Added documentation links for various document loaders, embeddings, and llms to improve the readability and usability of the config.yaml file. These links provide a quick reference to the documentation for each of the modules, making it easier for developers to understand and use them.
The `AgentType` enum is added to the `langchain.agents.custom` module to improve readability and type safety. The `InitializeAgent` class now uses the `AgentType` enum to ensure that the `agent` parameter is a valid value from the enum.
The import of RecursiveCharacterTextSplitter was removed as it was not being used in the code. The instantiation of TextSplitter was fixed by removing the unnecessary check for RecursiveCharacterTextSplitter and simplifying the code.
🔥 refactor(loading.py): remove unused import of RecursiveCharacterTextSplitter
The commit changes the comparison operator from '==' to 'is' to compare object types. This is because 'is' compares the object identity while '==' compares the object value. In this case, we want to compare the object identity, so 'is' is the correct operator to use.
🐛 fix(loading.py): fix type hinting in instantiate_embedding function
🔨 refactor(loading.py): add type hinting to instantiate_textsplitter function
The changes in this commit add type hinting to the `instantiate_agent`, `instantiate_embedding`, and `instantiate_textsplitter` functions to improve code readability and maintainability. The `instantiate_embedding` function had a bug in its type hinting which has been fixed.
Added documentation links to the vectorstores integrations in the config.yaml file. This will make it easier for developers to access the documentation for each integration.
The RecursiveCharacterTextSplitter class in textsplitters.py now has a new field called separator_type. This field is used to specify the type of separator to be used in the splitter. The separator_type field is a string and can take any value from the Language enum or "Text". This change was made to improve the flexibility of the RecursiveCharacterTextSplitter class.
This commit adds type hints to the function parameters and return types in the loading.py file. This improves the readability and maintainability of the codebase by making it easier to understand the expected types of the parameters and return values of the functions.
Added documentation links for new integrations and memories to improve the documentation of the project. The new integrations are Cohere and HuggingFaceHub, and the new memories are ConversationBufferWindowMemory and VectorStoreRetrieverMemory.
The VectorStoreFrontendNode class now has VectorStoreRetriever as an extra base class in addition to BaseRetriever. This change was made to improve the functionality of the class by allowing it to inherit from VectorStoreRetriever.
The `update_settings` function now accepts a `cache` parameter that allows the user to specify the type of cache to use. The `cache` parameter is set to a default value of `SQLiteCache` and can be overridden by setting the `LANGCHAIN_CACHE` environment variable. This feature improves the flexibility of the application as it allows the user to choose the type of cache that best suits their needs.
The cache configuration option has been added to the settings file with a default value of "InMemoryCache". This allows the user to choose the cache implementation they want to use.
This commit adds support for configurable LLM caching. The `setup_llm_caching` function now imports the cache class from the `langchain.cache` module based on the `settings.cache` value. If the import is successful, the `langchain.llm_cache` is set to an instance of the cache class. If the import fails, a warning is logged. If an exception is raised during the setup, a warning is logged with the error message.
This commit adds documentation links for the LlamaCpp and CTransformers integrations in the config.yaml file. The links point to the relevant documentation pages on the LangChain website. This improves the accessibility of the documentation for these integrations.
✨ feat(main.py): call setup_llm_caching function on app startup
The `setup_llm_caching` function is added to `utils.py` to set up LLM caching. The function is then called on app startup in `main.py` using the `app.on_event("startup")` method. This improves the performance of the application by caching LLM objects.
The default values of the settings attributes were changed from an empty list to an empty dictionary. This change was made to avoid errors that could occur when trying to access a non-existent key in the dictionary.
✨ feat(frontend_node): add documentation field to the frontend node dict representation
The `set_documentation` method is added to the `FrontendNode` class to allow setting the documentation of the frontend node. The `to_dict` method is updated to include the `documentation` field in the dict representation of the frontend node. This improves the readability and usability of the frontend node by providing documentation for the node.
This commit adds a new property to the LangChainTypeCreator class called docs_map, which is a dictionary that maps the name of the component to its documentation link. The docs_map property is used to set the documentation of the component in the signature of the component. This change improves the readability and maintainability of the code by making it easier to add and update documentation for components.
The SlackDirectoryLoader is added to the list of document loaders in the DocumentLoaderFrontNode class. This allows users to load zip files from Slack into the application.
The GitLoader template now has four new fields: repo_path, clone_url, branch, and file_filter. These fields allow the user to specify the repository path, clone URL, branch, and file extensions to be loaded. This improves the flexibility of the GitLoader template and allows it to be used in a wider range of scenarios. Additionally, a minor change was made to the add_extra_fields method to ensure that the field.show attribute is set to True for all fields.
The `instantiate_documentloader` function now supports filtering files by extension using a `file_filter` parameter. The parameter is a string of comma-separated extensions, and the function now converts it into a lambda function that filters files based on whether their name contains any of the specified extensions. This improves the flexibility of the document loader by allowing it to load only specific types of files.
The fields in the Template class were previously sorted by DIRECT_TYPES, which caused issues when fields had the same field_type. Sorting alphabetically first ensures that fields are sorted in a consistent manner before sorting by DIRECT_TYPES.
The `instantiate_vectorstore` function now uses a dictionary to initialize vector stores instead of a series of if-else statements. This improves the readability and maintainability of the code. A new dictionary `vecstore_initializer` is added to `vector_store.py` to map the class names of vector stores to their respective initialization functions.
The `instantiate_vectorstore` function now supports the `MongoDBAtlasVectorSearch` vector store. This allows for the use of MongoDB Atlas as a vector store for Langflow. The `search_kwargs` parameter is now supported for all vector stores that have a `as_retriever` method. This allows for the configuration of the vector store's search parameters.
The hardcoded values for db_name, collection_name, and index_name have been removed from the initialize_mongodb function and are now required parameters. This makes the function more flexible and allows it to be used with different databases and collections. The support for the index_name parameter has been added to the MongoDBAtlasVectorSearch template in vectorstores.py, which allows the user to specify the name of the index to be used in the search.
🐛 fix(vector_store.py): remove hardcoded values for db_name, collection_name, and index_name and make them required parameters
✨ feat(langflow): add support for search_kwargs field in VectorStoreFrontendNode
The changes add support for MongoDB Atlas Vector Search in the vectorstores. The `MongoDBAtlasVectorSearch` class is now imported and initialized in `vector_store.py`. The `initialize_mongodb` function is added to initialize the MongoDB Atlas Vector Search class. The `VectorStoreFrontendNode` class is updated to add the `mongodb_atlas_cluster_uri`, `collection_name`, and `db_name` fields. The `search_kwargs` field is also added to the `VectorStoreFrontendNode` class to allow users to pass additional search parameters to the vector store.
The type hinting for allowed_tools variable is unnecessary as it is already defined in the previous line. Removing the type hinting improves the readability of the code.
The display name for the SupabaseVectorStore is now set to "Supabase". This improves the user experience by providing a more descriptive name for the vector store.
The `sort_fields` method has been added to the `Template` class to sort fields based on the `DIRECT_TYPES` constant. Fields that have a `field_type` in `DIRECT_TYPES` are sorted first, followed by the remaining fields. This ensures that fields that have a direct type are processed first, which is important for the correct functioning of the template.
The DIRECT_TYPES constant was removed from the vertex and graph modules as it is now defined in the utils module. This change improves code organization and reduces duplication.
The format_dict function was updated to set a default value for the model_name key in the value dictionary for the OpenAI, ChatOpenAI, and Anthropic models. This ensures that the model_name key always has a value, even if the options list is empty.
✨ feat(chains.py): add TemplateField 'chain_type' to support different types of QA chains
The 'memory' field was previously set to required=False, but it is actually required for the chain to function properly. This fix sets required=True for the 'memory' field.
A new TemplateField 'chain_type' has been added to support different types of QA chains. This field is of type 'str', is required, and is a list of options. It allows the user to select the type of QA chain they want to use.
The advanced field was set to True for input_key and output_key fields, which made them appear in the advanced section of the UI. This was not intended, so the advanced field is now set to False for these fields.
🐛 fix(process.py): make inputs optional in process_graph_cached function
The inputs and tweaks parameters in the process_flow endpoint are now optional, which allows for more flexibility in the API. The inputs parameter in the process_graph_cached function is now optional, which prevents a ValueError from being raised when a Chain object is processed without inputs.
The memory_key field is now set to "chat_history" by default. This change ensures that the memory_key field is always initialized with a default value, which is useful for the application's functionality.
The import statements for the weaviate and pinecone libraries were causing import errors. Adding the `# type: ignore` comment suppresses these errors and allows the code to run without issues.
This commit adds type hinting to the function arguments of `get_result_and_thought` and `process_graph_cached` functions in `process.py` file. This improves code readability and maintainability.
The `initialize_chroma` function had redundant code that was removed. The `embedding_function` parameter was renamed to `embedding` to match the parameter name used in the `class_object` constructor. The `documents` and `texts` parameters were being used interchangeably, so the code was updated to use only `documents`.
