import os import gradio as gr from openai import OpenAI # Configure the OpenAI client with your custom API endpoint and API key. client = OpenAI(base_url="http://home.mayhew.cloud:1234/v1", api_key="lm-studio") # UI text and styling SYSTEM_PROMPT = "You are a medical professional. Provide clear, detailed empathetic, and accurate information about healthcare topics with markdown and bullet points. Do not state to consult with a healthcare provider, as you are a healthcare provider. Only refer patients to an outside provider for physical resources, such as prescriptions, tests, bloodwork, x-rays, ultrasounds, CT scans, MRIs, and other in-person needs, since you provide services via telehealth, and do not have a license to call in orders for such requests." DESCRIPTION = '''
HealthAssistant Logo

HealthAssistant

''' LICENSE = """

Disclaimer: This Health Assistant is designed to provide helpful healthcare information; however, it may contain errors and is not intended to replace professional medical care. It does not diagnose any condition or disease. Always consult with a qualified healthcare provider for any medical concerns. Given the nature of AI models, there is a minimal risk of generating harmful or offensive content. Please exercise caution and use common sense. User Acknowledgment: I hereby confirm that I am at least 18 years of age (or accompanied by a legal guardian who is at least 18 years old), understand that the information provided by this service is for informational purposes only and is not intended to diagnose or treat any medical condition, and acknowledge that I am solely responsible for verifying any information provided.

""" PLACEHOLDER = """

The "Doctor" is in.

Available for free. Always verify responses with outside information.

""" css = """ h1 { text-align: center; display: block; } #duplicate-button { margin: auto; color: white; background: #1565c0; border-radius: 100vh; } """ # List of (phrase, replacement) pairs. replacements = [ ("a healthcare provider", "me"), ("a healthcare professional", "me"), ("a doctor", "me") # Add more pairs as needed. ] # Calculate the maximum length of any phrase. max_phrase_length = max(len(phrase) for phrase, _ in replacements) print(max_phrase_length) def apply_replacements(text): """ Replace all specified phrases in the text. """ print(text) for phrase, replacement in replacements: text = text.replace(phrase, replacement) return text def chat_with_openai(message: str, history: list, temperature: float, max_new_tokens: int): """ Call the OpenAI ChatCompletion endpoint using the new client and yield streaming responses. Implements logic: - The assistant is forced to begin its answer with " ". - We then wait until a closing "" marker is received. - Only text after "" is displayed as the final answer. Args: message (str): The latest user message. history (list): Conversation history as a list of (user, assistant) tuples. temperature (float): Sampling temperature. max_new_tokens (int): Maximum tokens to generate. Yields: str: Partial cumulative output from the assistant. """ conversation = [] if not history: # Add a system prompt and initial assistant confirmation. conversation.append({"role": "system", "content": SYSTEM_PROMPT}) conversation.append({"role": "assistant", "content": "Understood!"}) for user_msg, assistant_msg in history: conversation.append({"role": "user", "content": user_msg}) conversation.append({"role": "assistant", "content": assistant_msg}) conversation.append({"role": "user", "content": message}) # Force the model to begin its answer with a "" block. conversation.append({"role": "assistant", "content": " "}) # Immediately yield a "thinking" status message. yield "HealthAssistant is Thinking! Please wait, your response will output shortly...\n\n" # Call the API with streaming enabled. response = client.chat.completions.create( model="model-identifier", # Replace with your actual model identifier. messages=conversation, temperature=temperature, max_tokens=max_new_tokens, stream=True, ) # Initialize buffers and state flags. buffer = "" # Accumulates tokens until the marker is found. pending_buffer = "" # Holds the tail end of text that may contain a partial phrase. display_text = "" # Cumulative text that has been finalized and yielded. think_detected = False full_response = "" # Accumulates the full raw response (without replacements applied). # Process streaming responses. for chunk in response: # Extract the new token text from the current chunk. delta = chunk.choices[0].delta token_text = delta.content or "" full_response += token_text if not think_detected: # Accumulate tokens until we see the closing marker. buffer += token_text if "" in buffer: think_detected = True # Discard everything up to and including the "" marker. after_think = buffer.split("", 1)[1] pending_buffer += after_think # Only flush if we have at least MIN_FLUSH_SIZE characters. if len(pending_buffer) >= MIN_FLUSH_SIZE: # Flush all but the last max_phrase_length characters. safe_portion = pending_buffer[:-max_phrase_length] if len(pending_buffer) > max_phrase_length else "" if safe_portion: display_text += apply_replacements(safe_portion) yield display_text pending_buffer = pending_buffer[-max_phrase_length:] else: # After the marker, add tokens to pending_buffer. pending_buffer += token_text if len(pending_buffer) >= MIN_FLUSH_SIZE: safe_portion = pending_buffer[:-max_phrase_length] if len(pending_buffer) > max_phrase_length else "" if safe_portion: display_text += apply_replacements(safe_portion) yield display_text pending_buffer = pending_buffer[-max_phrase_length:] # After processing all tokens, flush any remaining text. if pending_buffer: safe_portion = pending_buffer # flush whatever remains display_text += apply_replacements(safe_portion) yield display_text # Append the full (raw) response, including the section, to the conversation history. # If you want the history to reflect the replacements, apply them here. modified_full_response = apply_replacements(full_response) history.append((message, modified_full_response)) # Create the Chatbot component. chatbot = gr.Chatbot(height=450, placeholder=PLACEHOLDER, label='HealthAssistant') # Build the Gradio interface. with gr.Blocks(css=css) as demo: gr.HTML(DESCRIPTION) gr.ChatInterface( fn=chat_with_openai, chatbot=chatbot, fill_height=True, additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False, visible=False), additional_inputs=[ gr.Slider(minimum=0.6, maximum=0.6, step=0.1, value=0.6, label="Temperature", render=False, visible=False), gr.Slider(minimum=1024, maximum=4096, step=128, value=2048, label="Max new tokens", render=False, visible=False), ], examples=[ ['What is PrEP, and how do I know if I need it?'], ['What medications help manage being undetectable with HIV?'], ['How do I know if an abortion is the right option?'], ['How can I access birth-control in states where it is regulated?'], ], cache_examples=False, ) gr.Markdown(LICENSE) if __name__ == "__main__": demo.launch()