Update app.py
Browse files
app.py
CHANGED
@@ -37,7 +37,7 @@ MAX_NEW_TOKENS = 4096
|
|
37 |
MAX_CHUNK_TOKENS = 8192
|
38 |
BATCH_SIZE = 2
|
39 |
PROMPT_OVERHEAD = 300
|
40 |
-
SAFE_SLEEP = 0.5
|
41 |
|
42 |
# === Utility Functions ===
|
43 |
def estimate_tokens(text: str) -> int:
|
@@ -48,6 +48,17 @@ def clean_response(text: str) -> str:
|
|
48 |
text = re.sub(r"\n{3,}", "\n\n", text)
|
49 |
return text.strip()
|
50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
def extract_text_from_excel(path: str) -> str:
|
52 |
all_text = []
|
53 |
xls = pd.ExcelFile(path)
|
@@ -79,6 +90,9 @@ def extract_text_from_csv(path: str) -> str:
|
|
79 |
return "\n".join(all_text)
|
80 |
|
81 |
def extract_text_from_pdf(path: str) -> str:
|
|
|
|
|
|
|
82 |
all_text = []
|
83 |
try:
|
84 |
with pdfplumber.open(path) as pdf:
|
@@ -138,7 +152,6 @@ def init_agent() -> TxAgent:
|
|
138 |
agent.init_model()
|
139 |
return agent
|
140 |
|
141 |
-
# === Main Processing ===
|
142 |
def analyze_batches(agent, batches: List[List[str]]) -> List[str]:
|
143 |
results = []
|
144 |
for batch in batches:
|
@@ -172,7 +185,23 @@ def analyze_batches(agent, batches: List[List[str]]) -> List[str]:
|
|
172 |
return results
|
173 |
|
174 |
def generate_final_summary(agent, combined: str) -> str:
|
175 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
176 |
final_response = ""
|
177 |
for r in agent.run_gradio_chat(
|
178 |
message=final_prompt,
|
@@ -191,7 +220,10 @@ def generate_final_summary(agent, combined: str) -> str:
|
|
191 |
final_response += m.content
|
192 |
elif hasattr(r, "content"):
|
193 |
final_response += r.content
|
194 |
-
|
|
|
|
|
|
|
195 |
|
196 |
def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Dict[str, str]], Union[str, None]]:
|
197 |
if not file or not hasattr(file, "name"):
|
@@ -231,38 +263,18 @@ def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Di
|
|
231 |
|
232 |
def create_ui(agent):
|
233 |
with gr.Blocks(css="""
|
234 |
-
html, body, .gradio-container {
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
}
|
239 |
-
button.svelte-1ipelgc {
|
240 |
-
background: linear-gradient(to right, #1e88e5, #0d47a1) !important;
|
241 |
-
border: 1px solid #0d47a1 !important;
|
242 |
-
color: white !important;
|
243 |
-
font-weight: bold !important;
|
244 |
-
padding: 10px 20px !important;
|
245 |
-
border-radius: 8px !important;
|
246 |
-
}
|
247 |
-
button.svelte-1ipelgc:hover {
|
248 |
-
background: linear-gradient(to right, #2196f3, #1565c0) !important;
|
249 |
-
border: 1px solid #1565c0 !important;
|
250 |
-
}
|
251 |
-
.gr-column {
|
252 |
-
align-items: center !important;
|
253 |
-
gap: 12px;
|
254 |
-
}
|
255 |
-
.gr-file, .gr-button {
|
256 |
-
width: 100% !important;
|
257 |
-
max-width: 400px;
|
258 |
-
}
|
259 |
""") as demo:
|
260 |
gr.Markdown("""
|
261 |
<h2 style="text-align:center;">π CPS: Clinical Patient Support System</h2>
|
262 |
<p style="text-align:center;">Analyze and summarize unstructured medical files using AI (optimized for A100 GPU).</p>
|
263 |
""")
|
264 |
with gr.Column():
|
265 |
-
chatbot = gr.Chatbot(label="π§ CPS Assistant", height=480, type="messages")
|
266 |
upload = gr.File(label="π Upload Medical File", file_types=[".xlsx", ".csv", ".pdf"])
|
267 |
analyze = gr.Button("π§ Analyze")
|
268 |
download = gr.File(label="π₯ Download Report", visible=False, interactive=False)
|
@@ -281,4 +293,4 @@ def create_ui(agent):
|
|
281 |
if __name__ == "__main__":
|
282 |
agent = init_agent()
|
283 |
ui = create_ui(agent)
|
284 |
-
ui.launch(server_name="0.0.0.0", server_port=7860, allowed_paths=["/data/hf_cache/reports"], share=False)
|
|
|
37 |
MAX_CHUNK_TOKENS = 8192
|
38 |
BATCH_SIZE = 2
|
39 |
PROMPT_OVERHEAD = 300
|
40 |
+
SAFE_SLEEP = 0.5
|
41 |
|
42 |
# === Utility Functions ===
|
43 |
def estimate_tokens(text: str) -> int:
|
|
|
48 |
text = re.sub(r"\n{3,}", "\n\n", text)
|
49 |
return text.strip()
|
50 |
|
51 |
+
def remove_duplicate_paragraphs(text: str) -> str:
|
52 |
+
paragraphs = text.strip().split("\n\n")
|
53 |
+
seen = set()
|
54 |
+
unique_paragraphs = []
|
55 |
+
for p in paragraphs:
|
56 |
+
clean_p = p.strip()
|
57 |
+
if clean_p and clean_p not in seen:
|
58 |
+
unique_paragraphs.append(clean_p)
|
59 |
+
seen.add(clean_p)
|
60 |
+
return "\n\n".join(unique_paragraphs)
|
61 |
+
|
62 |
def extract_text_from_excel(path: str) -> str:
|
63 |
all_text = []
|
64 |
xls = pd.ExcelFile(path)
|
|
|
90 |
return "\n".join(all_text)
|
91 |
|
92 |
def extract_text_from_pdf(path: str) -> str:
|
93 |
+
import logging
|
94 |
+
logging.getLogger("pdfminer").setLevel(logging.ERROR)
|
95 |
+
|
96 |
all_text = []
|
97 |
try:
|
98 |
with pdfplumber.open(path) as pdf:
|
|
|
152 |
agent.init_model()
|
153 |
return agent
|
154 |
|
|
|
155 |
def analyze_batches(agent, batches: List[List[str]]) -> List[str]:
|
156 |
results = []
|
157 |
for batch in batches:
|
|
|
185 |
return results
|
186 |
|
187 |
def generate_final_summary(agent, combined: str) -> str:
|
188 |
+
combined = remove_duplicate_paragraphs(combined)
|
189 |
+
final_prompt = f"""
|
190 |
+
You are an expert clinical summarizer. Analyze the following summaries carefully and generate a **single final concise structured medical report**, avoiding any repetition or redundancy.
|
191 |
+
|
192 |
+
Summaries:
|
193 |
+
{combined}
|
194 |
+
|
195 |
+
Respond with:
|
196 |
+
- Diagnostic Patterns
|
197 |
+
- Medication Issues
|
198 |
+
- Missed Opportunities
|
199 |
+
- Inconsistencies
|
200 |
+
- Follow-up Recommendations
|
201 |
+
|
202 |
+
Avoid repeating the same points multiple times.
|
203 |
+
""".strip()
|
204 |
+
|
205 |
final_response = ""
|
206 |
for r in agent.run_gradio_chat(
|
207 |
message=final_prompt,
|
|
|
220 |
final_response += m.content
|
221 |
elif hasattr(r, "content"):
|
222 |
final_response += r.content
|
223 |
+
|
224 |
+
final_response = clean_response(final_response)
|
225 |
+
final_response = remove_duplicate_paragraphs(final_response)
|
226 |
+
return final_response
|
227 |
|
228 |
def process_report(agent, file, messages: List[Dict[str, str]]) -> Tuple[List[Dict[str, str]], Union[str, None]]:
|
229 |
if not file or not hasattr(file, "name"):
|
|
|
263 |
|
264 |
def create_ui(agent):
|
265 |
with gr.Blocks(css="""
|
266 |
+
html, body, .gradio-container { background: #0e1621; color: #e0e0e0; padding: 16px; }
|
267 |
+
button.svelte-1ipelgc { background: linear-gradient(to right, #1e88e5, #0d47a1) !important; border: 1px solid #0d47a1 !important; color: white !important; font-weight: bold !important; padding: 10px 20px !important; border-radius: 8px !important; }
|
268 |
+
button.svelte-1ipelgc:hover { background: linear-gradient(to right, #2196f3, #1565c0) !important; border: 1px solid #1565c0 !important; }
|
269 |
+
.gr-column { align-items: center !important; gap: 12px; }
|
270 |
+
.gr-file, .gr-button { width: 100% !important; max-width: 400px; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
271 |
""") as demo:
|
272 |
gr.Markdown("""
|
273 |
<h2 style="text-align:center;">π CPS: Clinical Patient Support System</h2>
|
274 |
<p style="text-align:center;">Analyze and summarize unstructured medical files using AI (optimized for A100 GPU).</p>
|
275 |
""")
|
276 |
with gr.Column():
|
277 |
+
chatbot = gr.Chatbot(label="π§ CPS Assistant", height=480, type="messages")
|
278 |
upload = gr.File(label="π Upload Medical File", file_types=[".xlsx", ".csv", ".pdf"])
|
279 |
analyze = gr.Button("π§ Analyze")
|
280 |
download = gr.File(label="π₯ Download Report", visible=False, interactive=False)
|
|
|
293 |
if __name__ == "__main__":
|
294 |
agent = init_agent()
|
295 |
ui = create_ui(agent)
|
296 |
+
ui.launch(server_name="0.0.0.0", server_port=7860, allowed_paths=["/data/hf_cache/reports"], share=False)
|