- name
- pywayne-llm-chat-bot
- description
- LLM chat interface using OpenAI-compatible APIs with streaming support and session management. Use when working with pywayne.llm.chat_bot module for creating LLM chat instances with custom configuration, managing multi-turn conversations with history, handling streaming responses, or managing multiple chat sessions with ChatManager
Pywayne LLM Chat Bot
This module provides a synchronous LLM chat interface compatible with OpenAI APIs (including local servers like Ollama).
Quick Start
from pywayne.llm.chat_bot import LLMChat
# Create chat instance
chat = LLMChat(
base_url="https://api.example.com/v1",
api_key="your_api_key",
model="deepseek-chat"
)
# Single-turn conversation (non-streaming)
response = chat.ask("Hello, LLM!", stream=False)
print(response)
# Streaming response
for token in chat.ask("Explain recursion", stream=True):
print(token, end='', flush=True)
Multi-turn Conversation
# Use chat() for history tracking
for token in chat.chat("What is a class in Python?"):
print(token, end='', flush=True)
# Continuation - remembers previous context
for token in chat.chat("How do I define a constructor?"):
print(token, end='', flush=True)
# View history
for msg in chat.history:
print(f"{msg['role']}: {msg['content']}")
# Clear history
chat.clear_history()
Configuration
LLMConfig Class
from pywayne.llm.chat_bot import LLMConfig
config = LLMConfig(
base_url="https://api.example.com/v1",
api_key="your_api_key",
model="deepseek-chat",
temperature=0.7,
max_tokens=8192,
top_p=1.0,
frequency_penalty=0.0,
presence_penalty=0.0,
system_prompt="You are a helpful assistant"
)
chat = LLMChat(**config.to_dict())
Dynamic System Prompt Update
chat.update_system_prompt("You are now a Python expert, provide code examples")
Managing Multiple Sessions
from pywayne.llm.chat_bot import ChatManager
manager = ChatManager(
base_url="https://api.example.com/v1",
api_key="your_api_key",
model="deepseek-chat",
timeout=300 # Session timeout in seconds
)
# Get or create chat instance (maintains per-session history)
chat1 = manager.get_chat("user1")
chat2 = manager.get_chat("user2")
# Sessions are independent
chat1.chat("Hello from user1")
chat2.chat("Hello from user2")
# Remove a session
manager.remove_chat("user1")
Custom Configuration per Session
custom_config = LLMConfig(
base_url=base_url,
api_key=api_key,
model="deepseek-chat",
temperature=0.9,
system_prompt="You are a creative writer"
)
chat3 = manager.get_chat("user3", config=custom_config)
API Reference
LLMChat
| Method | Description |
|---|
ask(prompt, stream=False) | Single-turn conversation without history |
chat(prompt, stream=True) | Multi-turn conversation with history tracking |
update_system_prompt(prompt) | Update system prompt in-place |
clear_history() | Clear conversation history (keeps system prompt) |
history (property) | Get copy of current conversation history |
ChatManager
| Method | Description |
|---|
get_chat(chat_id, stream=True, config=None) | Get or create chat instance by ID |
remove_chat(chat_id) | Remove chat session |
Parameters
| Parameter | Default | Description |
|---|
base_url | required | API base URL (e.g., https://api.deepseek.com/v1) |
api_key | required | API authentication key |
model | "deepseek-chat" | Model name |
temperature | 0.7 | Controls randomness (0-2) |
max_tokens | 2048/8192 | Maximum output tokens |
top_p | 1.0 | Nucleus sampling (0-1) |
frequency_penalty | 0.0 | Reduces repetition (-2 to 2) |
presence_penalty | 0.0 | Encourages new topics (-2 to 2) |
system_prompt | "你是一个严谨的助手" | System message |
timeout | inf | Session timeout in seconds (ChatManager only) |
