Releases: chatchat-space/Langchain-Chatchat
Releases · chatchat-space/Langchain-Chatchat
Langchain-Chatchat v0.2.0
重要提示
因 v0.2.0 版本与前序版本相比调整较大，使用 v0.2.0 版本时建议先新建环境，重新安装依赖包后，按照readme中开发部署操作顺序进行配置。
自版本 v0.2.0 起，项目正式更名为 Langchain-Chatchat，项目将基于 Langchain 和包含 ChatGLM-6B 在内的开源大语言模型提供持续优化的知识库问答应用。
变更信息
- 使用 FastChat 提供开源 LLM 模型的 API，以 OpenAI API 接口形式接入，提升 LLM 模型加载效果，并支持近期新出的baichuan-inc/baichuan-7B、internlm/internlm-chat-7b、Qwen/Qwen-7B-Chat等模型；
* 使用 langchain 中已有 Chain 的实现，便于后续接入不同类型 Chain，并将对 Agent 接入开展测试；
* 使用 FastAPI 提供 API 服务，全部接口可在 FastAPI 自动生成的 docs 中开展测试，且所有对话接口支持通过参数设置流式或非流式输出；
* 使用 Streamlit 提供 WebUI 服务，可选是否基于 API 服务启动 WebUI，后续可支持不同形式输出内容的显示；
* 项目中默认 LLM 模型改为 THUDM/chatglm2-6b，默认 Embedding 模型改为 moka-ai/m3e-base，文件加载方式与文段划分方式也有调整，后续将重新实现上下文扩充，并增加可选设置；
* 项目中扩充了对不同类型向量库的支持，除支持 FAISS 向量库外，还提供 Milvus, PGVector 向量库的接入；
* 项目中搜索引擎对话，除 Bing 搜索外，增加 DuckDuckGo 搜索选项，DuckDuckGo 搜索无需配置 API Key，在可访问国外服务环境下可直接使用。
主要贡献者
@imClumsyPanda
@liunux4odoo
@zqt996
@hzg0601
@glide-the
@inksong
Contributors
imClumsyPanda, inksong, and 4 other contributors
Assets 2
3
Join discussion
langchain-ChatGLM v0.1.17
重要提示
本版本为 langchain-ChatGLM v0.1.x 最后一个版本，接下来开发团队将全部精力投入到 v0.2.0 版本的开发，新版本将尽快完成上线。
变更信息
- Update MyFAISS.py by @foxmoon in #648
- 解决vue前端镜像构建失败的问题 by @hzhaoy in #636
- fix:前端知识库获取失败. 修改获取及删除接口至knowledge_base by @1729457433 in #653
- Revert 653 dev 1 by @fxjhello in #657
- 标题增强 by @kiddog99 in #631
- 整理了下issue的问题，方便新手入门 by @fengyunzaidushi in #669
- Fix #675: 解决因vue前端代码有误导致镜像构建失败的问题 by @hzhaoy in #679
- 前端知识库相关接口报错修复 by @sivdead in #651
- Dev streaming oom by @allseeeeeee in #722
- 向Readme添加Docker整合包描述 by @inksong in #746
- fix 2 bugs on streamlit demo by @neverless00 in #719
- Add device_map config to support chatglm2-6b by @Jingsong-Yan in #734
- 修复api接口websocket的伪流式输出 by @zzxs in #628
- Update model_config.py by @stc2001 in #768
- 添加启动API的参数，支持https、wss方式API调用 by @zfanswer in #728
- 解决 image_loader.py 运行时报错“ModuleNotFoundError: No module named 'configs'”的问题。 by @ysyx2008 in #795
- fix:The knowledge base name cannot be empty. by @kztao in #793
- Update README.md， 0.1.16版本已经支持删除知识库中的文件 by @kztao in #792
- 1.增加对llama-cpp模型的支持；2.增加对bloom/chatyuan/baichuan模型的支持；3. 修复多GPU部署的bug;4. 修复了moss_llm.py的bug；5. 增加对openai支持（没有api,未测试);6. 支持在多卡情况自定义设备GPU by @hzg0601 in #664
- feat: fastapi 接口优化 by @NieLamu in #684
- 修改默认的多卡部署方案，基本保证针对新模型也不会失败 by @hzg0601 in #788
- 测试OPENAI接口成功 by @hzg0601 in #813
- 添加新支持的LLM(bloomz-7b1,bloom-3b,baichuan-7b,vicuna-13b-delta-1.1) by @kztao in #845
- fix Non-UTF-8 code starting with \xe5 by @kztao in #849
- 更新 streamlit ui by @liunux4odoo in #850
- Fix LocalDocQA object has no attribute llm when adding docs by @benjaminhuo in #852
- fix issue#580: 新建知识库出错 by @liunux4odoo in #880
- 在args.py中增加ptuning相关的参数 by @hzg0601 in #838
- 1 修复model_config.py的type;2 修改默认的LLM_model;3. 移除chatglm_llm.py的无用模块 by @bones-zhu in #904
- 解决加载ptuning检查点失败的问题 by @chinainfant in #905
- [Fix] INSTALL.md by @JimmyMa99 in #907
- 完善api.py，增加LLM对话和知识库对话的流式输出功能 by @margox in #659
- 安装教程conda create新增默认的/envs下面，而不只是只能要额外指定路径 用-p by @KennyNg-19 in #482
- support chatglm2cpp by @cylee0909 in #859
- Update FAQ.md by @YikunHan42 in #1006
- 添加语音模块 by @nanayashiki1215 in #1039
- Update MyFAISS.py by @chiquitita-101 in #1041
- Adding an implementaion for involving with accessing Baichuan-13B-Chat（接入baichuan-13b-chat逻辑） by @Vincent-ch99 in #1005
- 线性简单实现了：vecorstores/MyFAISS.py TODO: 增加是否属于同一文档的判断 by @Wu-Fisher in #945
New Contributors
- @foxmoon made their first contribution in #648
- @hzhaoy made their first contribution in #636
- @1729457433 made their first contribution in #653
- @fxjhello made their first contribution in #657
- @kiddog99 made their first contribution in #631
- @sivdead made their first contribution in #651
- @allseeeeeee made their first contribution in #722
- @inksong made their first contribution in #746
- @neverless00 made their first contribution in #719
- @Jingsong-Yan made their first contribution in #734
- @stc2001 made their first contribution in #768
- @zfanswer made their first contribution in #728
- @ysyx2008 made their first contribution in #795
- @kztao made their first contribution in #793
- @benjaminhuo made their first contribution in #852
- @bones-zhu made their first contribution in #904
- @chinainfant made their first contribution in #905
- @JimmyMa99 made their first contribution in #907
- @margox made their first contribution in #659
- @KennyNg-19 made their first contribution in #482
- @cylee0909 made their first contribution in #859
- @YikunHan42 made their first contribution in #1006
- @nanayashiki1215 made their first contribution in #1039
- @chiquitita-101 made their first contribution in #1041
- @Vincent-ch99 made their first contribution in #1005
- @Wu-Fisher made their first contribution in #945
Full Changelog: v0.1.16...v0.1.17
Contributors
kztao, zfanswer, and 29 other contributors
Assets 2
langchain-ChatGLM v0.1.16
What's Changed
1. 新增功能
- 新增自定义 vectorstore 实现类 MyFAISS，增加对 FAISS 向量库中全部文件列表查询、指定文件对应向量删除、指定文件向量替换。
- 在 WebUI 和 API 中提供删除知识库、从知识库中删除指定文件的功能。
2. 问题修复
- 修复 pdf_loader 和 image_loader 进行测试时可能发生未找到 configs 的问题（感谢 @you-can-change 在 #112 (comment) 中提出的解决方案）
- 修复知识库多次检索时，内容可能发生叠加的问题（感谢 @jkmchinese 在 #613 (comment) 中提出的解决方案）
- 更新 README.md by @Lix1993 in #608
- 修复[bug] #598 #601，解决通过cli.py启动api、cli失败的问题 by @hzg0601 in #599
- 更新FAQ,model_config.py,bing_search.py文件，增加bing搜索问答相关问题的原因和可能解决方案，修复了一些typo;修改loader.py中load_in_8bit失败的原因和详细解决方案 by @hzg0601 in #610
New Contributors
Full Changelog: v0.1.15...v0.1.16
Contributors
jkmchinese, Lix1993, and 2 other contributors
Assets 2
langchain-ChatGLM v0.1.15
What's Changed
1. 新增功能
- WebUI 中现已支持刷新页面自动更新知识库列表 by @youyajike in #529
- 增加对 moka-ai/m3e-small 和 moka-ai/m3e-base 两种 embedding 模型的支持
2. 问题修复
- VUE 前端默认端口从 1002 调整为 8090 by @NieLamu in #555
- 修复中 WebUI 中仅显示一轮对话的BUG by @sysalong in #570
- 更新了requrements.txt和FAQ，解决了 #591 #592 by @hzg0601 in #593
- 修复 api.py 中 json 解析问题 by @zzxs in #540
New Contributors
- @youyajike made their first contribution in #529
- @zzxs made their first contribution in #540
- @NieLamu made their first contribution in #555
- @hzg0601 made their first contribution in #593
Full Changelog: v0.1.14...v0.1.15
Contributors
sysalong, youyajike, and 3 other contributors
Assets 2
langchain-ChatGLM v0.1.14
What's Changed
新增功能
- 新增通过 FastChat API 调用 LLM 的方式，实现了更多 LLM 的支持，并有效提升 LLM 回复速度 by @glide-the
- 新增 lru_cache 函数以提升本地向量库加载速度 by @liunux4odoo in #496
- Web UI 中增加更新知识库列表按钮，以解决刷新页面后 gradio 不支持自动更新知识库选项列表问题 by @imClumsyPanda
- 添加streamlit ui by @liunux4odoo in #480
- 添加重新初始化 uploadpath 内文档的按钮 by @zhoutongqing in #418
问题修复
- 更新 Dockerfile-cuda by @zhubao315 in #446
- 修改：读取本地知识文件时，能够读取给定目录下全部（包括直接和间接属于该目录）文件 by @DingJunyao in #471
- 修复 api 中 delete_docs 的 bug by @yihuaxiang in #499
- 修复上传 PDF 文件后可能提示 tmp.png 文件不存在的问题 by @imClumsyPanda
- 修复上传文件后可能提示本地缺少 faiss.index 导致文件上传至知识库失败的问题 by @imClumsyPanda
New Contributors
- @zhubao315 made their first contribution in #446
- @liunux4odoo made their first contribution in #480
- @DingJunyao made their first contribution in #471
- @yihuaxiang made their first contribution in #499
- @zhoutongqing made their first contribution in #418
Full Changelog: v0.1.13...v0.1.14
Contributors
zhubao315, imClumsyPanda, and 5 other contributors
Assets 2
langchain-ChatGLM v0.1.13
What's Changed
新增功能
- 新建llm接口类，并将chatglm与moss重新根据接口类定义实现类 by @glide-the
- 调整 model_config 中的 llm_model_dict 定义形式 by @glide-the
- 在 Web UI 和 vue 前端中增加bing搜索支持，目前仅支持使用知识库或使用bing搜索进行问答，同时使用知识库与bing搜索的功能将在后续进行支持。by @halfss in #378
- 添加命令行管理脚本 by @zqt996 in #355
- 增加删除知识库中原始文件 by @halfss in #365
- 更新 README.md
- 更新 install 说明，解决 PaddleOCR 依赖问题 by @changxubo in #387
问题修复
- 导入txt过程中可能存在的编码问题
- 导入pdf过程中可能存在的转换为txt后的编码问题，及可能出现unsupported colorspace for 'png’ 的问题
- 修复 vue 前端中上传文件的问题 by @halfss in #362
- 更新 api.py by @cd9807 in #357
- 修复 MOSS模型加载BUG by @changxubo in #389
- vue 前端 UI 修复 by @halfss in #394
- 修复历史记录最后一条丢失 by @zhenkaivip in #415
- 修复对话中换行连读造成歧义 by @zhenkaivip in #415
- 修复每次对话携带多余数据的问题 by @mayflyfy in #425
New Contributors
- @halfss made their first contribution in #362
- @zqt996 made their first contribution in #355
- @mayflyfy made their first contribution in #425
Full Changelog: v0.1.12...v0.1.13
Contributors
halfss, zhenkaivip, and 5 other contributors
Assets 2
langchain-ChatGLM v0.1.12
What's Changed
1. 新增功能
- 增加对 png 和 jpg 格式图片的 OCR 识别载入，基于 PaddleOCR 实现【使用前请先确认已卸载detectron2】by @zhenkaivip in #344
- 更新对 pdf 文件的读取方法，支持中英双语，基于 PaddleOCR 实现【使用前请先确认已卸载detectron2】by @zhenkaivip in #344
- 增加允许跨域调用API功能 by @hikariming in #279
- 增加对 MOSS 的支持 by @ykk648 in #317
- 增加传入单个文件与LLM对话接口 by @keenzhu
2. 问题修复
- 修复 chatglm 两次初始化的问题 by @ykk648 in #313
- 修复多卡运行时 lora 模型加载问题 by @ykk648 and @RainGather in #312 and #316
- 修复 local_doc_qa.py 中 logger 问题 by @RainGather in #323
- 修复 cli_demo.py 中可能存在无法正常显示流式输出内容的问题 by @imClumsyPanda
- 修复 api.py 中可能存在的现有接口运行错误问题 by @imClumsyPanda
New Contributors
- @hikariming made their first contribution in #279
- @ykk648 made their first contribution in #313
- @RainGather made their first contribution in #316
- @zhenkaivip made their first contribution in #342
Full Changelog: v0.1.11...v0.1.12
Contributors
keenzhu, RainGather, and 4 other contributors
Assets 2
langchain-ChatGLM v0.1.11
What's Changed
1. 新增功能
- Web UI 中新增加知识库 Beta 页面，提供知识库名分句、匹配测试 by @sysalong in #302
- 新增基于 Vue 的前端页面 by @fxjhello
- 新增 logging 格式日志输出 by @changxubo in #268
2. 问题修复
- 修复默认知识库获取的逻辑 by @mokeyish in #293
- 修复 logger bug by @oushu1zhangxiangxuan1 in #297
New Contributors
- @mokeyish made their first contribution in #293
- @oushu1zhangxiangxuan1 made their first contribution in #297
Full Changelog: v0.1.10...v0.1.11
Contributors
sysalong, changxubo, and 3 other contributors
Assets 2
langchain-ChatGLM v0.1.10
What's Changed
1. 新增功能
- 优化针对中文文本字符的分句方式 by @imClumsyPanda
- 增加基于语义的分句方式 by @cd9807 in #248
- 支持 Lora 权重加载 by @sysalong in #231
- webui 中增加 Flagging 日志收集问答效果。 by @changxubo in #250
- 更新 INSTALL.md by @lenosp in #223
2. 问题修复
- 修复 similarity_search_with_score_by_vector 函数返回多个文段时的 score 共用了相同结果的错误 by @imClumsyPanda
- 修复扩展文档的代码逻辑 by @winterggg in #227
- 修复知识库无法上载，NLTK_DATA_PATH路径错误 by @changxubo in #236
- 完善知识库路径问题，完善api接口 by @sysalong in #245
- 更新 requirements.txt 中 gradio 版本要求，修复知识库下拉框 change 事件获取值不是当前选中值的bug。 by @changxubo in #261
New Contributors
- @lenosp made their first contribution in #223
- @winterggg made their first contribution in #227
- @sysalong made their first contribution in #231
- @changxubo made their first contribution in #236
- @cd9807 made their first contribution in #248
Full Changelog: v0.1.9...v0.1.10
Contributors
imClumsyPanda, sysalong, and 4 other contributors
Assets 2
langchain-ChatGLM v0.1.9
What's Changed
1. 新增功能
- 使用 fastapi 实现 API by @BrightXiaoHan in #209
- 添加了Dockerfile-cuda , 提供GPU Docker支持 by @cocoMan in #197
- 更新 README.md 中原理介绍、AutoDL 镜像链接等。
2. 问题修复
- 修复上一版本中可能存在低版本 torch.mps 调用问题，改为仅在 mps 可用时导入相关函数。
New Contributors
- @BrightXiaoHan made their first contribution in #209
Full Changelog: v0.1.8...v0.1.9
Contributors
cocoMan and BrightXiaoHan