Update app.py
Browse files
app.py
CHANGED
@@ -3,42 +3,38 @@ import sys
|
|
3 |
import json
|
4 |
import re
|
5 |
import logging
|
|
|
|
|
6 |
from fastapi import FastAPI, HTTPException, UploadFile, File
|
7 |
from fastapi.responses import JSONResponse
|
8 |
from fastapi.middleware.cors import CORSMiddleware
|
9 |
-
from typing import List, Dict, Optional
|
10 |
-
from datetime import datetime
|
11 |
from pydantic import BaseModel
|
12 |
import markdown
|
13 |
import PyPDF2
|
14 |
|
15 |
-
#
|
16 |
logging.basicConfig(
|
17 |
level=logging.INFO,
|
18 |
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
19 |
)
|
20 |
logger = logging.getLogger("TxAgentAPI")
|
21 |
|
22 |
-
#
|
23 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
24 |
src_path = os.path.abspath(os.path.join(current_dir, "src"))
|
25 |
sys.path.insert(0, src_path)
|
26 |
|
27 |
-
# Import TxAgent
|
28 |
try:
|
29 |
from txagent.txagent import TxAgent
|
30 |
except ImportError as e:
|
31 |
logger.error(f"Failed to import TxAgent: {str(e)}")
|
32 |
raise
|
33 |
|
34 |
-
#
|
35 |
-
app = FastAPI(
|
36 |
-
title="TxAgent API",
|
37 |
-
description="API for TxAgent medical document analysis",
|
38 |
-
version="2.0.0"
|
39 |
-
)
|
40 |
|
41 |
-
# CORS
|
42 |
app.add_middleware(
|
43 |
CORSMiddleware,
|
44 |
allow_origins=["*"],
|
@@ -47,50 +43,48 @@ app.add_middleware(
|
|
47 |
allow_headers=["*"],
|
48 |
)
|
49 |
|
50 |
-
# Request
|
51 |
class ChatRequest(BaseModel):
|
52 |
message: str
|
53 |
temperature: float = 0.7
|
54 |
max_new_tokens: int = 512
|
55 |
history: Optional[List[Dict]] = None
|
56 |
-
format: Optional[str] = "clean"
|
57 |
|
58 |
-
# Response
|
59 |
def clean_text_response(text: str) -> str:
|
60 |
-
"""Basic text cleaning"""
|
61 |
text = re.sub(r'\n\s*\n', '\n\n', text)
|
62 |
text = re.sub(r'[ ]+', ' ', text)
|
63 |
text = text.replace("**", "").replace("__", "")
|
64 |
return text.strip()
|
65 |
|
66 |
def structure_medical_response(text: str) -> Dict:
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
|
|
83 |
agent = None
|
84 |
|
85 |
@app.on_event("startup")
|
86 |
async def startup_event():
|
87 |
global agent
|
88 |
try:
|
89 |
-
logger.info("Initializing TxAgent...")
|
90 |
agent = TxAgent(
|
91 |
model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B",
|
92 |
rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B",
|
93 |
-
tool_files_dict={},
|
94 |
enable_finish=True,
|
95 |
enable_rag=False,
|
96 |
force_finish=True,
|
@@ -98,17 +92,19 @@ async def startup_event():
|
|
98 |
step_rag_num=4,
|
99 |
seed=100
|
100 |
)
|
|
|
|
|
|
|
|
|
|
|
101 |
agent.init_model()
|
102 |
logger.info("TxAgent initialized successfully")
|
103 |
except Exception as e:
|
104 |
-
logger.error(f"
|
105 |
-
raise RuntimeError(f"Failed to initialize agent: {str(e)}")
|
106 |
|
107 |
@app.post("/chat")
|
108 |
async def chat_endpoint(request: ChatRequest):
|
109 |
-
"""Handle chat conversations with formatting options"""
|
110 |
try:
|
111 |
-
logger.info(f"Chat request received (format: {request.format})")
|
112 |
raw_response = agent.chat(
|
113 |
message=request.message,
|
114 |
history=request.history,
|
@@ -121,11 +117,10 @@ async def chat_endpoint(request: ChatRequest):
|
|
121 |
"structured": structure_medical_response(raw_response),
|
122 |
"html": markdown.markdown(raw_response)
|
123 |
}
|
124 |
-
response_content = formatted_response.get(request.format, formatted_response["clean"])
|
125 |
return JSONResponse({
|
126 |
"status": "success",
|
127 |
"format": request.format,
|
128 |
-
"response":
|
129 |
"timestamp": datetime.now().isoformat(),
|
130 |
"available_formats": list(formatted_response.keys())
|
131 |
})
|
@@ -135,36 +130,43 @@ async def chat_endpoint(request: ChatRequest):
|
|
135 |
|
136 |
@app.post("/upload")
|
137 |
async def upload_file(file: UploadFile = File(...)):
|
138 |
-
"""Handle file uploads and process with TxAgent"""
|
139 |
try:
|
140 |
logger.info(f"File upload received: {file.filename}")
|
141 |
content = ""
|
142 |
-
if file.filename.endswith(
|
143 |
pdf_reader = PyPDF2.PdfReader(file.file)
|
144 |
for page in pdf_reader.pages:
|
145 |
content += page.extract_text() or ""
|
146 |
else:
|
147 |
content = await file.read()
|
148 |
-
content = content.decode(
|
149 |
-
|
150 |
-
message = f"
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
157 |
formatted_response = {
|
158 |
"raw": raw_response,
|
159 |
"clean": clean_text_response(raw_response),
|
160 |
"structured": structure_medical_response(raw_response),
|
161 |
"html": markdown.markdown(raw_response)
|
162 |
}
|
163 |
-
response_content = formatted_response["clean"]
|
164 |
return JSONResponse({
|
165 |
"status": "success",
|
166 |
"format": "clean",
|
167 |
-
"response":
|
168 |
"timestamp": datetime.now().isoformat(),
|
169 |
"available_formats": list(formatted_response.keys())
|
170 |
})
|
@@ -175,12 +177,10 @@ async def upload_file(file: UploadFile = File(...)):
|
|
175 |
file.file.close()
|
176 |
|
177 |
@app.get("/status")
|
178 |
-
async def
|
179 |
-
"""Check service status"""
|
180 |
return {
|
181 |
"status": "running",
|
182 |
-
"version": "2.
|
183 |
"model": agent.model_name if agent else "not loaded",
|
184 |
-
"formats_available": ["raw", "clean", "structured", "html"],
|
185 |
"timestamp": datetime.now().isoformat()
|
186 |
-
}
|
|
|
3 |
import json
|
4 |
import re
|
5 |
import logging
|
6 |
+
from datetime import datetime
|
7 |
+
from typing import List, Dict, Optional
|
8 |
from fastapi import FastAPI, HTTPException, UploadFile, File
|
9 |
from fastapi.responses import JSONResponse
|
10 |
from fastapi.middleware.cors import CORSMiddleware
|
|
|
|
|
11 |
from pydantic import BaseModel
|
12 |
import markdown
|
13 |
import PyPDF2
|
14 |
|
15 |
+
# Setup logging
|
16 |
logging.basicConfig(
|
17 |
level=logging.INFO,
|
18 |
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
19 |
)
|
20 |
logger = logging.getLogger("TxAgentAPI")
|
21 |
|
22 |
+
# Adjust sys path
|
23 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
24 |
src_path = os.path.abspath(os.path.join(current_dir, "src"))
|
25 |
sys.path.insert(0, src_path)
|
26 |
|
27 |
+
# Import TxAgent
|
28 |
try:
|
29 |
from txagent.txagent import TxAgent
|
30 |
except ImportError as e:
|
31 |
logger.error(f"Failed to import TxAgent: {str(e)}")
|
32 |
raise
|
33 |
|
34 |
+
# Init FastAPI
|
35 |
+
app = FastAPI(title="TxAgent API", version="2.1.0")
|
|
|
|
|
|
|
|
|
36 |
|
37 |
+
# CORS
|
38 |
app.add_middleware(
|
39 |
CORSMiddleware,
|
40 |
allow_origins=["*"],
|
|
|
43 |
allow_headers=["*"],
|
44 |
)
|
45 |
|
46 |
+
# Request schema
|
47 |
class ChatRequest(BaseModel):
|
48 |
message: str
|
49 |
temperature: float = 0.7
|
50 |
max_new_tokens: int = 512
|
51 |
history: Optional[List[Dict]] = None
|
52 |
+
format: Optional[str] = "clean"
|
53 |
|
54 |
+
# Response formatting
|
55 |
def clean_text_response(text: str) -> str:
|
|
|
56 |
text = re.sub(r'\n\s*\n', '\n\n', text)
|
57 |
text = re.sub(r'[ ]+', ' ', text)
|
58 |
text = text.replace("**", "").replace("__", "")
|
59 |
return text.strip()
|
60 |
|
61 |
def structure_medical_response(text: str) -> Dict:
|
62 |
+
return {
|
63 |
+
"summary": extract_section(text, "Summary"),
|
64 |
+
"risks": extract_section(text, "Risks or Red Flags"),
|
65 |
+
"missed_issues": extract_section(text, "What the doctor might have missed"),
|
66 |
+
"recommendations": extract_section(text, "Suggested Clinical Actions")
|
67 |
+
}
|
68 |
+
|
69 |
+
def extract_section(text: str, heading: str) -> str:
|
70 |
+
try:
|
71 |
+
pattern = rf"{heading}:\n(.*?)(?=\n\w|\Z)"
|
72 |
+
match = re.search(pattern, text, re.DOTALL)
|
73 |
+
return clean_text_response(match.group(1)) if match else ""
|
74 |
+
except Exception as e:
|
75 |
+
logger.error(f"Section extraction failed: {e}")
|
76 |
+
return ""
|
77 |
+
|
78 |
+
# Agent init
|
79 |
agent = None
|
80 |
|
81 |
@app.on_event("startup")
|
82 |
async def startup_event():
|
83 |
global agent
|
84 |
try:
|
|
|
85 |
agent = TxAgent(
|
86 |
model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B",
|
87 |
rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B",
|
|
|
88 |
enable_finish=True,
|
89 |
enable_rag=False,
|
90 |
force_finish=True,
|
|
|
92 |
step_rag_num=4,
|
93 |
seed=100
|
94 |
)
|
95 |
+
agent.chat_prompt = (
|
96 |
+
"You are a clinical decision support assistant for doctors. "
|
97 |
+
"You analyze patient documents, detect medical issues, identify missed diagnoses, "
|
98 |
+
"and provide treatment suggestions with rationale in concise, readable language."
|
99 |
+
)
|
100 |
agent.init_model()
|
101 |
logger.info("TxAgent initialized successfully")
|
102 |
except Exception as e:
|
103 |
+
logger.error(f"Startup error: {str(e)}")
|
|
|
104 |
|
105 |
@app.post("/chat")
|
106 |
async def chat_endpoint(request: ChatRequest):
|
|
|
107 |
try:
|
|
|
108 |
raw_response = agent.chat(
|
109 |
message=request.message,
|
110 |
history=request.history,
|
|
|
117 |
"structured": structure_medical_response(raw_response),
|
118 |
"html": markdown.markdown(raw_response)
|
119 |
}
|
|
|
120 |
return JSONResponse({
|
121 |
"status": "success",
|
122 |
"format": request.format,
|
123 |
+
"response": formatted_response.get(request.format, formatted_response["clean"]),
|
124 |
"timestamp": datetime.now().isoformat(),
|
125 |
"available_formats": list(formatted_response.keys())
|
126 |
})
|
|
|
130 |
|
131 |
@app.post("/upload")
|
132 |
async def upload_file(file: UploadFile = File(...)):
|
|
|
133 |
try:
|
134 |
logger.info(f"File upload received: {file.filename}")
|
135 |
content = ""
|
136 |
+
if file.filename.endswith(".pdf"):
|
137 |
pdf_reader = PyPDF2.PdfReader(file.file)
|
138 |
for page in pdf_reader.pages:
|
139 |
content += page.extract_text() or ""
|
140 |
else:
|
141 |
content = await file.read()
|
142 |
+
content = content.decode("utf-8", errors="ignore")
|
143 |
+
|
144 |
+
message = f"""
|
145 |
+
You are a clinical decision support AI assisting physicians.
|
146 |
+
|
147 |
+
Given the following patient report, do the following:
|
148 |
+
1. Summarize the patient's main conditions and history.
|
149 |
+
2. Identify any potential clinical risks or red flags.
|
150 |
+
3. Highlight any important diagnoses or treatments the doctor might have missed.
|
151 |
+
4. Suggest next clinical steps, treatments, or referrals (if applicable).
|
152 |
+
5. Flag anything that could pose an urgent risk (e.g., suicide risk, untreated critical conditions).
|
153 |
+
|
154 |
+
Patient Document:
|
155 |
+
-----------------
|
156 |
+
{content[:10000]}
|
157 |
+
"""
|
158 |
+
|
159 |
+
raw_response = agent.chat(message=message, history=[], temperature=0.7, max_new_tokens=1024)
|
160 |
formatted_response = {
|
161 |
"raw": raw_response,
|
162 |
"clean": clean_text_response(raw_response),
|
163 |
"structured": structure_medical_response(raw_response),
|
164 |
"html": markdown.markdown(raw_response)
|
165 |
}
|
|
|
166 |
return JSONResponse({
|
167 |
"status": "success",
|
168 |
"format": "clean",
|
169 |
+
"response": formatted_response["clean"],
|
170 |
"timestamp": datetime.now().isoformat(),
|
171 |
"available_formats": list(formatted_response.keys())
|
172 |
})
|
|
|
177 |
file.file.close()
|
178 |
|
179 |
@app.get("/status")
|
180 |
+
async def status():
|
|
|
181 |
return {
|
182 |
"status": "running",
|
183 |
+
"version": "2.1.0",
|
184 |
"model": agent.model_name if agent else "not loaded",
|
|
|
185 |
"timestamp": datetime.now().isoformat()
|
186 |
+
}
|