Implemented better suited presets for smoother configuration.
Bug Fixes:
- Multiple bug fixes across the application, including shape-related issues, logic errors, and improved error messaging.
- Fixed a typo and conducted small refactoring in a few areas for code clarity.
- Fixed an indexing issue introduced after backend rework and documentation inaccuracies.
- Addressed a bug with embedding dimension handling and fixed query overloads.
- Rectified numerous minor issues that affected usability and stability.
Documentation and Infrastructure:
- Update README.md with new details and minor typing correction.
- Enhanced user experience with better error messages on file parse.
- Improved type passing and normalization of keyword arguments (kwargs) on expressions.
- Fixed issues regarding parameter passing to conversations.
- Improved shell script for app binary and handled symbolic sub-symbolic representation.
- Added Docker support to simplify the deployment process.
Refactoring:
- Major refactoring for initializing and purging processes, along with normalization of kwargs on expressions.
- Refactored code to improve dynamic context handling and facilitate type selection.
- Performed functional refactoring related to primitives and dynamic operator binding for efficiency and readability.
Miscellaneous Changes:
- Added local image support for the OCR engine.
- Provided more lenient declarative capabilities for callables and primitives.
- Enabled API key and model change arguments for flexible engine configurations.
- Resolved issues regarding path handling in various components such as the crawler and engine initialization.
Note: This changelog is a summary of the main changes and improvements made to the software, consolidated from the commits over the given period. Users are encouraged to update their versions to benefit from the new features and fixes introduced in this release.
v0.5.0
Release Notes
New release with reworked backend / engines code.
Enhancements:
- Introduced an improved default settings configuration to enhance user experience.
- Engine documentation has been updated for better clarity.
- Added wizard enhancements for better reading out existing settings.
- Enhanced the remedy function in the engines for more efficient error handling.
- Extended support for models and cost trackers with minor bug fixes.
- The replacement algorithm has been improved for handling paths.
Fixes:
- Resolved issues with missing variables that could potentially cause runtime errors.
- Fixed a bug causing the notebook popup to show the wizard twice.
- Fixed double memory inserts for Conversations to prevent data redundancy.
- Addressed multiple path-related issues, improving shell path parsing and path replacement.
- Resolved engine initialization issues for java files (tika) and fixed lazy loading of engines.
- Fixed an issue that caused improper handling of parameters when passing to conversations.
- Corrected errors in documentation to align with the current functionalities.
- Created measures to avoid throwing exceptions on data additions for reliability.
Backend and Internal Changes:
- Completed extensive refactoring of the complete codebase, potentially reaching a stable version.
- Refactoring included engines, search functions, pre and post-processors, and the core structure of components.
- Refined the engine bug fixes and consolidated the handler logic for more robust functionality.
- Added metadata objects and corrected serialization processes.
- Reworked the kwargs and method parameters.
File and Directory Changes:
- Shifted outputs directory to home for cleaner log handling.
Documentation and Misc:
- A new video tutorial has been added for user guidance.
- Updated README and TERMS_OF_SERVICE documents to reflect the latest operational parameters.
- Fixed encoding read for proper file handling.
Policy Updates:
- Implemented a revised and detailed Terms of Service document.
Noteworthy:
- Major codebase revision has been conducted with the possibility of this being the first stable release post-refactor.
- Significant restructuring and refactoring have been performed aimed at enhancing maintainability and scalability of the code.
Please note that while these updates have been thoroughly tested, we recommend users to report any unexpected behaviors or bugs to help us improve further.
v0.4.34
Release Notes
New Features
- Added setup wizard for improved user experience; logic to read out and utilize existing settings more efficiently.
- Created configurable splash screen.
- Introduced a new collection framework to streamline data handling.
- Created a data collection pipeline and consent messages for user agreements, transparency and clarity.
- Created API interface with extended capabilities for CRUD repository calls and component operation calls.
- Added a new experimental feature: an interpreter for better search engine result evaluations.
- Introduced the ability to start conversations with simple index references.
- Added terms of service and acknowledging the agreement process.
- Enhanced usability with improved notification system for users.
- Extended support for gpt-4-vision-preview.
- Added vision and function capabilities to the OpenAI LLM backend.
- Added more demo notebooks for better understanding of features.
- Included an experimental endpoint for style extensions in shell.
- Implemented bibtex reference formatting tool for users.
Bug Fixes
- Fixed bug with auto-on versus auto-off data sharing discrepancy.
- Fixed various typos and encoding issues for better system stability.
- Corrected deployment scripts to ensure smoother updates and installs.
- Resolved path issues related to terms of service documents.
- Fixed a bug in speech engine setup that could affect performance.
- Addressed ID type and serialization issues for accurate data processing.
- Fixed reference issue to sub-modules for seamless integration.
- Corrected Windows shell bugs and ensured compatibility with conda path.
- Solved encoding of files and escaped command mapping on lower-level model queries.
Improvements
- Upgraded properties to enhance system capabilities.
- Revised the model catalog to include the latest updates.
- Set default model to gpt-3.5-turbo for more general availability and better user experience.
- Removed extraneous debug lines for a cleaner code environment.
- Refined the build process and addressed issues related to build.os readthedocs.
- Improved the error message display to assist troubleshooting.
Documentation and Guides
- Updated README.md with the latest information on usage and features.
- Updated TERMS_OF_SERVICE.md to reflect the new terms and guidelines.
- Improved installation documentation for user guidance.
Repositories and Versioning
- Migrated repository to ExtensityAI for future support governed by non-profit organization.
- Improved repository handling with the addition of a trusted repositories file.
- Updated the list of repositories to ensure access to the latest components.
- Advanced the versioning system to reflect new changes and improvements.
v0.3.5
Release Notes
New Features
- Added better view formatting for enhanced user experience.
- Implemented start conversation with simple index reference to facilitate user interaction.
- Improved find and retrieve functionality is now mapped to conversation memory.
Enhancements
- Enhanced error messages for better user understanding when issues occur.
- Updated preview feature to provide more accurate representations.
- Added support for multiple browsers in selenium crawler, improving cross-platform usability.
Fixes
- Fixed issue with default index names to avoid conflicts.
- Resolved major problems with prompt composition for smoother operation.
- Addressed bug with reverse primitive operations to ensure consistency.
- Updated classes with minor fixes for overall stability improvements.
v0.3.0
Major Release to new OpenAI API
Major Enhancements
- Introduced support for GPT-4-Turbo Preview, a significant upgrade in the series of language models.
- Updated framework to align with the latest OpenAI API developments.
- Implemented text-to-speech functionality, providing a new dimension of interaction with the model.
Model Improvements
- Added a new default model to ensure more efficient and accurate responses.
- Refined memory retrieval capabilities, allowing for better context understanding and recall.
- Enhanced shell indexing to interact more seamlessly with the system's shell environment.
Usability Enhancements
- Fixed the maximum length computation issue for preview models, optimizing performance.
- Improved general shell interactions for a smoother user experience.
- Added improvements specifically targeted at shell command order and execution.
Fixes and Minor Tweaks
- Resolved issues with incorrect version numbers.
- Added missing package installations to prevent potential operational hiccups.
- Included minor tweaks to optimize existing functionalities.
Documentation and Help
- Enriched the system's manual pages to assist users in utilizing new and existing features.
v0.2.56
v0.2.54
Enhancements
- Data for fine-tuning models has been added alongside a refactoring of primitives for fine-tuning.
- Parallel execution has been introduced to enhance performance.
- Shell capabilities have been extended to provide more functionality.
- First working prototype of the new shell tool is now available.
Fixes
- Issue with pickle has been resolved.
- A fix for an issue discovered in the interactive shell has been added.
- Additional error handling for permission-related and filenotfound errors has been added.
Other Updates
- The way pickle is used in parallel has been updated using pathos.
- Default core usage has been reduced.
- Missing files for the interactive shell and the new shell tool prototype have been added.
- Styles for word completer have been implemented.
v0.2.50
Release Note
New Features
- Added support to stream over large audio files in batches of 30s.
- Added comments to symbol; added fine tuning support for openai.
- Support added to recover the transcript with timestamps.
- Added fine tuning capability.
- New release.
Bug Fixes
- Fixed issue with accessing properties when outside of class.
- Corrected the return type handling.
- Fixed some edge cases in timestamp sync.
- Fixed a bug related to object return and lazy loading.
(Fixed issue with BIG maggotrons; Big Daddy commit.)
Improvements
- Improved timestamp synchronization.
- Switched from mel based decoding to transcribe because mel detection.
- Added support for returning timestamps alongside the transcription.
- Made preview fix.
- Fixed readme video.
Other Changes
- Added top-k feature.
- Included minor fixes and better error handling.
v0.2.49
Release Notes
Major Stability Issue Fixed after Refactoring
New Features
- Added 'topk' feature
- Introduced 'os command'
Improvements
- Included better error handling
- Added fixes
Bug Fixes
- Fixed bug related to object return and lazy loading
- Resolved bug with stream processing
- Fixed table view