The LangChain project is an active Python project advancing applications for large language models. With an impressive user engagement showcased by 66,412 watchers and a high star count, the project evidences a strong community involvement.
The repository is extensive; over 5,637 commits spread across 383 branches have developed a sizeable codebase of 104,946 kB. Python is the predominant language of the project. The presence of CI and Experimental CI badges implies a commitment to solid development and testing practices.
Issues regarding breaking changes have been encountered, specifically affecting SQLDatabase chains. However, proactive action has been taken to provide mitigation plans and warnings. The repository also promotes open-source contributions, highlighting its collaborative spirit.
Notable issues range across various topics:
- Connection failures and timing issues point to complications in OpenAIEmbeddings.
- HTTP URI incompatibility has been reported in the Milvus wrapper.
- Actively reported bugs like exceptions when using
create_pandas_df_agent and similarity_score_threshold for ConversationalRetrievalChain not working as expected.
- Unclear documentation in several areas including roles in BaseCallbackHandler and BaseTracer, batch inference with
RetrievalQA.from_chain_type and Elastic VectorStore which can confuse users.
- Issues with functionalities of several components including UnstructuredFileLoader, Pinecone Vector Store, and
ChatTongyi class methods.
The activity level in the repository is high, but an accumulation of ongoing issues may discourage potential users or contributors. However, comprehensive documentation with step-by-step walkthroughs of functionality also exists, assisting newcomers in familiarizing themselves with the software.
Detailed Reports
Report on issues
Open Issues Overview:
- OpenAIEmbeddings issues: Connection failures (#12714) and timeout parameter malfunctioning (#12712).
- Milvus wrapper incompatibility with HTTP URIs (#12710).
- Exception thrown when using
create_pandas_df_agent (#12709), a potentially unusual error message.
similarity_score_threshold for ConversationalRetrievalChain not behaving as expected (#12707).- Confusion about managing conversation histories for different users in
ConversationBufferWindowMemory (#12706).
- Deprecation of
model_api_request in CerebriumAI (#12705).
- UnstructuredFileLoader malfunctioning (#12700).
- BaseCallbackHandler vs BaseTracer: Clarity needed in documentation of their roles (#12698).
- Pinecone Vector Store incorrectly assigning scores (#12697).
- Issue related to
ChatTongyi class method usage (#12695).
- Unclear documentation on batch inference with
RetrievalQA.from_chain_type and Elastic VectorStore (#12693).
- Certain models missing expected output format 'Action:' after 'Thought:' (#12689).
- A local variable reference issue in
PALChain.from_math_prompt (#12681).
VertexAIEmbeddings timing out (#12662
Report on pull requests
Open Pull Requests
- #12749: Dependency update, primarily for got and docusaurus; no issues flagged.
- #12747: Adds domain restriction to APIChain for security enhancements; some discussion about defaults. Notable because it addresses CVE-2023-32786.
- #12746: Improvements and fixes to demo server; no issues flagged.
- #12731: Enables
device_map parameter for huggingface pipeline; no issues flagged.
- #12726: Fixes issue regarding a problem with iterable 'Nonetype' for ObsidianLoader; no issues flagged.
- #12725: Adds support for vectorstores that return cosine similarity; no issues flagged.
- #12724: Adds model config to invocation params; no issues flagged.
- #12723: Corrects bugs in VertexAIModelGarden class; no issues flagged.
- #12719: Enhances support for varying type of containers for AWS Sagemaker LLM Streaming; no issues flagged.
- #12718: Introduces support for pgvecto.rs as a VectorStore; linter issues were addressed.
- #12717: Fixes an issue with the Vertex AI Search MultiTurn Retriever; no issues flagged.
- #12715: Fixes minor typo error in google_vertex_ai_palm.ipynb; no issues flagged.
- #12713: Fixes issue with number of elements in config list in
batch() and abatch() of BaseLLM; issue #12643.
- #12704: Handles deprecation of certain Cerebrium function; no issues flagged. Issue #12705.
- #12703: Fixes minor issue with VertexAI streaming; no issues flagged.
- #12702: Adds support for Xinference Chat, a range of fixes applied; no issues flagged.
- #12696: Adds functions to handle original queries in MultiQuery; no issues flagged.
- #12694: A work-in-progress related to self-query template; some discussion about defaults.
- #12690: Updates Zep's VectorStore to use native MMR; no issues flagged.
- #12688: Adds a basic critique revise cookbook; no issues flagged.
- #12687: Allows on_artifacts to be passed to a conversation; no issues flagged.
- #12686: Adds support for Zep to search over chat history summaries; no issues flagged.
- #12683: Adds a guide on RAG on biomedical data; no issues flagged.
- #12677: Updates linter for Python notebooks; failing test suggestions raised.
- #12666: Adds a feature for using metadata vars to form LLM prompts and a supporting test case; no issues flagged.
- #12633: Adds the ability to extract content and metadata from .pdf-saved emails; no issues flagged.
- #12618: Confirms log entries after the creation of a new project; no issues flagged.
- #12612: Adds a RAG example using Cassandra / Astra DB; no issues flagged.
- #12602: Automates the management of Docker inference server containers for the LLM; no issues flagged.
- #12595: Removes unnecessary DocCardList in indexes; no issues flagged.
- #12586: Introduces HuggingFaceTextGenInferenceAuto; no issues flagged.
- #12584: Adds environment variables section to missing template readmes; no issues flagged.
- #12579: Fixes a security issue in
_load_prompt_from_file function(s); no issues flagged.
- #12565: Fixes sources in LlamaIndexRetriever; issue #12563.
Future Attention Needed
Report on README and metadata
LangChain by LangChain AI is a Python-based project focusing on creating applications by leveraging the power of large language models (LLMs). The project provides interfaces and utilities for integrating LLMs with other sources of computation or knowledge. Its primary uses include creating question-answering applications, chatbots, and AI agents. The project is actively developed with its most recent update pushed on November 1, 2023.
The LangChain repository boasts a significant user engagement level, demonstrated by its 66,412 watchers and equivalently high star count. The project's codebase has grown to a size of 104,946 kB through 5,637 commits across 383 branches. The repository maintains its active status with 2,061 open issues, covering a wide range of topics, reflecting an active user base and ongoing maintenance. The project clearly relies on Python as its primary language, while the presence of CI and Experimental CI badges indicates its adherence to robust development and testing practices.
LangChain seems to grapple with some breaking changes, indicated by a clear warning and mitigation plan for select chains (SQLDatabase) on July 28, 2023 - an effort towards making LangChain leaner and safer. The project actively encourages open-source contributions, hinting at a collaborative and evolving nature of the software. Despite the high-activity level of the repository, the repository has many ongoing issues that might deter some potential users or contributors. Lastly, the project provides comprehensive documentation, complete with walkthroughs of specific functionalities, making it easier for newcomers to acclimate themselves with the software.
Summary
Detailed Reports