Releases: llm-workflow-engine/llm-workflow-engine
Releases · llm-workflow-engine/llm-workflow-engine
0.19.0
This release migrates from the legacy OpenAI function calling to general tool calling.
Any provider that also has Langchain tool integration should now be able to use a standardized tool calling interface.
🚒Breaking Changes🚒
- The configuration of 'functions' in presets has changed to a 'tools' configuration, see #345 for migration instructions.
- Provider.display_name is now a property instead of a method
Deprecations
The following are deprecated, and will be removed in a future release:
- '/functions' CLI command is now '/tools'
- Environment variable 'LWE_FUNCTION_DIR' has been renamed to 'LWE_TOOL_DIR'
- Configuration variable 'directories.functions' has been renamed to 'directories.tools'
Major fixes
- Refactored OpenAI function calling to general tool use
- Add gpt-4o model, use as default
- Add support for passing message list to make_request() Enables Python module usage to pass a string (one user message), or a list of messages
- Bump textract rev to support python 3.12
- NOTE: Existing installs will need to force reinstall the textract package for this upgrade, as it is a git install:
pip install --force-reinstall textract@git+https://github.com/thehunmonkgroup/textract@2109f34b4f3615004de1f2b2e635cfd61dae3cb7
- NOTE: Existing installs will need to force reinstall the textract package for this upgrade, as it is a git install:
- Support custom title in lwe_llm Ansible module
- Fixed broken command line args for --template-dir, --preset-dir, --plugin-dir, --workflow-dir, function-dir
- Renamed to --templates-dir, --presets-dir, --plugins-dir, --workflows-dir, --tools-dir
- More robust message type detection for display in CLI
- Add optional transform_tool() method to base Provider class
Commit log
- Sat May 18 2024: add optional transform_tool() method to base Provider class
- Sat May 18 2024: add func_to_json_schema_spec placeholder function
- Sat May 18 2024: script for quickly checking tool calling across providers
- Sat May 18 2024: tweak request unit tests
- Sat May 18 2024: pretty up upgrade output
- Sat May 18 2024: kill dead code
- Sat May 18 2024: add deprecation warnings for /functions -> /tools CLI command
- Sat May 18 2024: deprecation warning for directories.functions config option
- Sat May 18 2024: make tool config change in presets a breaking change
- Sat May 18 2024: remove unneeded metadata arg
- Sat May 18 2024: more robust message type detection for display in CLI
- Sat May 18 2024: clarify tool summary in doc
- Sat May 18 2024: upgrade request unit tests for tools
- Sat May 18 2024: fix broken util tests
- Sat May 18 2024: remove dead test code
- Sat May 18 2024: fix arg order
- Sat May 18 2024: tweak docs for tool use
- Fri May 17 2024: upgrade unit tests for request/token_manager/tool_cache/util for tool upgrade
- Fri May 17 2024: upgrade system tests for tools
- Fri May 17 2024: upgrade request integration tests for tools
- Fri May 17 2024: remove dead code
- Fri May 17 2024: clean up token counting calculation
- Fri May 17 2024: fix broken check for forced tool calls
- Fri May 17 2024: fix example workflow for new tools config
- Fri May 17 2024: more robust handling of user metadata fields
- Fri May 17 2024: fix streaming for tool calls
- Fri May 17 2024: clean out private customization keys before building LLM class
- Fri May 17 2024: abstract selection of title generation provider and llm
- Fri May 17 2024: abstract transform of AIMessage messages
- Fri May 17 2024: get tool call working in non-streaming case
- Thu May 16 2024: tool call submission working, return broken
- Thu May 16 2024: fix/rename broken CLI args
- Thu May 16 2024: schema upgrade code
- Thu May 16 2024: starting tool refactor, mostly renaming
- Wed May 15 2024: fix fake_llm plugin to support latest arg structure
- Wed May 15 2024: bump textract rev to support python 3.12
- Mon May 13 2024: add function docstring
- Mon May 13 2024: add gpt-4o model, use as default
- Fri May 10 2024: add support for passing message list to make_request() Enables Python module usage to pass a string (one user message), or a list of messages
- Fri May 10 2024: add build_message_from_template() support function to backend
- Fri May 03 2024: add doc for Fireworks chat provider plugin
- Tue Apr 30 2024: clarify file-summarizer new features in comments
- Tue Apr 30 2024: improvements to file-summarizer workflow
- Tue Apr 30 2024: support custom title in lwe_llm Ansible module
0.18.11
- Fri Apr 26 2024: pass missing config object during title generation
- Fri Apr 26 2024: add doc for backend_options.title_generation.provider
- Thu Apr 25 2024: fix broken request class unit tests
- Thu Apr 25 2024: improve check for custom provider in request logic
- Sat Apr 20 2024: add doc for openrouter provider plugin
- Sat Apr 20 2024: fix for Python 3.9 compat
- Sat Apr 20 2024: update chat_openai provider args
0.18.10
- Sat Apr 20 2024: migrate from pkg_resources to importlib.metadata
- Sat Apr 20 2024: add [dev] extra for development packages
- Mon Apr 15 2024: add link to chat groq provider plugin
- Tue Apr 09 2024: add latest gpt-4-turbo models to available models list
- Tue Apr 09 2024: switch to newest gpt-4-turbo
- Thu Mar 14 2024: add get_raw_template(), use in template edit/run
- Thu Feb 08 2024: check for full path in windows editor env var, more robust line splitting from where call
- Thu Feb 08 2024: fix #340: fix langchain deprecation warning
- Thu Feb 08 2024: add openai_api_base config option to chat_openai provider
- Thu Feb 08 2024: switch to gpt-3.5-turbo for default model
0.18.9
- Thu Feb 08 2024: enhance file-summarizer workflow, larger character limit for paper, add a fourth question
- Thu Feb 08 2024: adjust presets for new OpenAI models
- Thu Feb 08 2024: add gpt-4-0125-preview, gpt-4-turbo-preview, gpt-3.5-turbo-0125 models
- Wed Feb 07 2024: sync with patched FakeMessagesListChatModel langchain class
- Tue Jan 16 2024: Fix 404 link in templates.rst - "example templates"
- Mon Jan 08 2024: fix commit log link
- Mon Jan 08 2024: clean up formatting and date format
0.18.8
- Mon Jan 08 2024: fix bad reference to langchain tools
- Mon Jan 08 2024: fix broken token manager test
- Mon Jan 08 2024: allow capabilities override in provider_fake_llm
0.18.7
- Mon Jan 08 2024: fix edge cases around setting provider/model
- Mon Jan 08 2024: fix conditional logic
- Mon Jan 08 2024: check validate_models when verifying tokenizer
- Mon Jan 08 2024: fix docker compose command in docs
- Mon Jan 08 2024: move to langchain_openai partner package
0.18.6
- Mon Jan 08 2024: loosen up openai/langchain requirements
- Mon Jan 08 2024: fix test warnings
- Mon Jan 08 2024: bump langchain/openai deps
- Sat Dec 23 2023: add link for Github Gist plugin
- Tue Dec 19 2023: add plugin link to provider_chat_google_genai
- Tue Dec 19 2023: doc: add chat anthropic and chat mistralai plugin links
- Tue Dec 19 2023: fix doc build error
- Sun Dec 17 2023: remove dead pydantic-computed dep
0.18.5
- Fri Dec 15 2023: bump langchain to 0.0.350
- Sun Nov 26 2023: allow provider-specific token calculation
- Sat Nov 18 2023: add Ollama plugin to doc
- Fri Nov 17 2023: add chat-anyscale and chat-cohere plugin links
- Fri Nov 17 2023: fix #332: force backend setting, throw user warning on legacy settings
0.18.4
- Thu Nov 16 2023: exclude example dirs that start w/ an underscore
- Thu Nov 16 2023: fix examples plugin to work with non-dev installs
0.18.3
- Thu Nov 16 2023: upgrade to latest langchain, openai 1.x
- Thu Nov 16 2023: add provider_azure_openai_chat to plugins list
- Thu Nov 16 2023: lock openai package to legacy version for now
- Mon Nov 06 2023: upgrade default model to gpt-3.5-turbo-1106
- Mon Nov 06 2023: add new OpenAI provider models dated 1106, default to them in presets
- Fri Oct 20 2023: remove zap plugin from plugin list