File size: 8,797 Bytes
a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 c8d430c a444494 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 |
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from typing import List, Optional, Dict
import gradio as gr
import json
from enum import Enum
import re
import os
import time
from huggingface_hub import hf_hub_download
# We'll import llama_cpp in a way that provides better error messages
try:
from llama_cpp import Llama
LLAMA_IMPORT_ERROR = None
except Exception as e:
LLAMA_IMPORT_ERROR = str(e)
print(f"Warning: Failed to import llama_cpp: {e}")
class ConsultationState(Enum):
INITIAL = "initial"
GATHERING_INFO = "gathering_info"
DIAGNOSIS = "diagnosis"
class Message(BaseModel):
role: str
content: str
class ChatRequest(BaseModel):
messages: List[Message]
class ChatResponse(BaseModel):
response: str
finished: bool
# Standard health assessment questions
HEALTH_ASSESSMENT_QUESTIONS = [
"What are your current symptoms and how long have you been experiencing them?",
"Do you have any pre-existing medical conditions or chronic illnesses?",
"Are you currently taking any medications? If yes, please list them.",
"Is there any relevant family medical history I should know about?",
"Have you had any similar symptoms in the past? If yes, what treatments worked?"
]
NURSE_OGE_IDENTITY = """
You are Nurse Oge, a medical AI assistant focused on serving patients in Nigeria. Always be empathetic,
professional, and thorough in your assessments. When asked about your identity, explain that you are
Nurse Oge, a medical AI assistant serving Nigerian communities. Remember that you must gather complete
health information before providing any medical advice.
"""
class NurseOgeAssistant:
def __init__(self):
if LLAMA_IMPORT_ERROR:
raise ImportError(f"Cannot initialize NurseOgeAssistant due to llama_cpp import error: {LLAMA_IMPORT_ERROR}")
# Download the model file
try:
model_path = hf_hub_download(
repo_id="mradermacher/Llama3-Med42-8B-GGUF",
filename="Llama3-Med42-8B.IQ3_M.gguf",
resume_download=True
)
# Initialize the model with the downloaded file
self.llm = Llama(
model_path=model_path,
n_ctx=2048, # Context window
n_threads=4 # Number of CPU threads to use
)
except Exception as e:
raise RuntimeError(f"Failed to initialize the model: {str(e)}")
self.consultation_states = {}
self.gathered_info = {}
# ... (rest of the NurseOgeAssistant class methods remain the same)
def _is_identity_question(self, message: str) -> bool:
identity_patterns = [
r"who are you",
r"what are you",
r"your name",
r"what should I call you",
r"tell me about yourself"
]
return any(re.search(pattern, message.lower()) for pattern in identity_patterns)
def _is_location_question(self, message: str) -> bool:
location_patterns = [
r"where are you",
r"which country",
r"your location",
r"where do you work",
r"where are you based"
]
return any(re.search(pattern, message.lower()) for pattern in location_patterns)
def _get_next_assessment_question(self, conversation_id: str) -> Optional[str]:
if conversation_id not in self.gathered_info:
self.gathered_info[conversation_id] = []
questions_asked = len(self.gathered_info[conversation_id])
if questions_asked < len(HEALTH_ASSESSMENT_QUESTIONS):
return HEALTH_ASSESSMENT_QUESTIONS[questions_asked]
return None
async def process_message(self, conversation_id: str, message: str, history: List[Dict]) -> ChatResponse:
# Initialize state if new conversation
if conversation_id not in self.consultation_states:
self.consultation_states[conversation_id] = ConsultationState.INITIAL
# Handle identity questions
if self._is_identity_question(message):
return ChatResponse(
response="I am Nurse Oge, a medical AI assistant dedicated to helping patients in Nigeria. "
"I'm here to provide medical guidance while ensuring I gather all necessary health information "
"for accurate assessments.",
finished=True
)
# Handle location questions
if self._is_location_question(message):
return ChatResponse(
response="I am based in Nigeria and specifically trained to serve Nigerian communities, "
"taking into account local healthcare contexts and needs.",
finished=True
)
# Start health assessment if it's a medical query
if self.consultation_states[conversation_id] == ConsultationState.INITIAL:
self.consultation_states[conversation_id] = ConsultationState.GATHERING_INFO
next_question = self._get_next_assessment_question(conversation_id)
return ChatResponse(
response=f"Before I can provide any medical advice, I need to gather some important health information. "
f"{next_question}",
finished=False
)
# Continue gathering information
if self.consultation_states[conversation_id] == ConsultationState.GATHERING_INFO:
self.gathered_info[conversation_id].append(message)
next_question = self._get_next_assessment_question(conversation_id)
if next_question:
return ChatResponse(
response=f"Thank you for that information. {next_question}",
finished=False
)
else:
self.consultation_states[conversation_id] = ConsultationState.DIAGNOSIS
context = "\n".join([
f"Q: {q}\nA: {a}" for q, a in
zip(HEALTH_ASSESSMENT_QUESTIONS, self.gathered_info[conversation_id])
])
messages = [
{"role": "system", "content": NURSE_OGE_IDENTITY},
{"role": "user", "content": f"Based on the following patient information, provide a thorough assessment and recommendations:\n\n{context}\n\nOriginal query: {message}"}
]
response = self.llm.create_chat_completion(
messages=messages,
max_tokens=1024,
temperature=0.7
)
self.consultation_states[conversation_id] = ConsultationState.INITIAL
self.gathered_info[conversation_id] = []
return ChatResponse(
response=response['choices'][0]['message']['content'],
finished=True
)
# Initialize FastAPI
app = FastAPI()
# Create a global variable for our assistant
nurse_oge = None
@app.on_event("startup")
async def startup_event():
global nurse_oge
try:
nurse_oge = NurseOgeAssistant()
except Exception as e:
print(f"Failed to initialize NurseOgeAssistant: {e}")
# We'll continue running but the /chat endpoint will return errors
@app.post("/chat")
async def chat_endpoint(request: ChatRequest):
if nurse_oge is None:
raise HTTPException(
status_code=503,
detail="The medical assistant is not available at the moment. Please try again later."
)
conversation_id = "default"
if not request.messages:
raise HTTPException(status_code=400, detail="No messages provided")
latest_message = request.messages[-1].content
response = await nurse_oge.process_message(
conversation_id=conversation_id,
message=latest_message,
history=request.messages[:-1]
)
return response
# Gradio interface
def gradio_chat(message, history):
if nurse_oge is None:
return "The medical assistant is not available at the moment. Please try again later."
response = nurse_oge.process_message("gradio_user", message, history)
return response.response
demo = gr.ChatInterface(
fn=gradio_chat,
title="Nurse Oge",
description="Finetuned llama 3.0 for medical diagnosis and all. This is just a demo",
theme="soft"
)
# Mount both FastAPI and Gradio
app = gr.mount_gradio_app(app, demo, path="/gradio")
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000) |