Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -278,6 +278,175 @@
|
|
278 |
# ###################################################################################
|
279 |
|
280 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
281 |
import streamlit as st
|
282 |
import os
|
283 |
import re
|
@@ -286,7 +455,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
286 |
from PyPDF2 import PdfReader
|
287 |
from peft import get_peft_model, LoraConfig, TaskType
|
288 |
|
289 |
-
# β
Force CPU execution for
|
290 |
device = torch.device("cpu")
|
291 |
|
292 |
# πΉ Load IBM Granite Model (CPU-Compatible)
|
@@ -295,7 +464,7 @@ MODEL_NAME = "ibm-granite/granite-3.1-2b-instruct"
|
|
295 |
model = AutoModelForCausalLM.from_pretrained(
|
296 |
MODEL_NAME,
|
297 |
device_map="cpu", # Force CPU execution
|
298 |
-
torch_dtype=torch.float32 # Use float32 since
|
299 |
)
|
300 |
|
301 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
@@ -312,21 +481,6 @@ lora_config = LoraConfig(
|
|
312 |
model = get_peft_model(model, lora_config)
|
313 |
model.eval()
|
314 |
|
315 |
-
# π Function to Read & Extract Text from PDFs
|
316 |
-
# def read_files(file):
|
317 |
-
# file_context = ""
|
318 |
-
# try:
|
319 |
-
# reader = PdfReader(file)
|
320 |
-
# for page in reader.pages:
|
321 |
-
# text = page.extract_text()
|
322 |
-
# if text:
|
323 |
-
# file_context += text + "\n"
|
324 |
-
# except Exception as e:
|
325 |
-
# st.error(f"β οΈ Error reading PDF file: {e}")
|
326 |
-
# return ""
|
327 |
-
|
328 |
-
# return file_context.strip()
|
329 |
-
|
330 |
# π Function to Read & Extract Text from PDFs
|
331 |
def read_files(file):
|
332 |
file_context = ""
|
@@ -337,12 +491,8 @@ def read_files(file):
|
|
337 |
if text:
|
338 |
file_context += text + "\n"
|
339 |
|
340 |
-
if not file_context.strip():
|
341 |
-
return "β οΈ No text found. The document might be scanned or encrypted."
|
342 |
-
|
343 |
return file_context.strip()
|
344 |
|
345 |
-
|
346 |
# π Function to Format AI Prompts
|
347 |
def format_prompt(system_msg, user_msg, file_context=""):
|
348 |
if file_context:
|
@@ -354,9 +504,8 @@ def format_prompt(system_msg, user_msg, file_context=""):
|
|
354 |
|
355 |
# π Function to Generate AI Responses
|
356 |
def generate_response(input_text, max_tokens=1000, top_p=0.9, temperature=0.7):
|
357 |
-
st.write("π Generating response...") # Debugging message
|
358 |
model_inputs = tokenizer([input_text], return_tensors="pt").to(device)
|
359 |
-
|
360 |
with torch.no_grad():
|
361 |
output = model.generate(
|
362 |
**model_inputs,
|
@@ -367,10 +516,8 @@ def generate_response(input_text, max_tokens=1000, top_p=0.9, temperature=0.7):
|
|
367 |
num_return_sequences=1,
|
368 |
pad_token_id=tokenizer.eos_token_id
|
369 |
)
|
370 |
-
|
371 |
-
|
372 |
-
st.write("β
Response Generated!") # Debugging message
|
373 |
-
return response
|
374 |
|
375 |
# π Function to Clean AI Output
|
376 |
def post_process(text):
|
@@ -382,23 +529,18 @@ def post_process(text):
|
|
382 |
# π Function to Handle RAG with IBM Granite & Streamlit
|
383 |
def granite_simple(prompt, file):
|
384 |
file_context = read_files(file) if file else ""
|
385 |
-
|
386 |
-
# Debugging: Show extracted file content preview
|
387 |
-
if not file_context:
|
388 |
-
st.error("β οΈ No content extracted from the PDF. It might be a scanned image or encrypted.")
|
389 |
-
return "Error: No content found in the document."
|
390 |
-
|
391 |
system_message = "You are IBM Granite, a legal AI assistant specializing in contract analysis."
|
392 |
-
|
393 |
messages = format_prompt(system_message, prompt, file_context)
|
394 |
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
395 |
-
|
396 |
response = generate_response(input_text)
|
397 |
return post_process(response)
|
398 |
|
399 |
# πΉ Streamlit UI
|
400 |
def main():
|
401 |
-
st.set_page_config(page_title="Contract Analysis AI", page_icon="π")
|
402 |
|
403 |
st.title("π AI-Powered Contract Analysis Tool")
|
404 |
st.write("Upload a contract document (PDF) for a detailed AI-driven legal and technical analysis.")
|
@@ -413,38 +555,35 @@ def main():
|
|
413 |
# πΉ File Upload Section
|
414 |
uploaded_file = st.file_uploader("π Upload a contract document (PDF)", type="pdf")
|
415 |
|
416 |
-
if uploaded_file:
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
# Debugging: Show extracted text preview
|
421 |
-
pdf_text = read_files(uploaded_file)
|
422 |
-
if pdf_text:
|
423 |
-
st.write("**Extracted Sample Text:**")
|
424 |
-
st.code(pdf_text[:500]) # Show first 500 characters
|
425 |
-
else:
|
426 |
-
st.error("β οΈ No readable text found in the document.")
|
427 |
|
428 |
-
st.
|
429 |
|
430 |
-
#
|
431 |
-
|
432 |
|
433 |
if st.button("π Analyze Document"):
|
434 |
with st.spinner("Analyzing contract document... β³"):
|
435 |
-
final_answer = granite_simple(
|
436 |
-
"Perform a detailed technical analysis of the attached contract document, highlighting potential risks, legal pitfalls, compliance issues, and areas where contractual terms may lead to future disputes or operational challenges.",
|
437 |
-
uploaded_file
|
438 |
-
)
|
439 |
|
440 |
# πΉ Display Analysis Result
|
441 |
st.subheader("π Analysis Result")
|
442 |
st.write(final_answer)
|
443 |
|
|
|
|
|
|
|
444 |
# π₯ Run Streamlit App
|
445 |
if __name__ == '__main__':
|
446 |
main()
|
447 |
|
|
|
|
|
|
|
|
|
448 |
# import streamlit as st
|
449 |
# from PyPDF2 import PdfReader
|
450 |
|
|
|
278 |
# ###################################################################################
|
279 |
|
280 |
|
281 |
+
# import streamlit as st
|
282 |
+
# import os
|
283 |
+
# import re
|
284 |
+
# import torch
|
285 |
+
# from transformers import AutoModelForCausalLM, AutoTokenizer
|
286 |
+
# from PyPDF2 import PdfReader
|
287 |
+
# from peft import get_peft_model, LoraConfig, TaskType
|
288 |
+
|
289 |
+
# # β
Force CPU execution for Hugging Face Spaces
|
290 |
+
# device = torch.device("cpu")
|
291 |
+
|
292 |
+
# # πΉ Load IBM Granite Model (CPU-Compatible)
|
293 |
+
# MODEL_NAME = "ibm-granite/granite-3.1-2b-instruct"
|
294 |
+
|
295 |
+
# model = AutoModelForCausalLM.from_pretrained(
|
296 |
+
# MODEL_NAME,
|
297 |
+
# device_map="cpu", # Force CPU execution
|
298 |
+
# torch_dtype=torch.float32 # Use float32 since Hugging Face runs on CPU
|
299 |
+
# )
|
300 |
+
|
301 |
+
# tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
302 |
+
|
303 |
+
# # πΉ Apply LoRA Fine-Tuning Configuration
|
304 |
+
# lora_config = LoraConfig(
|
305 |
+
# r=8,
|
306 |
+
# lora_alpha=32,
|
307 |
+
# target_modules=["q_proj", "v_proj"],
|
308 |
+
# lora_dropout=0.1,
|
309 |
+
# bias="none",
|
310 |
+
# task_type=TaskType.CAUSAL_LM
|
311 |
+
# )
|
312 |
+
# model = get_peft_model(model, lora_config)
|
313 |
+
# model.eval()
|
314 |
+
|
315 |
+
# # π Function to Read & Extract Text from PDFs
|
316 |
+
# # def read_files(file):
|
317 |
+
# # file_context = ""
|
318 |
+
# # try:
|
319 |
+
# # reader = PdfReader(file)
|
320 |
+
# # for page in reader.pages:
|
321 |
+
# # text = page.extract_text()
|
322 |
+
# # if text:
|
323 |
+
# # file_context += text + "\n"
|
324 |
+
# # except Exception as e:
|
325 |
+
# # st.error(f"β οΈ Error reading PDF file: {e}")
|
326 |
+
# # return ""
|
327 |
+
|
328 |
+
# # return file_context.strip()
|
329 |
+
|
330 |
+
# # π Function to Read & Extract Text from PDFs
|
331 |
+
# def read_files(file):
|
332 |
+
# file_context = ""
|
333 |
+
# reader = PdfReader(file)
|
334 |
+
|
335 |
+
# for page in reader.pages:
|
336 |
+
# text = page.extract_text()
|
337 |
+
# if text:
|
338 |
+
# file_context += text + "\n"
|
339 |
+
|
340 |
+
# if not file_context.strip():
|
341 |
+
# return "β οΈ No text found. The document might be scanned or encrypted."
|
342 |
+
|
343 |
+
# return file_context.strip()
|
344 |
+
|
345 |
+
|
346 |
+
# # π Function to Format AI Prompts
|
347 |
+
# def format_prompt(system_msg, user_msg, file_context=""):
|
348 |
+
# if file_context:
|
349 |
+
# system_msg += f" The user has provided a contract document. Use its context to generate insights, but do not repeat or summarize the document itself."
|
350 |
+
# return [
|
351 |
+
# {"role": "system", "content": system_msg},
|
352 |
+
# {"role": "user", "content": user_msg}
|
353 |
+
# ]
|
354 |
+
|
355 |
+
# # π Function to Generate AI Responses
|
356 |
+
# def generate_response(input_text, max_tokens=1000, top_p=0.9, temperature=0.7):
|
357 |
+
# st.write("π Generating response...") # Debugging message
|
358 |
+
# model_inputs = tokenizer([input_text], return_tensors="pt").to(device)
|
359 |
+
|
360 |
+
# with torch.no_grad():
|
361 |
+
# output = model.generate(
|
362 |
+
# **model_inputs,
|
363 |
+
# max_new_tokens=max_tokens,
|
364 |
+
# do_sample=True,
|
365 |
+
# top_p=top_p,
|
366 |
+
# temperature=temperature,
|
367 |
+
# num_return_sequences=1,
|
368 |
+
# pad_token_id=tokenizer.eos_token_id
|
369 |
+
# )
|
370 |
+
|
371 |
+
# response = tokenizer.decode(output[0], skip_special_tokens=True)
|
372 |
+
# st.write("β
Response Generated!") # Debugging message
|
373 |
+
# return response
|
374 |
+
|
375 |
+
# # π Function to Clean AI Output
|
376 |
+
# def post_process(text):
|
377 |
+
# cleaned = re.sub(r'ζ₯+', '', text) # Remove unwanted symbols
|
378 |
+
# lines = cleaned.splitlines()
|
379 |
+
# unique_lines = list(dict.fromkeys([line.strip() for line in lines if line.strip()]))
|
380 |
+
# return "\n".join(unique_lines)
|
381 |
+
|
382 |
+
# # π Function to Handle RAG with IBM Granite & Streamlit
|
383 |
+
# def granite_simple(prompt, file):
|
384 |
+
# file_context = read_files(file) if file else ""
|
385 |
+
|
386 |
+
# # Debugging: Show extracted file content preview
|
387 |
+
# if not file_context:
|
388 |
+
# st.error("β οΈ No content extracted from the PDF. It might be a scanned image or encrypted.")
|
389 |
+
# return "Error: No content found in the document."
|
390 |
+
|
391 |
+
# system_message = "You are IBM Granite, a legal AI assistant specializing in contract analysis."
|
392 |
+
|
393 |
+
# messages = format_prompt(system_message, prompt, file_context)
|
394 |
+
# input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
395 |
+
|
396 |
+
# response = generate_response(input_text)
|
397 |
+
# return post_process(response)
|
398 |
+
|
399 |
+
# # πΉ Streamlit UI
|
400 |
+
# def main():
|
401 |
+
# st.set_page_config(page_title="Contract Analysis AI", page_icon="π")
|
402 |
+
|
403 |
+
# st.title("π AI-Powered Contract Analysis Tool")
|
404 |
+
# st.write("Upload a contract document (PDF) for a detailed AI-driven legal and technical analysis.")
|
405 |
+
|
406 |
+
# # πΉ Sidebar Settings
|
407 |
+
# with st.sidebar:
|
408 |
+
# st.header("βοΈ Settings")
|
409 |
+
# max_tokens = st.slider("Max Tokens", 50, 1000, 250, 50)
|
410 |
+
# top_p = st.slider("Top P (sampling)", 0.1, 1.0, 0.9, 0.1)
|
411 |
+
# temperature = st.slider("Temperature (creativity)", 0.1, 1.0, 0.7, 0.1)
|
412 |
+
|
413 |
+
# # πΉ File Upload Section
|
414 |
+
# uploaded_file = st.file_uploader("π Upload a contract document (PDF)", type="pdf")
|
415 |
+
|
416 |
+
# if uploaded_file:
|
417 |
+
# st.success(f"β
File uploaded successfully! File Name: {uploaded_file.name}")
|
418 |
+
# st.write(f"**File Size:** {uploaded_file.size / 1024:.2f} KB")
|
419 |
+
|
420 |
+
# # Debugging: Show extracted text preview
|
421 |
+
# pdf_text = read_files(uploaded_file)
|
422 |
+
# if pdf_text:
|
423 |
+
# st.write("**Extracted Sample Text:**")
|
424 |
+
# st.code(pdf_text[:500]) # Show first 500 characters
|
425 |
+
# else:
|
426 |
+
# st.error("β οΈ No readable text found in the document.")
|
427 |
+
|
428 |
+
# st.write("Click the button below to analyze the contract.")
|
429 |
+
|
430 |
+
# # Force button to always render
|
431 |
+
# st.markdown('<style>div.stButton > button {display: block; width: 100%;}</style>', unsafe_allow_html=True)
|
432 |
+
|
433 |
+
# if st.button("π Analyze Document"):
|
434 |
+
# with st.spinner("Analyzing contract document... β³"):
|
435 |
+
# final_answer = granite_simple(
|
436 |
+
# "Perform a detailed technical analysis of the attached contract document, highlighting potential risks, legal pitfalls, compliance issues, and areas where contractual terms may lead to future disputes or operational challenges.",
|
437 |
+
# uploaded_file
|
438 |
+
# )
|
439 |
+
|
440 |
+
# # πΉ Display Analysis Result
|
441 |
+
# st.subheader("π Analysis Result")
|
442 |
+
# st.write(final_answer)
|
443 |
+
|
444 |
+
# π₯ Run Streamlit App
|
445 |
+
# if __name__ == '__main__':
|
446 |
+
# main()
|
447 |
+
|
448 |
+
|
449 |
+
|
450 |
import streamlit as st
|
451 |
import os
|
452 |
import re
|
|
|
455 |
from PyPDF2 import PdfReader
|
456 |
from peft import get_peft_model, LoraConfig, TaskType
|
457 |
|
458 |
+
# β
Force CPU execution for Streamlit Cloud
|
459 |
device = torch.device("cpu")
|
460 |
|
461 |
# πΉ Load IBM Granite Model (CPU-Compatible)
|
|
|
464 |
model = AutoModelForCausalLM.from_pretrained(
|
465 |
MODEL_NAME,
|
466 |
device_map="cpu", # Force CPU execution
|
467 |
+
torch_dtype=torch.float32 # Use float32 since Streamlit runs on CPU
|
468 |
)
|
469 |
|
470 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
|
|
481 |
model = get_peft_model(model, lora_config)
|
482 |
model.eval()
|
483 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
484 |
# π Function to Read & Extract Text from PDFs
|
485 |
def read_files(file):
|
486 |
file_context = ""
|
|
|
491 |
if text:
|
492 |
file_context += text + "\n"
|
493 |
|
|
|
|
|
|
|
494 |
return file_context.strip()
|
495 |
|
|
|
496 |
# π Function to Format AI Prompts
|
497 |
def format_prompt(system_msg, user_msg, file_context=""):
|
498 |
if file_context:
|
|
|
504 |
|
505 |
# π Function to Generate AI Responses
|
506 |
def generate_response(input_text, max_tokens=1000, top_p=0.9, temperature=0.7):
|
|
|
507 |
model_inputs = tokenizer([input_text], return_tensors="pt").to(device)
|
508 |
+
|
509 |
with torch.no_grad():
|
510 |
output = model.generate(
|
511 |
**model_inputs,
|
|
|
516 |
num_return_sequences=1,
|
517 |
pad_token_id=tokenizer.eos_token_id
|
518 |
)
|
519 |
+
|
520 |
+
return tokenizer.decode(output[0], skip_special_tokens=True)
|
|
|
|
|
521 |
|
522 |
# π Function to Clean AI Output
|
523 |
def post_process(text):
|
|
|
529 |
# π Function to Handle RAG with IBM Granite & Streamlit
|
530 |
def granite_simple(prompt, file):
|
531 |
file_context = read_files(file) if file else ""
|
532 |
+
|
|
|
|
|
|
|
|
|
|
|
533 |
system_message = "You are IBM Granite, a legal AI assistant specializing in contract analysis."
|
534 |
+
|
535 |
messages = format_prompt(system_message, prompt, file_context)
|
536 |
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
537 |
+
|
538 |
response = generate_response(input_text)
|
539 |
return post_process(response)
|
540 |
|
541 |
# πΉ Streamlit UI
|
542 |
def main():
|
543 |
+
st.set_page_config(page_title="Contract Analysis AI", page_icon="π", layout="wide")
|
544 |
|
545 |
st.title("π AI-Powered Contract Analysis Tool")
|
546 |
st.write("Upload a contract document (PDF) for a detailed AI-driven legal and technical analysis.")
|
|
|
555 |
# πΉ File Upload Section
|
556 |
uploaded_file = st.file_uploader("π Upload a contract document (PDF)", type="pdf")
|
557 |
|
558 |
+
if uploaded_file is not None:
|
559 |
+
temp_file_path = "temp_uploaded_contract.pdf"
|
560 |
+
with open(temp_file_path, "wb") as f:
|
561 |
+
f.write(uploaded_file.getbuffer())
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
562 |
|
563 |
+
st.success("β
File uploaded successfully!")
|
564 |
|
565 |
+
# πΉ User Input for Analysis
|
566 |
+
user_prompt = "Perform a detailed technical analysis of the attached contract document, highlighting potential risks, legal pitfalls, compliance issues, and areas where contractual terms may lead to future disputes or operational challenges."
|
567 |
|
568 |
if st.button("π Analyze Document"):
|
569 |
with st.spinner("Analyzing contract document... β³"):
|
570 |
+
final_answer = granite_simple(user_prompt, temp_file_path)
|
|
|
|
|
|
|
571 |
|
572 |
# πΉ Display Analysis Result
|
573 |
st.subheader("π Analysis Result")
|
574 |
st.write(final_answer)
|
575 |
|
576 |
+
# πΉ Remove Temporary File
|
577 |
+
os.remove(temp_file_path)
|
578 |
+
|
579 |
# π₯ Run Streamlit App
|
580 |
if __name__ == '__main__':
|
581 |
main()
|
582 |
|
583 |
+
|
584 |
+
|
585 |
+
|
586 |
+
|
587 |
# import streamlit as st
|
588 |
# from PyPDF2 import PdfReader
|
589 |
|