| import gradio as gr | |
| from ollama import chat, ChatResponse | |
| import subprocess | |
| def interact(message: str, history: list): | |
| message_dct = { | |
| "role": "user", | |
| "content": message | |
| } | |
| chat_history = [msg for msg in history] | |
| chat_history.append(message_dct) | |
| response: ChatResponse = chat( | |
| model="deepseek-r1:1.5b", | |
| messages=chat_history, | |
| stream=True | |
| ) | |
| text_response = "" | |
| thinking_response = gr.ChatMessage(content="", metadata={"title":"Thinking Cloud"}) | |
| thinking = False | |
| for chunk in response: | |
| bit = chunk["message"]["content"] | |
| if(bit == "<think>"): | |
| thinking = True | |
| continue | |
| elif(bit == "</think>"): | |
| thinking = False | |
| continue | |
| if(thinking): | |
| thinking_response.content += bit | |
| else: | |
| text_response += bit | |
| final_response = [thinking_response, text_response] | |
| yield final_response | |
| interface = gr.ChatInterface( | |
| fn=interact, | |
| type="messages", | |
| title="Deepseek-R1 Chat Interface" | |
| ) | |
| if __name__ == "__main__": | |
| subprocess.run(["ollama", "run", "deepseek-r1:1.5b"]) | |
| interface.launch() |