Langchain-Chatchat is an open-source, Python-based language model Q&A application. It's popular (18262 stars, 187 watchers, 3089 forks) and active (1539 commits, 12 branches, 42 open issues). The project is designed for offline deployment and emphasizes data security and privacy. It supports multiple models and APIs, with plans for expansion.
There are 42 open PRs, with a mix of new features, enhancements, and bug fixes. Notably:
Some PRs, like #552 and #1200, have been open for a while with no recent updates.
1966 PRs have been closed. Recent merges include:
Some PRs, like #2026 and #2000, were closed without merging.
Recent issues highlight bugs/errors, language-specific issues (particularly Chinese), UI problems, and model handling issues. Older open issues and recently closed issues share similar themes.
Langchain-Chatchat is an active, popular project with ongoing enhancements and bug fixes. Some PRs remain open for extended periods, and language-specific, UI, and model handling issues are recurrent themes in user-reported issues.
The recently opened issues for the software project are mainly related to bugs and errors. A significant number of these issues are related to the use of the software in non-English languages, particularly Chinese, as seen in issues #2074, #2073, #2072, and #2067. This suggests that there might be some language-specific bugs or compatibility issues in the software.
Another common theme among the recent issues is problems with the software's user interface, as seen in issues #2070, #2069, and #2068. These issues suggest that there might be some bugs or design flaws in the software's user interface that are causing problems for users.
A third common theme among the recent issues is problems with the software's model handling, as seen in issues #2066, #2065, and #2064. These issues suggest that there might be some bugs or design flaws in the software's model handling mechanisms that are causing problems for users.
The older open issues for the software project are mainly related to bugs and errors, similar to the recently opened issues. Some of these older issues, such as #2059, #2057, and #2055, are still open, possibly because they are complex or difficult to resolve, or because they are not considered high priority by the project's maintainers.
Recently closed issues include #14, which was about optimizing the software's performance. This issue was closed after a discussion about possible ways to improve the software's performance.
In summary, the common themes among all open and recently closed issues are bugs and errors, problems with the software's user interface, and problems with the software's model handling. These themes suggest that these are areas where the software could be improved.
There are 42 open pull requests. The most recent ones are #2071, #2046, #2013, and #1981. The oldest ones are #552, #1200, #1757, #1943.
PR #2071 is a new feature adding file dialogue mode. It adds new API endpoints and configuration options. It was created 0 days ago and is actively being discussed.
PR #2046 is a feature that adds db memory. It introduces a callback_handler to automatically save chat records to db. It was created 2 days ago and is actively being discussed.
PR #2013 is a feature enhancement for OCR recognition of PPT and DOC files in the knowledge base. It was created 5 days ago and is actively being discussed.
PR #1981 is a bug fix for not deleting the knowledge base directory when deleting the pg knowledge base. It was created 8 days ago and is actively being discussed.
PR #552 is a feature that uses multiple processes to import multiple PDF files and improve paddleorc recognition. It was created 162 days ago and hasn't been updated recently.
PR #1200 is an optimization of EventSource response. It was created 85 days ago and hasn't been updated recently.
PR #1757 is a new feature that adds audio support using Alibaba's ASR SOTA model. It was created 31 days ago and hasn't been updated recently.
PR #1943 is a revision that adds index support for documents based on optional configuration files. It was created 14 days ago and hasn't been updated recently.
There are 1966 closed pull requests. The most recent ones are #2060, #2058, #2049, #2041, #2034, #2033, #2026, #2021, #2002, and #2000.
PR #2060 was merged 1 day ago. It includes non-model worker started online models (such as openai-api) and already downloaded local models in the webui model list.
PR #2058 was merged 1 day ago. It adds support for passing prompt words to the ChatGLM3-6B agent.
PR #2049 was merged 1 day ago. It is a simple fix for a typo in the documentation.
PR #2041 was merged 2 days ago. It updates the documentation and prompt words for the ChatGLM3-6B agent and fixes a few issues.
PR #2034 was merged 2 days ago. It is a simple fix for a typo in the README.
PR #2033 was merged 2 days ago. It updates the comments in the requirements files to English.
PR #2026 was not merged. It was created 3 days ago and closed 2 days ago. It aimed to add multi kb and custom prompt.
PR #2021 was merged 5 days ago. It added ES support and fixed several typo bugs.
PR #2002 was merged 6 days ago. It supports multiple models being started at the same time through configuration items and includes Wiki in the samples knowledge base.
PR #2000 was not merged. It was created 6 days ago and closed 3 days ago. It aimed to support Alibaba Cloud's text_embedding_v1 model.
There are several pull requests related to adding new features or enhancing existing ones, such as file dialogue mode (#2071), db memory (#2046), OCR recognition of PPT and DOC files (#2013), and audio support (#1757).
There are also several pull requests related to bug fixes, such as not deleting the knowledge base directory when deleting the pg knowledge base (#1981).
Some pull requests are related to improving the performance of the software, such as using multiple processes to import multiple PDF files and improve paddleorc recognition (#552).
Some pull requests are related to improving the user interface and user experience, such as including non-model worker started online models and already downloaded local models in the webui model list (#2060).
Some pull requests are related to improving the documentation and comments in the code, such as fixing typos (#2049, #2034) and updating the comments in the requirements files to English (#2033).
Some pull requests have been open for a long time without being updated or merged, such as #552, #1200, #1757, and #1943. This may indicate that these pull requests are not being actively worked on or that there are issues with the proposed changes.
Some pull requests were not merged, such as #2026 and #2000. This may indicate that there were issues with the proposed changes or that the changes were not accepted by the project maintainers.
Langchain-Chatchat is a Python-based project developed by the organization chatchat-space. It is a local knowledge-based language model question and answer application, built on Langchain and ChatGLM. The project is designed to be open-source and capable of offline deployment. The software is licensed under the Apache License 2.0. The README provides a comprehensive guide on how to set up and use the application, including environment setup, model download, initialization of knowledge base and configuration files, and application launch.
The repository is quite mature and active, with 1539 total commits, 12 branches, and 42 open issues at the time of analysis. It has garnered significant popularity, as evidenced by its 18262 stars and 187 watchers. The repository size is 56627 kB and it has been forked 3089 times. The README details the software's technical architecture and software stack, which includes Langchain, ChatGLM, FastChat, FastAPI, and Streamlit. The project also supports open-source LLM and Embedding models, and it can be deployed offline using open-source models.
The README highlights the project's ability to solve data security and privacy issues, making it an ideal solution for enterprises. It also mentions the project's inspiration from other projects like document.ai and ChatGLM-6B Pull Request. The project provides a Docker image for easy deployment and has a comprehensive Wiki for users seeking a deeper understanding of the project. The project does not involve fine-tuning or training processes, but these can be used to optimize the project's performance. The project also supports the call of the OpenAI GPT API and plans to continue expanding the access to various models and model APIs.
Summary
Detailed Reports