Dev (#1652)

* 更新上agent提示词代码 * 更新部分文档，修复了issue中提到的bge匹配超过1 的bug * 按需修改 * 解决了部分最新用户用依赖的bug，加了两个工具，移除google工具
chatchat-space · Oct 4, 2023 · fc6a3b0 · fc6a3b0
1 parent 332f8be
commit fc6a3b0
Show file tree

Hide file tree

Showing 11 changed files with 108 additions and 31 deletions.
diff --git a/README.md b/README.md
@@ -168,7 +168,8 @@ docker run -d --gpus all -p 80:8501 registry.cn-beijing.aliyuncs.com/chatchat/ch
 - [BAAI/bge-base-zh](https://huggingface.co/BAAI/bge-base-zh)
 - [BAAI/bge-large-zh](https://huggingface.co/BAAI/bge-large-zh)
 - [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5)
-- [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5)- [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5)
+- [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5)
+- [BAAI/bge-base-zh-v1.5](https://huggingface.co/BAAI/bge-base-zh-v1.5)
 - [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5)
 - [BAAI/bge-large-zh-noinstruct](https://huggingface.co/BAAI/bge-large-zh-noinstruct)
 - [sensenova/piccolo-base-zh](https://huggingface.co/sensenova/piccolo-base-zh)

diff --git a/configs/__init__.py b/configs/__init__.py
@@ -5,4 +5,4 @@
 from .prompt_config import *
 
 
-VERSION = "v0.2.5-preview"
+VERSION = "v0.2.6-preview"
diff --git a/requirements.txt b/requirements.txt
@@ -1,17 +1,17 @@
 langchain>=0.0.302
-fschat[model_worker]==0.2.29
+fschat[model_worker]==0.2.30
 openai
 sentence_transformers
 transformers>=4.33.0
 torch>=2.0.1
 torchvision
 torchaudio
-fastapi>=0.103.1
+fastapi>=0.103.2
 nltk~=3.8.1
 uvicorn~=0.23.1
 starlette~=0.27.0
 pydantic~=1.10.11
-unstructured[all-docs]>=0.10.4
+unstructured[all-docs]>=0.10.12
 python-magic-bin; sys_platform == 'win32'
 SQLAlchemy==2.0.19
 faiss-cpu
@@ -25,7 +25,7 @@ pathlib
 pytest
 scikit-learn
 numexpr
-vllm==0.1.7; sys_platform == "linux"
+vllm>=0.2.0; sys_platform == "linux"
 # online api libs
 # zhipuai
 # dashscope>=1.10.0 # qwen

diff --git a/requirements_api.txt b/requirements_api.txt
@@ -1,9 +1,9 @@
 langchain>=0.0.302
-fschat[model_worker]==0.2.29
+fschat[model_worker]==0.2.30
 openai
 sentence_transformers
-transformers>=4.33.0
-torch >=2.0.1
+transformers>=4.33.3
+torch>=2.0.1
 torchvision
 torchaudio
 fastapi>=0.103.1
@@ -24,14 +24,16 @@ pathlib
 pytest
 scikit-learn
 numexpr
-vllm==0.1.7; sys_platform == "linux"
+
+vllm>=0.2.0; sys_platform == "linux"
 
 
 # online api libs
 # zhipuai
 # dashscope>=1.10.0 # qwen
 # qianfan
 # volcengine>=1.0.106 # fangzhou
+# duckduckgo-searchd #duckduckgo搜索
 
 # uncomment libs if you want to use corresponding vector store
 # pymilvus==2.1.3 # requires milvus==2.1.3

diff --git a/server/agent/google_search.py b/server/agent/google_search.py
diff --git a/server/agent/search_internet.py b/server/agent/search_internet.py
@@ -0,0 +1,35 @@
+## 单独运行的时候需要添加
+import sys
+import os
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+import json
+from server.chat import search_engine_chat
+from configs import LLM_MODEL, TEMPERATURE, VECTOR_SEARCH_TOP_K, SCORE_THRESHOLD
+
+import asyncio
+
+
+async def search_engine_iter(query: str):
+    response = await search_engine_chat(query=query,
+                                         search_engine_name="bing",
+                                         model_name=LLM_MODEL,
+                                         temperature=TEMPERATURE,
+                                         history=[],
+                                         top_k = VECTOR_SEARCH_TOP_K,
+                                         prompt_name = "knowledge_base_chat",
+                                         stream=False)
+
+    contents = ""
+    async for data in response.body_iterator: # 这里的data是一个json字符串
+        data = json.loads(data)
+        contents = data["answer"]
+        docs = data["docs"]
+    return contents
+
+def search_internet(query: str):
+    return asyncio.run(search_engine_iter(query))
+
+
+if __name__ == "__main__":
+    result = search_internet("大数据男女比例")
+    print("答案:",result)
diff --git a/server/agent/search_knowledge.py b/server/agent/search_knowledge.py
@@ -0,0 +1,37 @@
+## 单独运行的时候需要添加
+import sys
+import os
+import json
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))))
+from server.chat.knowledge_base_chat import knowledge_base_chat
+from configs import LLM_MODEL, TEMPERATURE, VECTOR_SEARCH_TOP_K, SCORE_THRESHOLD
+
+import asyncio
+
+
+async def search_knowledge_base_iter(query: str):
+    response = await knowledge_base_chat(query=query,
+                                         knowledge_base_name="tcqa",
+                                         model_name=LLM_MODEL,
+                                         temperature=TEMPERATURE,
+                                         history=[],
+                                         top_k = VECTOR_SEARCH_TOP_K,
+                                         prompt_name = "knowledge_base_chat",
+                                         score_threshold = SCORE_THRESHOLD,
+                                         stream=False)
+
+    contents = ""
+    async for data in response.body_iterator: # 这里的data是一个json字符串
+        data = json.loads(data)
+        contents = data["answer"]
+        docs = data["docs"]
+    return contents
+
+def search_knowledge(query: str):
+    return asyncio.run(search_knowledge_base_iter(query))
+
+
+if __name__ == "__main__":
+    result = search_knowledge("大数据男女比例")
+    print("答案:",result)
diff --git a/server/agent/tools.py b/server/agent/tools.py
@@ -7,8 +7,9 @@
 from server.agent.translator import translate
 from server.agent.weather import weathercheck
 from server.agent.shell import shell
-from server.agent.google_search import google_search
 from langchain.agents import Tool
+from server.agent.search_knowledge import search_knowledge
+from server.agent.search_internet import search_internet
 
 tools = [
     Tool.from_function(
@@ -32,9 +33,15 @@
         description="使用命令行工具输出",
     ),
     Tool.from_function(
-        func=google_search,
-        name="谷歌搜索工具",
-        description="使用谷歌搜索",
-    )
+        func=search_knowledge,
+        name="知识库查询工具",
+        description="使用西交利物浦大学大数据专业的本专业数据库来解答问题",
+    ),
+    Tool.from_function(
+        func=search_internet,
+        name="互联网查询工具",
+        description="访问Bing互联网来解答问题",
+    ),
+
 ]
 tool_names = [tool.name for tool in tools]
diff --git a/server/agent/translator.py b/server/agent/translator.py
@@ -52,4 +52,4 @@ def translate(query: str):
     llm_translate = LLMChain(llm=model, prompt=PROMPT)
     ans = llm_translate.run(query)
 
-    return ans
+    return ans
diff --git a/startup.py b/startup.py
@@ -114,7 +114,7 @@ def create_model_worker_app(log_level: str = "INFO", **kwargs) -> FastAPI:
             args.block_size = 16
             args.swap_space = 4  # GiB
             args.gpu_memory_utilization = 0.90
-            args.max_num_batched_tokens = 2560
+            args.max_num_batched_tokens = 16384 # 一个批次中的最大令牌（tokens）数量，这个取决于你的显卡和大模型设置，设置太大显存会不够
             args.max_num_seqs = 256
             args.disable_log_stats = False
             args.conv_template = None
@@ -123,6 +123,13 @@ def create_model_worker_app(log_level: str = "INFO", **kwargs) -> FastAPI:
             args.num_gpus = 1 # vllm worker的切分是tensor并行，这里填写显卡的数量
             args.engine_use_ray = False
             args.disable_log_requests = False
+
+            # 0.2.0 vllm后要加的参数
+            args.max_model_len = 8192 # 模型可以处理的最大序列长度。请根据你的大模型设置，
+            args.revision = None
+            args.quantization = None
+            args.max_log_len = None
+
             if args.model_path:
                 args.model = args.model_path
             if args.num_gpus > 1:

diff --git a/webui_pages/dialogue/dialogue.py b/webui_pages/dialogue/dialogue.py
@@ -57,7 +57,7 @@ def on_mode_change():
                                       "搜索引擎问答",
                                       "自定义Agent问答",
                                       ],
-                                     index=1,
+                                     index=3,
                                      on_change=on_mode_change,
                                      key="dialogue_mode",
                                      )
@@ -101,9 +101,8 @@ def llm_model_format_func(x):
                     st.success(msg)
                     st.session_state["prev_llm_model"] = llm_model
 
-        temperature = st.slider("Temperature：", 0.0, 1.0, TEMPERATURE, 0.01)
+        temperature = st.slider("Temperature：", 0.0, 1.0, TEMPERATURE, 0.05)
 
-        ## 部分模型可以超过10抡对话
         history_len = st.number_input("历史对话轮数：", 0, 20, HISTORY_LEN)
 
         def on_kb_change():
@@ -119,10 +118,7 @@ def on_kb_change():
                     key="selected_kb",
                 )
                 kb_top_k = st.number_input("匹配知识条数：", 1, 20, VECTOR_SEARCH_TOP_K)
-
-                ## Bge 模型会超过1
                 score_threshold = st.slider("知识匹配分数阈值：", 0.0, 1.0, float(SCORE_THRESHOLD), 0.01)
-
                 # chunk_content = st.checkbox("关联上下文", False, disabled=True)
                 # chunk_size = st.slider("关联长度：", 0, 500, 250, disabled=True)
         elif dialogue_mode == "搜索引擎问答":