Self-host llmapi server

中文文档

llmapi-server is an abstract backend that encapsulates a variety of large language models (LLM, such as ChatGPT, GPT-3, GPT-4, etc.), and provides simple access services through OpenAPI

🌟 If it is helpful to you,please star it 🌟

graph LR

subgraph llmapi server
OpenAPI --> session
OpenAPI --> pre_post
subgraph backend
style backend fill:#f9f
pre_post-->chatgpt
pre_post-->dall-e
pre_post-->llama
pre_post-->...
end
end
text-->OpenAPI
image-->OpenAPI
embedding-->OpenAPI
others--> OpenAPI

• chatgpt: openai's official ChatGPT interface
• gpt3: openai's official GPT-3 interface
• gpt-embedding: openai's official Embedding interface
• dall-e: openai's official DALL·E interface
• welm: wechat's llm interface
• newbing: New Bing search based on ChatGPT（unofficial)

• llama
• stable diffusion
• controlNet
• SAM(meta)

1. run locally

# python >= 3.8
python3 -m pip install -r requirements.txt

python3 run_api_server.py

2. run with docker

./build_docker.sh
./start_docker.sh

1. Use the curl command to access:

# 1. Start a new session
curl -X POST -H "Content-Type: application/json" -d '{"bot_type":"mock"}' http://127.0.0.1:5050/v1/chat/start
# response sample: {"code":0,"msg":"Success","session":"123456"}

# 2. chat with LLMs
curl -X POST -H "Content-Type: application/json" -d '{"session":"123456","content":"hello"}' http://127.0.0.1:5050/v1/chat/ask
# response sample: {"code":0,"msg":"Success","reply":"Text mock reply for your prompt:hello","timestamp":1678865301.0842562}

# 3. Close the session and end chat
curl -X POST -H "Content-Type: application/json" -d '{"session":"123456"}' http://127.0.0.1:5050/v1/chat/end
# response: {"code":0,"msg":"Success"}

2. Using command line tools:llmapi_cli

llmapi_cli --host="http://127.0.0.1:5050" --bot=mock

3. Integrate in your python code with llmapi_cli module

from llmapi_cli import LLMClient

client = LLMClient(host = "http://127.0.0.1:5050", bot = "mock")

rep = client.ask("hello")

print(rep)

1. You need to create a new backend name in the backend directory (assumed to be newllm), you can directly cp -r mock newllm
2. Referring to the implementation of mock, change the backend name to newllm
3. In the newllm directory, add the necessary dependencies, and all related development is bound to this directory
4. Add support for newllm in backend.py