weiweiw
/
Langchain-Chatchat
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

修复 Langchain 更新之后OpenAI在线embed无法正常使用问题

This commit is contained in:
zR 2023-12-23 11:36:11 +08:00
parent 4e5bc8b399
commit 2e1442a5c1
5 changed files with 84 additions and 139 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

73
requirements.txt
View File

@ -1,76 +1,77 @@
# API requirements
langchain==0.0.344
langchain-experimental>=0.0.42
# On Windows system, install the cuda version manually from https://pytorch.org/
torch>=2.1.2
torchvision>=0.16.2
torchaudio>=2.1.2
langchain>=0.0.352
langchain-experimental>=0.0.47
pydantic==1.10.13
fschat>=0.2.33
xformers>=0.0.22.post7
sentence_transformers
transformers>=4.35.2
torch==2.1.0 ##on Windows system, install the cuda version manually from https://pytorch.org/
torchvision #on Windows system, install the cuda version manually from https://pytorch.org/
torchaudio #on Windows system, install the cuda version manually from https://pytorch.org/
openai>=1.3.7
fastapi>=0.104
sse_starlette
xformers>=0.0.23.post1
openai>=1.6.0
sentence_transformers>=2.2.2
transformers>=4.36.2
fastapi>=0.105
nltk>=3.8.1
uvicorn>=0.24.0.post1
starlette~=0.27.0
unstructured[all-docs]==0.11.0
python-magic-bin; sys_platform == 'win32'
SQLAlchemy==2.0.19
faiss-cpu # `conda install faiss-gpu -c conda-forge` if you want to accelerate with gpus
faiss-cpu>=1.7.4
accelerate>=0.24.1
spacy>=3.7.2
PyMuPDF
rapidocr_onnxruntime
PyMuPDF>=1.23.8
rapidocr_onnxruntime>=1.3.8
requests>=2.31.0
pathlib>=1.0.1
pytest>=7.4.3
numexpr>=2.8.7
numexpr>=2.8.8
strsimpy>=0.2.1
markdownify>=0.11.6
tiktoken>=0.5.1
tiktoken>=0.5.2
tqdm>=4.66.1
websockets
numpy~=1.24.4
pandas~=2.0.3
websockets>=12.0
numpy~=1.26.2
pandas~=2.1.4
einops>=0.7.0
transformers_stream_generator==0.0.4
vllm==0.2.2; sys_platform == "linux"
httpx[brotli,http2,socks]>=0.25.2
transformers_stream_generator>=0.0.4
vllm>=0.2.6; sys_platform == "linux"
# optional document loaders
# rapidocr_paddle[gpu] # gpu accelleration for ocr of pdf and image files
jq # for .json and .jsonl files. suggest `conda install jq` on windows
# html2text # for .enex files
# beautifulsoup4 # for .mhtml files
# pysrt # for .srt files
rapidocr_paddle[gpu]>=1.3.0.post5
jq>=1.6.0
html2text
beautifulsoup4>=4.12.2
pysrt>=1.1.2
# Online api libs dependencies
# zhipuai>=1.0.7 # zhipu
# dashscope>=1.10.0 # qwen
# volcengine>=1.0.106 # fangzhou
# uncomment libs if you want to use corresponding vector store
# pymilvus==2.1.3 # requires milvus==2.1.3
# psycopg2
zhipuai>=1.0.7
dashscope>=1.13.6
volcengine>=1.0.119
pymilvus>=2.3.4
# psycopg2==2.9.9
# pgvector>=0.2.4
# Agent and Search Tools
arxiv>=2.0.0
youtube-search>=2.1.2
duckduckgo-search>=3.9.9
duckduckgo-search>=4.1.0
metaphor-python>=0.1.23
# WebUI requirements
streamlit>=1.29.0 # do remember to add streamlit to environment variables if you use windows
streamlit>=1.29.0
streamlit-option-menu>=0.3.6
streamlit-antd-components>=0.2.4
streamlit-chatbox>=1.1.11
streamlit-modal>=0.1.0
streamlit-aggrid>=0.3.4.post3
httpx[brotli,http2,socks]>=0.25.2
watchdog>=3.0.0

70
requirements_api.txt
View File

@ -1,66 +1,64 @@
# API requirements
langchain==0.0.344
langchain-experimental>=0.0.42
torch>=2.1.2
torchvision>=0.16.2
torchaudio>=2.1.2
langchain>=0.0.352
langchain-experimental>=0.0.47
pydantic==1.10.13
fschat>=0.2.33
xformers>=0.0.22.post7
sentence_transformers
transformers>=4.35.2
torch==2.1.0 ##on Windows system, install the cuda version manually from https://pytorch.org/
torchvision #on Windows system, install the cuda version manually from https://pytorch.org/
torchaudio #on Windows system, install the cuda version manually from https://pytorch.org/
openai>=1.3.7
fastapi>=0.104
sse_starlette
fschat>=0.2.34
xformers>=0.0.23.post1
openai>=1.6.0
sentence_transformers>=2.2.2
transformers>=4.36.2
fastapi>=0.105
nltk>=3.8.1
uvicorn>=0.24.0.post1
starlette~=0.27.0
unstructured[all-docs]==0.11.0
python-magic-bin; sys_platform == 'win32'
SQLAlchemy==2.0.19
faiss-cpu # `conda install faiss-gpu -c conda-forge` if you want to accelerate with gpus
faiss-cpu>=1.7.4
accelerate>=0.24.1
spacy>=3.7.2
PyMuPDF
rapidocr_onnxruntime
PyMuPDF>=1.23.8
rapidocr_onnxruntime>=1.3.8
requests>=2.31.0
pathlib>=1.0.1
pytest>=7.4.3
numexpr>=2.8.7
numexpr>=2.8.8
strsimpy>=0.2.1
markdownify>=0.11.6
tiktoken>=0.5.1
tiktoken>=0.5.2
tqdm>=4.66.1
websockets
numpy~=1.24.4
pandas~=2.0.3
websockets>=12.0
numpy~=1.26.2
pandas~=2.1.4
einops>=0.7.0
transformers_stream_generator==0.0.4
vllm==0.2.2; sys_platform == "linux"
httpx[brotli,http2,socks]>=0.25.2
transformers_stream_generator>=0.0.4
vllm>=0.2.6; sys_platform == "linux"
# optional document loaders
# rapidocr_paddle[gpu] # gpu accelleration for ocr of pdf and image files
jq # for .json and .jsonl files. suggest `conda install jq` on windows
# html2text # for .enex files
# beautifulsoup4 # for .mhtml files
# pysrt # for .srt files
rapidocr_paddle[gpu]>=1.3.0.post5
jq>=1.6.0
html2text
beautifulsoup4>=4.12.2
pysrt>=1.1.2
# Online api libs dependencies
# zhipuai>=1.0.7 # zhipu
# dashscope>=1.10.0 # qwen
# volcengine>=1.0.106 # fangzhou
# uncomment libs if you want to use corresponding vector store
# pymilvus==2.1.3 # requires milvus==2.1.3
# psycopg2
zhipuai>=1.0.7，<=2.0.0
dashscope>=1.13.6
volcengine>=1.0.119
pymilvus>=2.3.4
# psycopg2==2.9.9
# pgvector>=0.2.4
# Agent and Search Tools
arxiv>=2.0.0
youtube-search>=2.1.2
duckduckgo-search>=3.9.9
duckduckgo-search>=4.1.0
metaphor-python>=0.1.23

23
requirements_lite.txt
View File

@ -3,15 +3,9 @@
langchain==0.0.344
langchain-experimental>=0.0.42
pydantic==1.10.13
fschat>=0.2.33
# xformers>=0.0.22.post7
# sentence_transformers
# transformers>=4.35.2
# torch==2.1.0 ##on Windows system, install the cuda version manually from https://pytorch.org/
# torchvision #on Windows system, install the cuda version manually from https://pytorch.org/
# torchaudio #on Windows system, install the cuda version manually from https://pytorch.org/
openai>=1.3.7
fastapi>=0.104
fschat>=0.2.34
openai>=1.6.0
fastapi>=0.105
sse_starlette
nltk>=3.8.1
uvicorn>=0.24.0.post1
@ -20,25 +14,16 @@ unstructured[docx,csv]==0.11.0 # add pdf if need
python-magic-bin; sys_platform == 'win32'
SQLAlchemy==2.0.19
faiss-cpu # `conda install faiss-gpu -c conda-forge` if you want to accelerate with gpus
# accelerate>=0.24.1
# spacy>=3.7.2
# PyMuPDF
# rapidocr_onnxruntime
requests>=2.31.0
pathlib>=1.0.1
pytest>=7.4.3
numexpr>=2.8.7
strsimpy>=0.2.1
markdownify>=0.11.6
# tiktoken>=0.5.1
tqdm>=4.66.1
websockets
numpy~=1.24.4
pandas~=2.0.3
# einops>=0.7.0
# transformers_stream_generator==0.0.4
# vllm==0.2.2; sys_platform == "linux"
httpx[brotli,http2,socks]>=0.25.2
# optional document loaders
# rapidocr_paddle[gpu] # gpu accelleration for ocr of pdf and image files
@ -49,7 +34,7 @@ jq # for .json and .jsonl files. suggest `conda install jq` on windows
# Online api libs dependencies
zhipuai>=1.0.7 # zhipu
zhipuai>=1.0.7，<=2.0.0 # zhipu
dashscope>=1.10.0 # qwen
# volcengine>=1.0.106 # fangzhou

17
server/knowledge_base/kb_cache/base.py
View File

@ -2,7 +2,7 @@ from langchain.embeddings.base import Embeddings
from langchain.vectorstores.faiss import FAISS
import threading
from configs import (EMBEDDING_MODEL, CHUNK_SIZE,
logger, log_verbose)
logger, log_verbose)
from server.utils import embedding_device, get_model_path, list_online_embed_models
from contextlib import contextmanager
from collections import OrderedDict
@ -99,10 +99,10 @@ class CachePool:
return cache
def load_kb_embeddings(
self,
kb_name: str,
embed_device: str = embedding_device(),
default_embed_model: str = EMBEDDING_MODEL,
self,
kb_name: str,
embed_device: str = embedding_device(),
default_embed_model: str = EMBEDDING_MODEL,
) -> Embeddings:
from server.db.repository.knowledge_base_repository import get_kb_detail
from server.knowledge_base.kb_service.base import EmbeddingsFunAdapter
@ -129,7 +129,7 @@ class EmbeddingsPool(CachePool):
self.atomic.release()
if model == "text-embedding-ada-002": # openai text-embedding-ada-002
from langchain.embeddings.openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings(model_name=model,
embeddings = OpenAIEmbeddings(model=model,
openai_api_key=get_model_path(model),
chunk_size=CHUNK_SIZE)
elif 'bge-' in model:
@ -145,12 +145,13 @@ class EmbeddingsPool(CachePool):
query_instruction = ""
embeddings = HuggingFaceBgeEmbeddings(model_name=get_model_path(model),
model_kwargs={'device': device},
query_instruction=query_instruction)
query_instruction=query_instruction)
if model == "bge-large-zh-noinstruct": # bge large -noinstruct embedding
embeddings.query_instruction = ""
else:
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
embeddings = HuggingFaceEmbeddings(model_name=get_model_path(model), model_kwargs={'device': device})
embeddings = HuggingFaceEmbeddings(model_name=get_model_path(model),
model_kwargs={'device': device})
item.obj = embeddings
item.finish_loading()
else:

40
tests/agent/test_agent_function.py
View File

@ -1,40 +0,0 @@
import sys
import os
sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
from configs import LLM_MODELS, TEMPERATURE
from server.utils import get_ChatOpenAI
from langchain.chains import LLMChain
from langchain.agents import LLMSingleActionAgent, AgentExecutor
from server.agent.tools import tools, tool_names
from langchain.memory import ConversationBufferWindowMemory
memory = ConversationBufferWindowMemory(k=5)
model = get_ChatOpenAI(
model_name=LLM_MODELS[0],
temperature=TEMPERATURE,
)
from server.agent.custom_template import CustomOutputParser, prompt
output_parser = CustomOutputParser()
llm_chain = LLMChain(llm=model, prompt=prompt)
agent = LLMSingleActionAgent(
llm_chain=llm_chain,
output_parser=output_parser,
stop=["\nObservation:"],
allowed_tools=tool_names
)
agent_executor = AgentExecutor.from_agent_and_tools(agent=agent, tools=tools, memory=memory, verbose=True)
import pytest
@pytest.mark.parametrize("text_prompt",
["北京市朝阳区未来24小时天气如何？", # 天气功能函数
"计算 (2 + 2312312)/4 是多少？", # 计算功能函数
"翻译这句话成中文：Life is the art of drawing sufficient conclusions form insufficient premises."] # 翻译功能函数
)
def test_different_agent_function(text_prompt):
try:
text_answer = agent_executor.run(text_prompt)
assert text_answer is not None
except Exception as e:
pytest.fail(f"agent_function failed with {text_prompt}, error: {str(e)}")