app.py

import os
import warnings
from typing import *
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from langchain.llms import HuggingFacePipeline

from langgraph.checkpoint.memory import MemorySaver

from interface import create_demo
from medrax.agent import *
from medrax.tools import *
from medrax.utils import *

warnings.filterwarnings("ignore")

def initialize_agent(
    prompt_file,
    tools_to_use=None,
    model_dir="./model-weights",
    temp_dir="temp",
    device="cuda",
    temperature=0.7,
    top_p=0.95
):
    """Initialize the MedRAX agent with specified tools and configuration."""

    prompts = load_prompts_from_file(prompt_file)
    prompt = prompts["MEDICAL_ASSISTANT"]

    all_tools = {
        "ChestXRayClassifierTool": lambda: ChestXRayClassifierTool(device=device),
        "ChestXRaySegmentationTool": lambda: ChestXRaySegmentationTool(device=device),
        "LlavaMedTool": lambda: LlavaMedTool(cache_dir=model_dir, device=device, load_in_8bit=True),
        "XRayVQATool": lambda: XRayVQATool(cache_dir=model_dir, device=device),
        "ChestXRayReportGeneratorTool": lambda: ChestXRayReportGeneratorTool(
            cache_dir=model_dir, device=device
        ),
        "XRayPhraseGroundingTool": lambda: XRayPhraseGroundingTool(
            cache_dir=model_dir, temp_dir=temp_dir, load_in_8bit=True, device=device
        ),
        "ChestXRayGeneratorTool": lambda: ChestXRayGeneratorTool(
            model_path=f"{model_dir}/roentgen", temp_dir=temp_dir, device=device
        ),
        "ImageVisualizerTool": lambda: ImageVisualizerTool(),
        "DicomProcessorTool": lambda: DicomProcessorTool(temp_dir=temp_dir),
    }

    tools_dict = {}
    tools_to_use = tools_to_use or all_tools.keys()
    for tool_name in tools_to_use:
        if tool_name in all_tools:
            tools_dict[tool_name] = all_tools[tool_name]()

    checkpointer = MemorySaver()

    # Load local Hugging Face model
    hf_model_id = model_id = "TheBloke/Mistral-7B-Instruct-v0.2-GGUF"
    tokenizer = AutoTokenizer.from_pretrained(hf_model_id)
    raw_model = AutoModelForCausalLM.from_pretrained(hf_model_id, device_map="auto")

    pipe = pipeline(
        "text-generation",
        model=raw_model,
        tokenizer=tokenizer,
        max_new_tokens=512,
        temperature=temperature,`
        top_p=top_p,
        return_full_text=False,
    )

    model = HuggingFacePipeline(pipeline=pipe)

    agent = Agent(
        model,
        tools=list(tools_dict.values()),
        log_tools=True,
        log_dir="logs",
        system_prompt=prompt,
        checkpointer=checkpointer,
    )

    print("Agent initialized")
    return agent, tools_dict


if __name__ == "__main__":
    print("Starting server...")

    selected_tools = [
        "ImageVisualizerTool",
        "DicomProcessorTool",
        "ChestXRayClassifierTool",
        "ChestXRaySegmentationTool",
        "ChestXRayReportGeneratorTool",
        "XRayVQATool",
        # "LlavaMedTool",
        # "XRayPhraseGroundingTool",
        # "ChestXRayGeneratorTool",
    ]

    agent, tools_dict = initialize_agent(
        "medrax/docs/system_prompts.txt",
        tools_to_use=selected_tools,
        model_dir="./model-weights",
        temp_dir="temp",
        device="cuda",
        temperature=0.7,
        top_p=0.95
    )

    demo = create_demo(agent, tools_dict)
    demo.launch(debug=True, ssr_mode=False)