Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -23,20 +23,43 @@ if 'summaries' not in st.session_state:
|
|
23 |
if 'text_processor' not in st.session_state:
|
24 |
st.session_state.text_processor = None
|
25 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
def load_model(model_type):
|
27 |
-
"""Load appropriate model based on type"""
|
|
|
|
|
28 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
if model_type == "summarize":
|
30 |
base_model = AutoModelForSeq2SeqLM.from_pretrained(
|
31 |
"facebook/bart-large-cnn",
|
32 |
-
cache_dir="./models"
|
|
|
|
|
33 |
)
|
34 |
model = PeftModel.from_pretrained(
|
35 |
base_model,
|
36 |
"pendar02/results",
|
37 |
-
|
38 |
-
|
39 |
-
torch_dtype=torch.float32
|
40 |
)
|
41 |
tokenizer = AutoTokenizer.from_pretrained(
|
42 |
"facebook/bart-large-cnn",
|
@@ -45,14 +68,15 @@ def load_model(model_type):
|
|
45 |
else: # question_focused
|
46 |
base_model = AutoModelForSeq2SeqLM.from_pretrained(
|
47 |
"GanjinZero/biobart-base",
|
48 |
-
cache_dir="./models"
|
|
|
|
|
49 |
)
|
50 |
model = PeftModel.from_pretrained(
|
51 |
base_model,
|
52 |
"pendar02/biobart-finetune",
|
53 |
-
|
54 |
-
|
55 |
-
torch_dtype=torch.float32
|
56 |
)
|
57 |
tokenizer = AutoTokenizer.from_pretrained(
|
58 |
"GanjinZero/biobart-base",
|
@@ -148,23 +172,10 @@ def generate_focused_summary(question, abstracts, model, tokenizer):
|
|
148 |
|
149 |
return tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
150 |
|
|
|
151 |
def main():
|
152 |
st.title("π¬ Biomedical Papers Analysis")
|
153 |
|
154 |
-
# Sidebar
|
155 |
-
st.sidebar.header("About")
|
156 |
-
st.sidebar.info(
|
157 |
-
"This app analyzes biomedical research papers. Upload an Excel file "
|
158 |
-
"containing paper details and abstracts to:"
|
159 |
-
"\n- Generate individual summaries"
|
160 |
-
"\n- Get question-focused insights"
|
161 |
-
)
|
162 |
-
|
163 |
-
# Initialize text processor if not already done
|
164 |
-
if st.session_state.text_processor is None:
|
165 |
-
with st.spinner("Loading NLP models..."):
|
166 |
-
st.session_state.text_processor = TextProcessor()
|
167 |
-
|
168 |
# File upload section
|
169 |
uploaded_file = st.file_uploader(
|
170 |
"Upload Excel file containing papers",
|
@@ -179,74 +190,66 @@ def main():
|
|
179 |
df = process_excel(uploaded_file)
|
180 |
if df is not None:
|
181 |
st.session_state.processed_data = df.dropna(subset=["Abstract"])
|
|
|
|
|
|
|
|
|
182 |
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
|
|
|
|
194 |
model, tokenizer = load_model("summarize")
|
195 |
|
196 |
-
|
197 |
progress_bar = st.progress(0)
|
198 |
-
|
199 |
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
progress_bar.progress((i + 1) / len(df))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
204 |
|
205 |
st.session_state.summaries = summaries
|
206 |
|
207 |
-
# Clear
|
208 |
del model
|
209 |
del tokenizer
|
210 |
torch.cuda.empty_cache()
|
211 |
gc.collect()
|
212 |
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
sort_column = st.selectbox("Sort by:", df.columns)
|
221 |
-
with col2:
|
222 |
-
ascending = st.checkbox("Ascending order", True)
|
223 |
-
|
224 |
-
# Create display dataframe
|
225 |
-
display_df = df.copy()
|
226 |
-
display_df['Summary'] = st.session_state.summaries
|
227 |
-
sorted_df = display_df.sort_values(by=sort_column, ascending=ascending)
|
228 |
-
|
229 |
-
# Show interactive table
|
230 |
-
st.dataframe(sorted_df, hide_index=True)
|
231 |
-
|
232 |
-
# Question-focused Summary Section
|
233 |
-
st.header("β Question-focused Summary")
|
234 |
-
question = st.text_input("Enter your research question:")
|
235 |
-
|
236 |
-
if question and st.button("Generate Focused Summary"):
|
237 |
-
try:
|
238 |
-
with st.spinner("Analyzing relevant papers..."):
|
239 |
-
# Find relevant abstracts
|
240 |
results = st.session_state.text_processor.find_most_relevant_abstracts(
|
241 |
question,
|
242 |
df['Abstract'].tolist(),
|
243 |
top_k=5
|
244 |
)
|
245 |
|
246 |
-
# Load question-focused model
|
247 |
model, tokenizer = load_model("question_focused")
|
248 |
|
249 |
-
# Get relevant abstracts and generate summary
|
250 |
relevant_abstracts = df['Abstract'].iloc[results['top_indices']].tolist()
|
251 |
focused_summary = generate_focused_summary(
|
252 |
question,
|
@@ -255,26 +258,68 @@ def main():
|
|
255 |
tokenizer
|
256 |
)
|
257 |
|
258 |
-
|
259 |
-
st.subheader("Summary")
|
260 |
st.write(focused_summary)
|
261 |
|
262 |
-
# Show relevant papers
|
263 |
st.subheader("Most Relevant Papers")
|
264 |
relevant_papers = df.iloc[results['top_indices']][
|
265 |
['Article Title', 'Authors', 'Publication Year', 'DOI']
|
266 |
]
|
267 |
relevant_papers['Relevance Score'] = results['scores']
|
268 |
-
|
269 |
|
270 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
271 |
del model
|
272 |
del tokenizer
|
273 |
torch.cuda.empty_cache()
|
274 |
gc.collect()
|
275 |
-
|
276 |
-
|
277 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
278 |
|
279 |
if __name__ == "__main__":
|
280 |
main()
|
|
|
23 |
if 'text_processor' not in st.session_state:
|
24 |
st.session_state.text_processor = None
|
25 |
|
26 |
+
def manage_resources():
|
27 |
+
"""Clear memory and ensure resources are available"""
|
28 |
+
# Force garbage collection
|
29 |
+
gc.collect()
|
30 |
+
|
31 |
+
# Clear CUDA cache if available
|
32 |
+
if torch.cuda.is_available():
|
33 |
+
torch.cuda.empty_cache()
|
34 |
+
|
35 |
+
# Set torch to use CPU
|
36 |
+
torch.set_num_threads(8) # Use half of available CPU threads for each model
|
37 |
+
|
38 |
def load_model(model_type):
|
39 |
+
"""Load appropriate model based on type with resource management"""
|
40 |
+
manage_resources()
|
41 |
+
|
42 |
try:
|
43 |
+
# Set lower precision to reduce memory usage
|
44 |
+
torch_dtype = torch.float32
|
45 |
+
if torch.cuda.is_available():
|
46 |
+
device = "cuda"
|
47 |
+
else:
|
48 |
+
device = "cpu"
|
49 |
+
torch_dtype = torch.float32 # Use float32 for CPU
|
50 |
+
|
51 |
if model_type == "summarize":
|
52 |
base_model = AutoModelForSeq2SeqLM.from_pretrained(
|
53 |
"facebook/bart-large-cnn",
|
54 |
+
cache_dir="./models",
|
55 |
+
torch_dtype=torch_dtype,
|
56 |
+
low_cpu_mem_usage=True
|
57 |
)
|
58 |
model = PeftModel.from_pretrained(
|
59 |
base_model,
|
60 |
"pendar02/results",
|
61 |
+
device_map=device,
|
62 |
+
torch_dtype=torch_dtype
|
|
|
63 |
)
|
64 |
tokenizer = AutoTokenizer.from_pretrained(
|
65 |
"facebook/bart-large-cnn",
|
|
|
68 |
else: # question_focused
|
69 |
base_model = AutoModelForSeq2SeqLM.from_pretrained(
|
70 |
"GanjinZero/biobart-base",
|
71 |
+
cache_dir="./models",
|
72 |
+
torch_dtype=torch_dtype,
|
73 |
+
low_cpu_mem_usage=True
|
74 |
)
|
75 |
model = PeftModel.from_pretrained(
|
76 |
base_model,
|
77 |
"pendar02/biobart-finetune",
|
78 |
+
device_map=device,
|
79 |
+
torch_dtype=torch_dtype
|
|
|
80 |
)
|
81 |
tokenizer = AutoTokenizer.from_pretrained(
|
82 |
"GanjinZero/biobart-base",
|
|
|
172 |
|
173 |
return tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
174 |
|
175 |
+
|
176 |
def main():
|
177 |
st.title("π¬ Biomedical Papers Analysis")
|
178 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
179 |
# File upload section
|
180 |
uploaded_file = st.file_uploader(
|
181 |
"Upload Excel file containing papers",
|
|
|
190 |
df = process_excel(uploaded_file)
|
191 |
if df is not None:
|
192 |
st.session_state.processed_data = df.dropna(subset=["Abstract"])
|
193 |
+
|
194 |
+
if st.session_state.processed_data is not None:
|
195 |
+
df = st.session_state.processed_data
|
196 |
+
st.write(f"π Loaded {len(df)} papers")
|
197 |
|
198 |
+
# Individual Summaries Section
|
199 |
+
st.header("π Individual Paper Summaries")
|
200 |
+
|
201 |
+
# Question input before the unified generate button
|
202 |
+
st.header("β Question-focused Summary (Optional)")
|
203 |
+
question = st.text_input("Enter your research question (optional):")
|
204 |
+
|
205 |
+
# Unified generate button
|
206 |
+
if st.button("Generate Analysis"):
|
207 |
+
try:
|
208 |
+
# Step 1: Generate Individual Summaries
|
209 |
+
if st.session_state.summaries is None:
|
210 |
+
with st.spinner("Generating individual summaries..."):
|
211 |
model, tokenizer = load_model("summarize")
|
212 |
|
213 |
+
progress_text = st.empty()
|
214 |
progress_bar = st.progress(0)
|
215 |
+
summary_display = st.container()
|
216 |
|
217 |
+
summaries = []
|
218 |
+
for i, (_, row) in enumerate(df.iterrows()):
|
219 |
+
progress_text.text(f"Processing paper {i+1} of {len(df)}")
|
220 |
progress_bar.progress((i + 1) / len(df))
|
221 |
+
|
222 |
+
summary = generate_summary(row['Abstract'], model, tokenizer)
|
223 |
+
summaries.append(summary)
|
224 |
+
|
225 |
+
with summary_display:
|
226 |
+
st.write(f"**Paper {i+1}:** {row['Article Title']}")
|
227 |
+
st.write(summary)
|
228 |
+
st.divider()
|
229 |
|
230 |
st.session_state.summaries = summaries
|
231 |
|
232 |
+
# Clear memory after individual summaries
|
233 |
del model
|
234 |
del tokenizer
|
235 |
torch.cuda.empty_cache()
|
236 |
gc.collect()
|
237 |
|
238 |
+
# Step 2: Generate Question-Focused Summary (only if question is provided)
|
239 |
+
if question.strip():
|
240 |
+
with st.spinner("Generating question-focused summary..."):
|
241 |
+
# Clear memory before question processing
|
242 |
+
torch.cuda.empty_cache()
|
243 |
+
gc.collect()
|
244 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
245 |
results = st.session_state.text_processor.find_most_relevant_abstracts(
|
246 |
question,
|
247 |
df['Abstract'].tolist(),
|
248 |
top_k=5
|
249 |
)
|
250 |
|
|
|
251 |
model, tokenizer = load_model("question_focused")
|
252 |
|
|
|
253 |
relevant_abstracts = df['Abstract'].iloc[results['top_indices']].tolist()
|
254 |
focused_summary = generate_focused_summary(
|
255 |
question,
|
|
|
258 |
tokenizer
|
259 |
)
|
260 |
|
261 |
+
st.subheader("Question-Focused Summary")
|
|
|
262 |
st.write(focused_summary)
|
263 |
|
|
|
264 |
st.subheader("Most Relevant Papers")
|
265 |
relevant_papers = df.iloc[results['top_indices']][
|
266 |
['Article Title', 'Authors', 'Publication Year', 'DOI']
|
267 |
]
|
268 |
relevant_papers['Relevance Score'] = results['scores']
|
269 |
+
relevant_papers['Publication Year'] = relevant_papers['Publication Year'].astype(int)
|
270 |
|
271 |
+
st.dataframe(
|
272 |
+
relevant_papers,
|
273 |
+
column_config={
|
274 |
+
'Publication Year': st.column_config.NumberColumn('Year', format="%d"),
|
275 |
+
'Relevance Score': st.column_config.NumberColumn('Relevance', format="%.3f")
|
276 |
+
},
|
277 |
+
hide_index=True
|
278 |
+
)
|
279 |
+
|
280 |
+
# Clear memory after question processing
|
281 |
del model
|
282 |
del tokenizer
|
283 |
torch.cuda.empty_cache()
|
284 |
gc.collect()
|
285 |
+
|
286 |
+
except Exception as e:
|
287 |
+
st.error(f"Error in analysis: {str(e)}")
|
288 |
+
|
289 |
+
# Display sorted summaries if they exist
|
290 |
+
if st.session_state.summaries is not None:
|
291 |
+
st.subheader("All Paper Summaries")
|
292 |
+
sort_options = {
|
293 |
+
'Article Title': 'Article Title',
|
294 |
+
'Authors': 'Authors',
|
295 |
+
'Publication Year': 'Publication Year',
|
296 |
+
'Source Title': 'Source Title'
|
297 |
+
}
|
298 |
+
|
299 |
+
col1, col2 = st.columns(2)
|
300 |
+
with col1:
|
301 |
+
sort_column = st.selectbox("Sort by:", list(sort_options.keys()))
|
302 |
+
with col2:
|
303 |
+
ascending = st.checkbox("Ascending order", True)
|
304 |
+
|
305 |
+
display_df = df.copy()
|
306 |
+
display_df['Summary'] = st.session_state.summaries
|
307 |
+
display_df['Publication Year'] = display_df['Publication Year'].astype(int)
|
308 |
+
sorted_df = display_df.sort_values(by=sort_options[sort_column], ascending=ascending)
|
309 |
+
|
310 |
+
st.dataframe(
|
311 |
+
sorted_df[['Article Title', 'Authors', 'Source Title',
|
312 |
+
'Publication Year', 'DOI', 'Summary']],
|
313 |
+
column_config={
|
314 |
+
'Article Title': st.column_config.TextColumn('Article Title', width='medium'),
|
315 |
+
'Authors': st.column_config.TextColumn('Authors', width='medium'),
|
316 |
+
'Source Title': st.column_config.TextColumn('Source Title', width='medium'),
|
317 |
+
'Publication Year': st.column_config.NumberColumn('Year', format="%d"),
|
318 |
+
'DOI': st.column_config.TextColumn('DOI', width='small'),
|
319 |
+
'Summary': st.column_config.TextColumn('Summary', width='large'),
|
320 |
+
},
|
321 |
+
hide_index=True
|
322 |
+
)
|
323 |
|
324 |
if __name__ == "__main__":
|
325 |
main()
|