240908-联合DBGPT与Ollama实现RAG本地知识检索增强

A. 最终效果

B. 配景说明

• DBGPT在0.5.6版本中开始支持Ollama：v0.5.6 版本更新
  
  
  
  
• 网友对其Web端及界面端的设置进行了分享：
  
  
  
  • feat(model): support ollama as an optional llm & embedding proxy by GITHUBear · Pull Request #1475 · eosphoros-ai/DB-GPT
    
  • DB-GPT+Ollama构建本地智能数据平台_db-gpt ollama-CSDN博客
    
  
  

C. 情况设置

• 参考官网教程完成情况设置
  
  
  
  

D. 设置文件

• ⚠️：留意下面带⭐️的操作
  

1. #*******************************************************************#
2. #**             DB-GPT  - GENERAL SETTINGS                        **#  
3. #*******************************************************************#
5. #*******************************************************************#
6. #**                        Webserver Port                         **#
7. #*******************************************************************#
8. # DBGPT_WEBSERVER_PORT=5670
9. ## Whether to enable the new web UI, enabled by default,False use old ui
10. # USE_NEW_WEB_UI=True
11. #*******************************************************************#
12. #***                       LLM PROVIDER                          ***#
13. #*******************************************************************#
15. # TEMPERATURE=0
17. #*******************************************************************#
18. #**                         LLM MODELS                            **#
19. #*******************************************************************#
20. # ⭐️ 添加Ollama配置
21. LLM_MODEL=ollama_proxyllm
22. PROXY_SERVER_URL=http://127.0.0.1:11434
23. PROXYLLM_BACKEND="qwen2:1.5b"
24. PROXY_API_KEY=not_used
25. EMBEDDING_MODEL=proxy_ollama
26. proxy_ollama_proxy_server_url=http://127.0.0.1:11434
27. proxy_ollama_proxy_backend="nomic-embed-text:latest"
29. # LLM_MODEL=ollama_proxyllm
30. # MODEL_SERVER=http://127.0.0.1:11434
31. # PROXYLLM_BACKEND=llama3.1:8b
32. # EMBEDDING_MODEL=proxy_ollama
33. # proxy_ollama_proxy_server_url=http://127.0.0.1:11434
34. # proxy_ollama_proxy_backend=llama3.1:8b
36. # LLM_MODEL=ollama_proxyllm
37. # PROXY_SERVER_URL=http://127.0.0.1:11434
38. # PROXYLLM_BACKEND="qwen:0.5b"
39. # PROXY_API_KEY=not_used
40. # EMBEDDING_MODEL=proxy_ollama
41. # proxy_ollama_proxy_server_url=http://127.0.0.1:11434
42. # proxy_ollama_proxy_backend="nomic-embed-text:latest"   
46. # # LLM_MODEL, see dbgpt/configs/model_config.LLM_MODEL_CONFIG
47. # LLM_MODEL=glm-4-9b-chat
48. # ## LLM model path, by default, DB-GPT will read the model path from LLM_MODEL_CONFIG based on the LLM_MODEL.
49. # ## Of course you can specify your model path according to LLM_MODEL_PATH
50. # ## In DB-GPT, the priority from high to low to read model path:
51. # ##    1. environment variable with key: {LLM_MODEL}_MODEL_PATH (Avoid multi-model conflicts)
52. # ##    2. environment variable with key: MODEL_PATH
53. # ##    3. environment variable with key: LLM_MODEL_PATH
54. # ##    4. the config in dbgpt/configs/model_config.LLM_MODEL_CONFIG
55. # # LLM_MODEL_PATH=/app/models/glm-4-9b-chat
56. # # LLM_PROMPT_TEMPLATE=vicuna_v1.1
57. # MODEL_SERVER=http://127.0.0.1:8000
58. # LIMIT_MODEL_CONCURRENCY=5
59. # MAX_POSITION_EMBEDDINGS=4096
60. # QUANTIZE_QLORA=True
61. # QUANTIZE_8bit=True
62. # # QUANTIZE_4bit=False
63. # ## SMART_LLM_MODEL - Smart language model (Default: vicuna-13b)
64. # ## FAST_LLM_MODEL - Fast language model (Default: chatglm-6b)
65. # # SMART_LLM_MODEL=vicuna-13b
66. # # FAST_LLM_MODEL=chatglm-6b
67. # ## Proxy llm backend, this configuration is only valid when "LLM_MODEL=proxyllm", When we use the rest API provided by deployment frameworks like fastchat as a proxyllm,
68. # ## "PROXYLLM_BACKEND" is the model they actually deploy. We can use "PROXYLLM_BACKEND" to load the prompt of the corresponding scene.
69. # # PROXYLLM_BACKEND=
71. # ### You can configure parameters for a specific model with {model name}_{config key}=xxx
72. # ### See dbgpt/model/parameter.py
73. # ## prompt template for current model
74. # # llama_cpp_prompt_template=vicuna_v1.1
75. # ## llama-2-70b must be 8
76. # # llama_cpp_n_gqa=8
77. # ## Model path
78. # # llama_cpp_model_path=/data/models/TheBloke/vicuna-13B-v1.5-GGUF/vicuna-13b-v1.5.Q4_K_M.gguf
80. # ### LLM cache
81. # ## Enable Model cache
82. # # MODEL_CACHE_ENABLE=True
83. # ## The storage type of model cache, now supports: memory, disk
84. # # MODEL_CACHE_STORAGE_TYPE=disk
85. # ## The max cache data in memory, we always store cache data in memory fist for high speed.
86. # # MODEL_CACHE_MAX_MEMORY_MB=256
87. # ## The dir to save cache data, this configuration is only valid when MODEL_CACHE_STORAGE_TYPE=disk
88. # ## The default dir is pilot/data/model_cache
89. # # MODEL_CACHE_STORAGE_DISK_DIR=
91. #*******************************************************************#
92. #**                         EMBEDDING SETTINGS                    **#
93. #*******************************************************************#
94. # ⭐️ 取消非Ollama的Embedding设置
95. # EMBEDDING_MODEL=text2vec
96. # #EMBEDDING_MODEL=m3e-large
97. # #EMBEDDING_MODEL=bge-large-en
98. # #EMBEDDING_MODEL=bge-large-zh
99. # KNOWLEDGE_CHUNK_SIZE=500
100. # KNOWLEDGE_SEARCH_TOP_SIZE=5
101. # KNOWLEDGE_GRAPH_SEARCH_TOP_SIZE=200
102. # ## Maximum number of chunks to load at once, if your single document is too large,
103. # ## you can set this value to a higher value for better performance.
104. # ## if out of memory when load large document, you can set this value to a lower value.
105. # # KNOWLEDGE_MAX_CHUNKS_ONCE_LOAD=10
106. # #KNOWLEDGE_CHUNK_OVERLAP=50
107. # # Control whether to display the source document of knowledge on the front end.
108. # KNOWLEDGE_CHAT_SHOW_RELATIONS=False
109. # # Whether to enable Chat Knowledge Search Rewrite Mode
110. # KNOWLEDGE_SEARCH_REWRITE=False
111. # ## EMBEDDING_TOKENIZER   - Tokenizer to use for chunking large inputs
112. # ## EMBEDDING_TOKEN_LIMIT - Chunk size limit for large inputs
113. # # EMBEDDING_MODEL=all-MiniLM-L6-v2
114. # # EMBEDDING_TOKENIZER=all-MiniLM-L6-v2
115. # # EMBEDDING_TOKEN_LIMIT=8191
117. # ## Openai embedding model, See dbgpt/model/parameter.py
118. # # EMBEDDING_MODEL=proxy_openai
119. # # proxy_openai_proxy_server_url=https://api.openai.com/v1
120. # # proxy_openai_proxy_api_key={your-openai-sk}
121. # # proxy_openai_proxy_backend=text-embedding-ada-002
124. # ## qwen embedding model, See dbgpt/model/parameter.py
125. # # EMBEDDING_MODEL=proxy_tongyi
126. # # proxy_tongyi_proxy_backend=text-embedding-v1
127. # # proxy_tongyi_proxy_api_key={your-api-key}
129. # ## qianfan embedding model, See dbgpt/model/parameter.py
130. # #EMBEDDING_MODEL=proxy_qianfan
131. # #proxy_qianfan_proxy_backend=bge-large-zh
132. # #proxy_qianfan_proxy_api_key={your-api-key}
133. # #proxy_qianfan_proxy_api_secret={your-secret-key}
136. # ## Common HTTP embedding model
137. # # EMBEDDING_MODEL=proxy_http_openapi
138. # # proxy_http_openapi_proxy_server_url=http://localhost:8100/api/v1/embeddings
139. # # proxy_http_openapi_proxy_api_key=1dce29a6d66b4e2dbfec67044edbb924
140. # # proxy_http_openapi_proxy_backend=text2vec
142. #*******************************************************************#
143. #**                         RERANK SETTINGS                       **#
144. #*******************************************************************#
145. ## Rerank model
146. # RERANK_MODEL=bge-reranker-base
147. ## If you not set RERANK_MODEL_PATH, DB-GPT will read the model path from EMBEDDING_MODEL_CONFIG based on the RERANK_MODEL.
148. # RERANK_MODEL_PATH=
149. ## The number of rerank results to return
150. # RERANK_TOP_K=3
152. ## Common HTTP rerank model
153. # RERANK_MODEL=rerank_proxy_http_openapi
154. # rerank_proxy_http_openapi_proxy_server_url=http://127.0.0.1:8100/api/v1/beta/relevance
155. # rerank_proxy_http_openapi_proxy_api_key={your-api-key}
156. # rerank_proxy_http_openapi_proxy_backend=bge-reranker-base
161. #*******************************************************************#
162. #**                  DB-GPT METADATA DATABASE SETTINGS            **#
163. #*******************************************************************#
164. ### SQLite database (Current default database)
165. LOCAL_DB_TYPE=sqlite
167. ### MYSQL database
168. # LOCAL_DB_TYPE=mysql
169. # LOCAL_DB_USER=root
170. # LOCAL_DB_PASSWORD={your_password}
171. # LOCAL_DB_HOST=127.0.0.1
172. # LOCAL_DB_PORT=3306
173. # LOCAL_DB_NAME=dbgpt
174. ### This option determines the storage location of conversation records. The default is not configured to the old version of duckdb. It can be optionally db or file (if the value is db, the database configured by LOCAL_DB will be used)
175. #CHAT_HISTORY_STORE_TYPE=db
177. #*******************************************************************#
178. #**                         COMMANDS                              **#
179. #*******************************************************************#
180. EXECUTE_LOCAL_COMMANDS=False
182. #*******************************************************************#
183. #**            VECTOR STORE / KNOWLEDGE GRAPH SETTINGS            **#
184. #*******************************************************************#
185. VECTOR_STORE_TYPE=Chroma
186. GRAPH_STORE_TYPE=TuGraph
187. GRAPH_COMMUNITY_SUMMARY_ENABLED=True
188. KNOWLEDGE_GRAPH_EXTRACT_SEARCH_TOP_SIZE=5
189. KNOWLEDGE_GRAPH_EXTRACT_SEARCH_RECALL_SCORE=0.3
190. KNOWLEDGE_GRAPH_COMMUNITY_SEARCH_TOP_SIZE=20
191. KNOWLEDGE_GRAPH_COMMUNITY_SEARCH_RECALL_SCORE=0.0
193. ### Chroma vector db config
194. #CHROMA_PERSIST_PATH=/root/DB-GPT/pilot/data
196. ### Milvus vector db config
197. #VECTOR_STORE_TYPE=Milvus
198. #MILVUS_URL=127.0.0.1
199. #MILVUS_PORT=19530
200. #MILVUS_USERNAME
201. #MILVUS_PASSWORD
202. #MILVUS_SECURE=
204. ### Weaviate vector db config
205. #VECTOR_STORE_TYPE=Weaviate
206. #WEAVIATE_URL=https://kt-region-m8hcy0wc.weaviate.network
208. ## ElasticSearch vector db config
209. #VECTOR_STORE_TYPE=ElasticSearch
210. ElasticSearch_URL=127.0.0.1
211. ElasticSearch_PORT=9200
212. ElasticSearch_USERNAME=elastic
213. ElasticSearch_PASSWORD={your_password}
215. ### TuGraph config
216. #TUGRAPH_HOST=127.0.0.1
217. #TUGRAPH_PORT=7687
218. #TUGRAPH_USERNAME=admin
219. #TUGRAPH_PASSWORD=73@TuGraph
220. #TUGRAPH_VERTEX_TYPE=entity
221. #TUGRAPH_EDGE_TYPE=relation
222. #TUGRAPH_PLUGIN_NAMES=leiden
224. #*******************************************************************#
225. #**                  WebServer Language Support                   **#
226. #*******************************************************************#
227. # en, zh, fr, ja, ko, ru
228. LANGUAGE=en
229. #LANGUAGE=zh
232. #*******************************************************************#
233. # **    PROXY_SERVER (openai interface | chatGPT proxy service), use chatGPT as your LLM.
234. # ⭐️ 注释掉Ollama之外的PROXY_SERVER_URL
235. # ** if your server can visit openai, please set PROXY_SERVER_URL=https://api.openai.com/v1/chat/completions
236. # ** else if you have a chatgpt proxy server, you can set PROXY_SERVER_URL={your-proxy-serverip:port/xxx}
237. #*******************************************************************#
238. # PROXY_API_KEY={your-openai-sk}
239. # PROXY_SERVER_URL=https://api.openai.com/v1/chat/completions
241. # # from https://bard.google.com/     f12-> application-> __Secure-1PSID
242. # BARD_PROXY_API_KEY={your-bard-token}
244. #*******************************************************************#
245. # **  PROXY_SERVER +                                              **#
246. #*******************************************************************#
248. # Aliyun tongyi
249. TONGYI_PROXY_API_KEY={your-tongyi-sk}
251. ## Baidu wenxin
252. #WEN_XIN_MODEL_VERSION={version}
253. #WEN_XIN_API_KEY={your-wenxin-sk}
254. #WEN_XIN_API_SECRET={your-wenxin-sct}
256. ## Zhipu
257. #ZHIPU_MODEL_VERSION={version}
258. #ZHIPU_PROXY_API_KEY={your-zhipu-sk}
260. ## Baichuan
261. #BAICHUN_MODEL_NAME={version}
262. #BAICHUAN_PROXY_API_KEY={your-baichuan-sk}
263. #BAICHUAN_PROXY_API_SECRET={your-baichuan-sct}
265. # Xunfei Spark
266. #XUNFEI_SPARK_API_VERSION={version}
267. #XUNFEI_SPARK_APPID={your_app_id}
268. #XUNFEI_SPARK_API_KEY={your_api_key}
269. #XUNFEI_SPARK_API_SECRET={your_api_secret}
271. ## Yi Proxyllm, https://platform.lingyiwanwu.com/docs
272. #YI_MODEL_VERSION=yi-34b-chat-0205
273. #YI_API_BASE=https://api.lingyiwanwu.com/v1
274. #YI_API_KEY={your-yi-api-key}
276. ## Moonshot Proxyllm, https://platform.moonshot.cn/docs/
277. # MOONSHOT_MODEL_VERSION=moonshot-v1-8k
278. # MOONSHOT_API_BASE=https://api.moonshot.cn/v1
279. # MOONSHOT_API_KEY={your-moonshot-api-key}
281. ## Deepseek Proxyllm, https://platform.deepseek.com/api-docs/
282. # DEEPSEEK_MODEL_VERSION=deepseek-chat
283. # DEEPSEEK_API_BASE=https://api.deepseek.com/v1
284. # DEEPSEEK_API_KEY={your-deepseek-api-key}
287. #*******************************************************************#
288. #**    SUMMARY_CONFIG                                             **#
289. #*******************************************************************#
290. SUMMARY_CONFIG=FAST
292. #*******************************************************************#
293. #**    MUlti-GPU                                                  **#
294. #*******************************************************************#
295. ## See https://developer.nvidia.com/blog/cuda-pro-tip-control-gpu-visibility-cuda_visible_devices/
296. ## If CUDA_VISIBLE_DEVICES is not configured, all available gpus will be used
297. # CUDA_VISIBLE_DEVICES=0
298. ## You can configure the maximum memory used by each GPU.
299. # MAX_GPU_MEMORY=16Gib
301. #*******************************************************************#
302. #**                         LOG                                   **#
303. #*******************************************************************#
304. # FATAL, ERROR, WARNING, WARNING, INFO, DEBUG, NOTSET
305. DBGPT_LOG_LEVEL=INFO
306. # LOG dir, default: ./logs
307. #DBGPT_LOG_DIR=
310. #*******************************************************************#
311. #**                         API_KEYS                              **#
312. #*******************************************************************#
313. # API_KEYS - The list of API keys that are allowed to access the API. Each of the below are an option, separated by commas.
314. # API_KEYS=dbgpt
316. #*******************************************************************#
317. #**                         ENCRYPT                               **#
318. #*******************************************************************#
319. # ENCRYPT KEY - The key used to encrypt and decrypt the data
320. # ENCRYPT_KEY=your_secret_key
322. #*******************************************************************#
323. #**                         File Server                           **#
324. #*******************************************************************#
325. ## The local storage path of the file server, the default is pilot/data/file_server
326. # FILE_SERVER_LOCAL_STORAGE_PATH =
328. #*******************************************************************#
329. #**                     Application Config                        **#
330. #*******************************************************************#
331. ## Non-streaming scene retries
332. # DBGPT_APP_SCENE_NON_STREAMING_RETRIES_BASE=1
333. ## Non-streaming scene parallelism
334. # DBGPT_APP_SCENE_NON_STREAMING_PARALLELISM_BASE=1
336. #*******************************************************************#
337. #**                   Observability Config                        **#
338. #*******************************************************************#
339. ## Whether to enable DB-GPT send trace to OpenTelemetry
340. # TRACER_TO_OPEN_TELEMETRY=False
341. ## Following configurations are only valid when TRACER_TO_OPEN_TELEMETRY=True
342. ## More details see https://opentelemetry-python.readthedocs.io/en/latest/exporter/otlp/otlp.html
343. # OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://localhost:4317
344. # OTEL_EXPORTER_OTLP_TRACES_INSECURE=False
345. # OTEL_EXPORTER_OTLP_TRACES_CERTIFICATE=
346. # OTEL_EXPORTER_OTLP_TRACES_HEADERS=
347. # OTEL_EXPORTER_OTLP_TRACES_TIMEOUT=
348. # OTEL_EXPORTER_OTLP_TRACES_COMPRESSION=
350. #*******************************************************************#
351. #**                     FINANCIAL CHAT Config                     **#
352. #*******************************************************************#
353. # FIN_REPORT_MODEL=/app/models/bge-large-zh

复制代码

• 错误排查
  

E. 运行利用

E.1 启动

1. python dbgpt/app/dbgpt_server.py

复制代码

E.2 利用

F. 引文出处的设置

   在新版本中，引文出处转移到了应用，通过创建应用绑定知识库，然后在应用内里对话后就会显示出处。
  F.1 点击创建应用

F.2 绑定知识库

F.3 选择应用

F.4对话引文检察

免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！更多信息从访问主页：qidao123.com:ToB企服之家，中国第一个企服评测及商务社交产业平台。