Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -24,7 +24,7 @@ logger = logging.getLogger(__name__)
|
|
24 |
def load_embeddings_model():
|
25 |
logger.info("Loading embeddings model")
|
26 |
try:
|
27 |
-
return SentenceTransformer("all-MiniLM-
|
28 |
except Exception as e:
|
29 |
logger.error(f"Embeddings load error: {str(e)}")
|
30 |
st.error(f"Embedding model error: {str(e)}")
|
@@ -39,7 +39,7 @@ def load_qa_pipeline():
|
|
39 |
fine_tuned_pipeline = fine_tune_qa_model(dataset)
|
40 |
if fine_tuned_pipeline:
|
41 |
return fine_tuned_pipeline
|
42 |
-
return pipeline("text2text-generation", model="google/flan-t5-
|
43 |
except Exception as e:
|
44 |
logger.error(f"QA model load error: {str(e)}")
|
45 |
st.error(f"QA model error: {str(e)}")
|
@@ -49,7 +49,7 @@ def load_qa_pipeline():
|
|
49 |
def load_summary_pipeline():
|
50 |
logger.info("Loading summary pipeline")
|
51 |
try:
|
52 |
-
return pipeline("summarization", model="
|
53 |
except Exception as e:
|
54 |
logger.error(f"Summary model load error: {str(e)}")
|
55 |
st.error(f"Summary model error: {str(e)}")
|
@@ -79,7 +79,7 @@ def load_and_prepare_dataset(dataset_name="squad", max_samples=1000):
|
|
79 |
def fine_tune_qa_model(dataset):
|
80 |
logger.info("Starting fine-tuning")
|
81 |
try:
|
82 |
-
model_name = "google/flan-t5-
|
83 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
84 |
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
|
85 |
|
@@ -93,13 +93,13 @@ def fine_tune_qa_model(dataset):
|
|
93 |
|
94 |
training_args = TrainingArguments(
|
95 |
output_dir="./fine_tuned_model",
|
96 |
-
num_train_epochs=2,
|
97 |
per_device_train_batch_size=4,
|
98 |
save_steps=500,
|
99 |
logging_steps=100,
|
100 |
evaluation_strategy="no",
|
101 |
learning_rate=3e-5,
|
102 |
-
fp16=False,
|
103 |
)
|
104 |
|
105 |
trainer = Trainer(
|
@@ -118,14 +118,14 @@ def fine_tune_qa_model(dataset):
|
|
118 |
return None
|
119 |
|
120 |
# Augment vector store with dataset
|
121 |
-
def augment_vector_store(vector_store, dataset_name="squad", max_samples=
|
122 |
logger.info(f"Augmenting vector store with dataset: {dataset_name}")
|
123 |
try:
|
124 |
dataset = load_dataset(dataset_name, split="train").select(range(min(max_samples, len(dataset))))
|
125 |
chunks = [f"Context: {c}\nAnswer: {a['text'][0]}" for c, a in zip(dataset['context'], dataset['answers'])]
|
126 |
embeddings_model = load_embeddings_model()
|
127 |
if embeddings_model and vector_store:
|
128 |
-
embeddings = embeddings_model.encode(chunks, batch_size=
|
129 |
vector_store.add_embeddings(zip(chunks, embeddings))
|
130 |
return vector_store
|
131 |
except Exception as e:
|
@@ -139,16 +139,19 @@ def process_pdf(uploaded_file):
|
|
139 |
text = ""
|
140 |
code_blocks = []
|
141 |
with pdfplumber.open(BytesIO(uploaded_file.getvalue())) as pdf:
|
142 |
-
for page in pdf.pages[:
|
143 |
extracted = page.extract_text(layout=False)
|
144 |
-
if not extracted:
|
145 |
try:
|
146 |
img = page.to_image(resolution=150).original
|
147 |
extracted = pytesseract.image_to_string(img, config='--psm 6')
|
148 |
except Exception as ocr_e:
|
149 |
logger.warning(f"OCR failed: {str(ocr_e)}")
|
150 |
if extracted:
|
151 |
-
text
|
|
|
|
|
|
|
152 |
for char in page.chars:
|
153 |
if 'fontname' in char and 'mono' in char['fontname'].lower():
|
154 |
code_blocks.append(char['text'])
|
@@ -168,20 +171,20 @@ def process_pdf(uploaded_file):
|
|
168 |
if not text:
|
169 |
raise ValueError("No text extracted from PDF")
|
170 |
|
171 |
-
text_splitter = CharacterTextSplitter(separator="\n\n", chunk_size=
|
172 |
-
text_chunks = text_splitter.split_text(text)[:
|
173 |
-
code_chunks = text_splitter.split_text(code_text)[:
|
174 |
|
175 |
embeddings_model = load_embeddings_model()
|
176 |
if not embeddings_model:
|
177 |
return None, None, text, code_text
|
178 |
|
179 |
text_vector_store = FAISS.from_embeddings(
|
180 |
-
zip(text_chunks, [embeddings_model.encode(chunk, show_progress_bar=False) for chunk in text_chunks]),
|
181 |
embeddings_model.encode
|
182 |
) if text_chunks else None
|
183 |
code_vector_store = FAISS.from_embeddings(
|
184 |
-
zip(code_chunks, [embeddings_model.encode(chunk, show_progress_bar=False) for chunk in code_chunks]),
|
185 |
embeddings_model.encode
|
186 |
) if code_chunks else None
|
187 |
|
@@ -195,7 +198,7 @@ def process_pdf(uploaded_file):
|
|
195 |
st.error(f"PDF error: {str(e)}")
|
196 |
return None, None, "", ""
|
197 |
|
198 |
-
# Summarize PDF with ROUGE metrics
|
199 |
def summarize_pdf(text):
|
200 |
logger.info("Generating summary")
|
201 |
try:
|
@@ -203,23 +206,39 @@ def summarize_pdf(text):
|
|
203 |
if not summary_pipeline:
|
204 |
return "Summary model unavailable."
|
205 |
|
206 |
-
text_splitter = CharacterTextSplitter(separator="\n\n", chunk_size=
|
207 |
-
chunks = text_splitter.split_text(text)
|
208 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
|
210 |
-
|
211 |
-
|
|
|
212 |
summaries.append(summary.strip())
|
213 |
|
214 |
combined_summary = " ".join(summaries)
|
215 |
-
if len(combined_summary.split()) >
|
216 |
-
combined_summary = " ".join(combined_summary.split()[:
|
217 |
|
|
|
218 |
scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
|
219 |
-
scores = scorer.score(text[:
|
220 |
logger.info(f"ROUGE scores: {scores}")
|
221 |
|
222 |
-
return f"**Summary
|
223 |
except Exception as e:
|
224 |
logger.error(f"Summary error: {str(e)}")
|
225 |
return f"Oops, something went wrong summarizing: {str(e)}"
|
@@ -285,7 +304,7 @@ try:
|
|
285 |
""", unsafe_allow_html=True)
|
286 |
|
287 |
st.markdown('<div class="header"><h1>Smart PDF Q&A</h1></div>', unsafe_allow_html=True)
|
288 |
-
st.markdown("Upload a PDF to ask questions,
|
289 |
|
290 |
# Initialize session state
|
291 |
if "messages" not in st.session_state:
|
@@ -307,7 +326,7 @@ try:
|
|
307 |
if st.button("Fine-Tune Model"):
|
308 |
progress_bar = st.progress(0)
|
309 |
for i in range(100):
|
310 |
-
time.sleep(0.
|
311 |
progress_bar.progress(i + 1)
|
312 |
dataset = load_and_prepare_dataset(dataset_name=dataset_name)
|
313 |
if dataset:
|
@@ -319,6 +338,15 @@ try:
|
|
319 |
if st.button("Clear Chat"):
|
320 |
st.session_state.messages = []
|
321 |
st.experimental_rerun()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
322 |
st.markdown('</div>', unsafe_allow_html=True)
|
323 |
|
324 |
# PDF upload and processing
|
@@ -329,7 +357,7 @@ try:
|
|
329 |
progress_bar = st.progress(0)
|
330 |
with st.spinner("Processing PDF..."):
|
331 |
for i in range(100):
|
332 |
-
time.sleep(0.
|
333 |
progress_bar.progress(i + 1)
|
334 |
st.session_state.text_vector_store, st.session_state.code_vector_store, st.session_state.pdf_text, st.session_state.code_text = process_pdf(uploaded_file)
|
335 |
if st.session_state.text_vector_store or st.session_state.code_vector_store:
|
@@ -342,7 +370,7 @@ try:
|
|
342 |
progress_bar = st.progress(0)
|
343 |
with st.spinner("Summarizing..."):
|
344 |
for i in range(100):
|
345 |
-
time.sleep(0.
|
346 |
progress_bar.progress(i + 1)
|
347 |
summary = summarize_pdf(st.session_state.pdf_text)
|
348 |
st.session_state.messages.append({"role": "assistant", "content": summary})
|
@@ -360,7 +388,7 @@ try:
|
|
360 |
progress_bar = st.progress(0)
|
361 |
with st.spinner('<div class="spinner">⏳ Processing...</div>'):
|
362 |
for i in range(100):
|
363 |
-
time.sleep(0.
|
364 |
progress_bar.progress(i + 1)
|
365 |
answer = answer_question(st.session_state.text_vector_store, st.session_state.code_vector_store, prompt)
|
366 |
st.markdown(answer, unsafe_allow_html=True)
|
|
|
24 |
def load_embeddings_model():
|
25 |
logger.info("Loading embeddings model")
|
26 |
try:
|
27 |
+
return SentenceTransformer("all-MiniLM-L6-v2")
|
28 |
except Exception as e:
|
29 |
logger.error(f"Embeddings load error: {str(e)}")
|
30 |
st.error(f"Embedding model error: {str(e)}")
|
|
|
39 |
fine_tuned_pipeline = fine_tune_qa_model(dataset)
|
40 |
if fine_tuned_pipeline:
|
41 |
return fine_tuned_pipeline
|
42 |
+
return pipeline("text2text-generation", model="google/flan-t5-small", max_length=300)
|
43 |
except Exception as e:
|
44 |
logger.error(f"QA model load error: {str(e)}")
|
45 |
st.error(f"QA model error: {str(e)}")
|
|
|
49 |
def load_summary_pipeline():
|
50 |
logger.info("Loading summary pipeline")
|
51 |
try:
|
52 |
+
return pipeline("summarization", model="facebook/bart-large-cnn", max_length=250)
|
53 |
except Exception as e:
|
54 |
logger.error(f"Summary model load error: {str(e)}")
|
55 |
st.error(f"Summary model error: {str(e)}")
|
|
|
79 |
def fine_tune_qa_model(dataset):
|
80 |
logger.info("Starting fine-tuning")
|
81 |
try:
|
82 |
+
model_name = "google/flan-t5-small"
|
83 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
84 |
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
|
85 |
|
|
|
93 |
|
94 |
training_args = TrainingArguments(
|
95 |
output_dir="./fine_tuned_model",
|
96 |
+
num_train_epochs = 2,
|
97 |
per_device_train_batch_size=4,
|
98 |
save_steps=500,
|
99 |
logging_steps=100,
|
100 |
evaluation_strategy="no",
|
101 |
learning_rate=3e-5,
|
102 |
+
fp16=False,
|
103 |
)
|
104 |
|
105 |
trainer = Trainer(
|
|
|
118 |
return None
|
119 |
|
120 |
# Augment vector store with dataset
|
121 |
+
def augment_vector_store(vector_store, dataset_name="squad", max_samples=300):
|
122 |
logger.info(f"Augmenting vector store with dataset: {dataset_name}")
|
123 |
try:
|
124 |
dataset = load_dataset(dataset_name, split="train").select(range(min(max_samples, len(dataset))))
|
125 |
chunks = [f"Context: {c}\nAnswer: {a['text'][0]}" for c, a in zip(dataset['context'], dataset['answers'])]
|
126 |
embeddings_model = load_embeddings_model()
|
127 |
if embeddings_model and vector_store:
|
128 |
+
embeddings = embeddings_model.encode(chunks, batch_size=128, show_progress_bar=False)
|
129 |
vector_store.add_embeddings(zip(chunks, embeddings))
|
130 |
return vector_store
|
131 |
except Exception as e:
|
|
|
139 |
text = ""
|
140 |
code_blocks = []
|
141 |
with pdfplumber.open(BytesIO(uploaded_file.getvalue())) as pdf:
|
142 |
+
for page in pdf.pages[:8]:
|
143 |
extracted = page.extract_text(layout=False)
|
144 |
+
if not extracted:
|
145 |
try:
|
146 |
img = page.to_image(resolution=150).original
|
147 |
extracted = pytesseract.image_to_string(img, config='--psm 6')
|
148 |
except Exception as ocr_e:
|
149 |
logger.warning(f"OCR failed: {str(ocr_e)}")
|
150 |
if extracted:
|
151 |
+
# Clean text: remove headers/footers (simple heuristic)
|
152 |
+
lines = extracted.split("\n")
|
153 |
+
cleaned_lines = [line for line in lines if not re.match(r'^\s*(Page \d+|.*\d{4}-\d{4}|Copyright.*)\s*$', line, re.I)]
|
154 |
+
text += "\n".join(cleaned_lines) + "\n"
|
155 |
for char in page.chars:
|
156 |
if 'fontname' in char and 'mono' in char['fontname'].lower():
|
157 |
code_blocks.append(char['text'])
|
|
|
171 |
if not text:
|
172 |
raise ValueError("No text extracted from PDF")
|
173 |
|
174 |
+
text_splitter = CharacterTextSplitter(separator="\n\n", chunk_size=250, chunk_overlap=40, keep_separator=True)
|
175 |
+
text_chunks = text_splitter.split_text(text)[:25]
|
176 |
+
code_chunks = text_splitter.split_text(code_text)[:10] if code_text else []
|
177 |
|
178 |
embeddings_model = load_embeddings_model()
|
179 |
if not embeddings_model:
|
180 |
return None, None, text, code_text
|
181 |
|
182 |
text_vector_store = FAISS.from_embeddings(
|
183 |
+
zip(text_chunks, [embeddings_model.encode(chunk, show_progress_bar=False, batch_size=128) for chunk in text_chunks]),
|
184 |
embeddings_model.encode
|
185 |
) if text_chunks else None
|
186 |
code_vector_store = FAISS.from_embeddings(
|
187 |
+
zip(code_chunks, [embeddings_model.encode(chunk, show_progress_bar=False, batch_size=128) for chunk in code_chunks]),
|
188 |
embeddings_model.encode
|
189 |
) if code_chunks else None
|
190 |
|
|
|
198 |
st.error(f"PDF error: {str(e)}")
|
199 |
return None, None, "", ""
|
200 |
|
201 |
+
# Summarize PDF with ROUGE metrics and improved topic focus
|
202 |
def summarize_pdf(text):
|
203 |
logger.info("Generating summary")
|
204 |
try:
|
|
|
206 |
if not summary_pipeline:
|
207 |
return "Summary model unavailable."
|
208 |
|
209 |
+
text_splitter = CharacterTextSplitter(separator="\n\n", chunk_size=250, chunk_overlap=40)
|
210 |
+
chunks = text_splitter.split_text(text)
|
211 |
+
|
212 |
+
# Hybrid search for relevant chunks
|
213 |
+
embeddings_model = load_embeddings_model()
|
214 |
+
if embeddings_model and chunks:
|
215 |
+
temp_vector_store = FAISS.from_embeddings(
|
216 |
+
zip(chunks, [embeddings_model.encode(chunk, show_progress_bar=False) for chunk in chunks]),
|
217 |
+
embeddings_model.encode
|
218 |
+
)
|
219 |
+
bm25 = BM25Okapi([chunk.split() for chunk in chunks])
|
220 |
+
query = "main topic and key points"
|
221 |
+
bm25_docs = bm25.get_top_n(query.split(), chunks, n=4)
|
222 |
+
faiss_docs = temp_vector_store.similarity_search(query, k=4)
|
223 |
+
selected_chunks = list(set(bm25_docs + [doc.page_content for doc in faiss_docs]))[:4]
|
224 |
+
else:
|
225 |
+
selected_chunks = chunks[:4]
|
226 |
|
227 |
+
summaries = []
|
228 |
+
for chunk in selected_chunks:
|
229 |
+
summary = summary_pipeline(f"Summarize the main topic and key points in detail: {chunk[:250]}", max_length=100, min_length=50, do_sample=False)[0]['summary_text']
|
230 |
summaries.append(summary.strip())
|
231 |
|
232 |
combined_summary = " ".join(summaries)
|
233 |
+
if len(combined_summary.split()) > 250:
|
234 |
+
combined_summary = " ".join(combined_summary.split()[:250])
|
235 |
|
236 |
+
word_count = len(combined_summary.split())
|
237 |
scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
|
238 |
+
scores = scorer.score(text[:500], combined_summary)
|
239 |
logger.info(f"ROUGE scores: {scores}")
|
240 |
|
241 |
+
return f"**Main Topic Summary** ({word_count} words):\n{combined_summary}\n\n**ROUGE-1**: {scores['rouge1'].fmeasure:.2f}"
|
242 |
except Exception as e:
|
243 |
logger.error(f"Summary error: {str(e)}")
|
244 |
return f"Oops, something went wrong summarizing: {str(e)}"
|
|
|
304 |
""", unsafe_allow_html=True)
|
305 |
|
306 |
st.markdown('<div class="header"><h1>Smart PDF Q&A</h1></div>', unsafe_allow_html=True)
|
307 |
+
st.markdown("Upload a PDF to ask questions, get a ~200-word summary, or extract code with 'give me code'. Optimized for speed and accuracy!")
|
308 |
|
309 |
# Initialize session state
|
310 |
if "messages" not in st.session_state:
|
|
|
326 |
if st.button("Fine-Tune Model"):
|
327 |
progress_bar = st.progress(0)
|
328 |
for i in range(100):
|
329 |
+
time.sleep(0.008)
|
330 |
progress_bar.progress(i + 1)
|
331 |
dataset = load_and_prepare_dataset(dataset_name=dataset_name)
|
332 |
if dataset:
|
|
|
338 |
if st.button("Clear Chat"):
|
339 |
st.session_state.messages = []
|
340 |
st.experimental_rerun()
|
341 |
+
if st.button("Retry Summarization") and st.session_state.pdf_text:
|
342 |
+
progress_bar = st.progress(0)
|
343 |
+
with st.spinner("Retrying summarization..."):
|
344 |
+
for i in range(100):
|
345 |
+
time.sleep(0.008)
|
346 |
+
progress_bar.progress(i + 1)
|
347 |
+
summary = summarize_pdf(st.session_state.pdf_text)
|
348 |
+
st.session_state.messages.append({"role": "assistant", "content": summary})
|
349 |
+
st.markdown(summary, unsafe_allow_html=True)
|
350 |
st.markdown('</div>', unsafe_allow_html=True)
|
351 |
|
352 |
# PDF upload and processing
|
|
|
357 |
progress_bar = st.progress(0)
|
358 |
with st.spinner("Processing PDF..."):
|
359 |
for i in range(100):
|
360 |
+
time.sleep(0.02)
|
361 |
progress_bar.progress(i + 1)
|
362 |
st.session_state.text_vector_store, st.session_state.code_vector_store, st.session_state.pdf_text, st.session_state.code_text = process_pdf(uploaded_file)
|
363 |
if st.session_state.text_vector_store or st.session_state.code_vector_store:
|
|
|
370 |
progress_bar = st.progress(0)
|
371 |
with st.spinner("Summarizing..."):
|
372 |
for i in range(100):
|
373 |
+
time.sleep(0.008)
|
374 |
progress_bar.progress(i + 1)
|
375 |
summary = summarize_pdf(st.session_state.pdf_text)
|
376 |
st.session_state.messages.append({"role": "assistant", "content": summary})
|
|
|
388 |
progress_bar = st.progress(0)
|
389 |
with st.spinner('<div class="spinner">⏳ Processing...</div>'):
|
390 |
for i in range(100):
|
391 |
+
time.sleep(0.004)
|
392 |
progress_bar.progress(i + 1)
|
393 |
answer = answer_question(st.session_state.text_vector_store, st.session_state.code_vector_store, prompt)
|
394 |
st.markdown(answer, unsafe_allow_html=True)
|