app.py

import os
import sys
import random
import gradio as gr

# Add `src` to import path
sys.path.append(os.path.join(os.path.dirname(__file__), "src"))

from txagent.txagent import TxAgent

# === Environment setup ===
current_dir = os.path.dirname(os.path.abspath(__file__))
os.environ["MKL_THREADING_LAYER"] = "GNU"
os.environ["TOKENIZERS_PARALLELISM"] = "false"

# === UI constants ===
DESCRIPTION = '''
<div>
<h1 style="text-align: center;">TxAgent: An AI Agent for Therapeutic Reasoning Across a Universe of Tools</h1>
</div>
'''

INTRO = "Precision therapeutics require multimodal adaptive models..."
LICENSE = "DISCLAIMER: THIS WEBSITE DOES NOT PROVIDE MEDICAL ADVICE..."

PLACEHOLDER = '''
<div style="padding: 30px; text-align: center;">
   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">TxAgent</h1>
   <p style="font-size: 18px;">Click clear 🗑️ before asking a new question.</p>
   <p style="font-size: 18px;">Click retry 🔄 to see another answer.</p>
</div>
'''

css = """
h1 { text-align: center; }
#duplicate-button {
  margin: auto;
  color: white;
  background: #1565c0;
  border-radius: 100vh;
}
.gradio-accordion {
    margin-top: 0px !important;
    margin-bottom: 0px !important;
}
"""

chat_css = """
.gr-button { font-size: 20px !important; }
.gr-button svg { width: 32px !important; height: 32px !important; }
"""

# === Model and tool config ===
model_name = "mims-harvard/TxAgent-T1-Llama-3.1-8B"
rag_model_name = "mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B"
new_tool_files = {
    "new_tool": os.path.join(current_dir, "data", "new_tool.json")
}

question_examples = [
    ["Given a 50-year-old patient experiencing severe acute pain and considering the use of the newly approved medication, Journavx, how should the dosage be adjusted considering moderate hepatic impairment?"],
    ["A 30-year-old patient is on Prozac for depression and now diagnosed with WHIM syndrome. Is Xolremdi suitable?"]
]

agent = None  # global agent initialized later


# === Create Gradio UI ===
def create_ui():
    with gr.Blocks(css=css) as demo:
        gr.Markdown(DESCRIPTION)
        gr.Markdown(INTRO)

        temperature = gr.Slider(0, 1, step=0.1, value=0.3, label="Temperature")
        max_new_tokens = gr.Slider(128, 4096, step=1, value=1024, label="Max New Tokens")
        max_tokens = gr.Slider(128, 32000, step=1, value=8192, label="Max Total Tokens")
        max_round = gr.Slider(1, 50, step=1, value=30, label="Max Rounds")
        multi_agent = gr.Checkbox(label="Enable Multi-agent Reasoning", value=False)
        conversation_state = gr.State([])

        chatbot = gr.Chatbot(
            label="TxAgent",
            placeholder=PLACEHOLDER,
            height=700,
            type="messages",
            show_copy_button=True
        )

        # === Retry handler (fixed)
        def handle_retry(history, retry_data, temperature, max_new_tokens, max_tokens, multi_agent, conversation, max_round):
            agent.update_parameters(seed=random.randint(0, 10000))
            new_history = history[:retry_data.index]
            prompt = history[retry_data.index]["content"]
            result = agent.run_gradio_chat(
                new_history + [{"role": "user", "content": prompt}],
                temperature, max_new_tokens, max_tokens,
                multi_agent, conversation, max_round
            )
            if hasattr(result, "__iter__") and not isinstance(result, (str, list, dict)):
                result = list(result)
            return result

        chatbot.retry(
            handle_retry,
            chatbot, chatbot,
            temperature, max_new_tokens, max_tokens,
            multi_agent, conversation_state, max_round
        )

        # === Chat handler
        def handle_chat(message, history, temperature, max_new_tokens, max_tokens, multi_agent, conversation, max_round):
            result = agent.run_gradio_chat(message, history, temperature, max_new_tokens, max_tokens, multi_agent, conversation, max_round)
            if hasattr(result, "__iter__") and not isinstance(result, (str, list, dict)):
                result = list(result)
            return result

        # === Chat Interface setup
        gr.ChatInterface(
            fn=handle_chat,
            chatbot=chatbot,
            additional_inputs=[
                temperature, max_new_tokens, max_tokens,
                multi_agent, conversation_state, max_round
            ],
            examples=question_examples,
            css=chat_css,
            cache_examples=False,
            fill_height=True,
            fill_width=True,
            stop_btn=True
        )

        gr.Markdown(LICENSE)
    return demo


# === HF Spaces + vLLM-safe entrypoint ===
if __name__ == "__main__":
    agent = TxAgent(
        model_name,
        rag_model_name,
        tool_files_dict=new_tool_files,
        force_finish=True,
        enable_checker=True,
        step_rag_num=10,
        seed=100,
        additional_default_tools=["DirectResponse", "RequireClarification"]
    )
    agent.init_model()

    demo = create_ui()
    demo.launch()