Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,195 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import gradio as gr
|
3 |
+
import pdfplumber
|
4 |
+
import requests
|
5 |
+
import faiss
|
6 |
+
import json
|
7 |
+
import torch
|
8 |
+
from bs4 import BeautifulSoup
|
9 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
10 |
+
from sentence_transformers import SentenceTransformer
|
11 |
+
import numpy as np
|
12 |
+
import tempfile
|
13 |
+
import logging
|
14 |
+
from datetime import datetime
|
15 |
+
from typing import List, Dict
|
16 |
+
|
17 |
+
# Optimize CUDA memory management
|
18 |
+
os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'expandable_segments:True'
|
19 |
+
|
20 |
+
# Configure logging
|
21 |
+
logging.basicConfig(level=logging.INFO)
|
22 |
+
logger = logging.getLogger(__name__)
|
23 |
+
|
24 |
+
class CaseStudyGenerator:
|
25 |
+
def __init__(self):
|
26 |
+
self.model_name = "facebook/opt-2.7b"
|
27 |
+
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
|
28 |
+
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
29 |
+
|
30 |
+
# Clear any reserved memory
|
31 |
+
if self.device == "cuda":
|
32 |
+
torch.cuda.empty_cache()
|
33 |
+
torch.cuda.ipc_collect()
|
34 |
+
|
35 |
+
model_kwargs = {
|
36 |
+
'torch_dtype': torch.float16 if self.device == "cuda" else torch.float32
|
37 |
+
}
|
38 |
+
|
39 |
+
try:
|
40 |
+
self.model = AutoModelForCausalLM.from_pretrained(self.model_name, **model_kwargs)
|
41 |
+
if self.device == "cuda":
|
42 |
+
self.model = self.model.to(self.device)
|
43 |
+
self.model.gradient_checkpointing_enable()
|
44 |
+
except RuntimeError as e:
|
45 |
+
logger.warning(f"Memory issue detected: {e}, attempting 8-bit loading.")
|
46 |
+
|
47 |
+
try:
|
48 |
+
from transformers import BitsAndBytesConfig
|
49 |
+
quantization_config = BitsAndBytesConfig(load_in_8bit=True)
|
50 |
+
self.model = AutoModelForCausalLM.from_pretrained(self.model_name, quantization_config=quantization_config)
|
51 |
+
except ImportError:
|
52 |
+
logger.error("Missing 'bitsandbytes'. Install it using 'pip install -U bitsandbytes'")
|
53 |
+
logger.info("Switching to CPU to continue operations.")
|
54 |
+
self.device = "cpu"
|
55 |
+
self.model = AutoModelForCausalLM.from_pretrained(self.model_name, torch_dtype=torch.float32)
|
56 |
+
|
57 |
+
self.generator = pipeline(
|
58 |
+
"text-generation",
|
59 |
+
model=self.model,
|
60 |
+
tokenizer=self.tokenizer,
|
61 |
+
device=0 if self.device == "cuda" else -1,
|
62 |
+
max_length=2048,
|
63 |
+
num_return_sequences=1,
|
64 |
+
temperature=0.8,
|
65 |
+
top_p=0.95,
|
66 |
+
do_sample=True,
|
67 |
+
pad_token_id=self.tokenizer.eos_token_id
|
68 |
+
)
|
69 |
+
|
70 |
+
self.embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
|
71 |
+
self.dimension = 384
|
72 |
+
self.index = faiss.IndexFlatL2(self.dimension)
|
73 |
+
self.stored_texts: List[Dict] = []
|
74 |
+
|
75 |
+
def clean_url(self, url: str) -> str:
|
76 |
+
if not url.startswith(('http://', 'https://')):
|
77 |
+
return ""
|
78 |
+
return url.split('?')[0][:100]
|
79 |
+
|
80 |
+
def fetch_articles(self, topic: str) -> List[str]:
|
81 |
+
try:
|
82 |
+
search_url = f"https://www.google.com/search?q={topic.replace(' ', '+')}+case+study+manufacturing+strategy"
|
83 |
+
headers = {"User-Agent": "Mozilla/5.0"}
|
84 |
+
response = requests.get(search_url, headers=headers, timeout=10)
|
85 |
+
response.raise_for_status()
|
86 |
+
|
87 |
+
soup = BeautifulSoup(response.text, "html.parser")
|
88 |
+
articles = [self.clean_url(link.get("href", "")) for link in soup.find_all("a") if "google" not in link.get("href", "")]
|
89 |
+
return articles[:5] or ["No articles found"]
|
90 |
+
except Exception as e:
|
91 |
+
logger.error(f"Error fetching articles: {str(e)}")
|
92 |
+
return ["Error fetching articles"]
|
93 |
+
|
94 |
+
def process_pdf(self, pdf_file) -> str:
|
95 |
+
try:
|
96 |
+
if pdf_file is None:
|
97 |
+
return "No PDF provided"
|
98 |
+
|
99 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix='.pdf') as temp_pdf:
|
100 |
+
temp_pdf.write(pdf_file.read())
|
101 |
+
temp_path = temp_pdf.name
|
102 |
+
|
103 |
+
text = []
|
104 |
+
with pdfplumber.open(temp_path) as pdf:
|
105 |
+
text = [page.extract_text().strip() for page in pdf.pages if page.extract_text()]
|
106 |
+
|
107 |
+
os.unlink(temp_path)
|
108 |
+
return "\n".join(text) or "No text extracted from PDF"
|
109 |
+
except Exception as e:
|
110 |
+
logger.error(f"Error processing PDF: {str(e)}")
|
111 |
+
return "Error processing PDF"
|
112 |
+
|
113 |
+
def generate_case_study(self, topic: str, pdf=None) -> str:
|
114 |
+
try:
|
115 |
+
if self.device == "cuda":
|
116 |
+
torch.cuda.empty_cache()
|
117 |
+
|
118 |
+
articles = self.fetch_articles(topic)
|
119 |
+
pdf_text = self.process_pdf(pdf) if pdf else "No PDF provided"
|
120 |
+
|
121 |
+
prompt = f"""Write a professional case study about {topic}.
|
122 |
+
Background Information:
|
123 |
+
- Topic: {topic}
|
124 |
+
- Supporting Documents: {pdf_text[:500]}
|
125 |
+
- Related Sources: {', '.join(articles)}
|
126 |
+
|
127 |
+
Format your response as:
|
128 |
+
1. Executive Summary
|
129 |
+
2. Company Background
|
130 |
+
3. Challenge Analysis
|
131 |
+
4. Strategic Implementation
|
132 |
+
5. Results and Impact
|
133 |
+
6. Key Learnings
|
134 |
+
"""
|
135 |
+
|
136 |
+
output = self.generator(
|
137 |
+
prompt,
|
138 |
+
max_new_tokens=1024,
|
139 |
+
num_return_sequences=1,
|
140 |
+
temperature=0.8,
|
141 |
+
top_p=0.95,
|
142 |
+
do_sample=True,
|
143 |
+
repetition_penalty=1.2,
|
144 |
+
no_repeat_ngram_size=3
|
145 |
+
)
|
146 |
+
|
147 |
+
case_study = output[0]['generated_text'].replace(prompt, "").strip()
|
148 |
+
embedding = self.embedding_model.encode([case_study])[0]
|
149 |
+
self.index.add(embedding.reshape(1, -1))
|
150 |
+
|
151 |
+
self.stored_texts.append({
|
152 |
+
"timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
153 |
+
"topic": topic,
|
154 |
+
"content": case_study
|
155 |
+
})
|
156 |
+
|
157 |
+
return case_study
|
158 |
+
except Exception as e:
|
159 |
+
logger.error(f"Error generating case study: {str(e)}")
|
160 |
+
return f"Error generating case study: {str(e)}"
|
161 |
+
|
162 |
+
def retrieve_past_case_studies(self) -> str:
|
163 |
+
try:
|
164 |
+
if not self.stored_texts:
|
165 |
+
return "No case studies generated yet."
|
166 |
+
|
167 |
+
result = ""
|
168 |
+
for idx, case in enumerate(self.stored_texts[-5:], start=1):
|
169 |
+
result += f"Case Study {idx}\nTopic: {case['topic']}\nGenerated on: {case['timestamp']}\n\n{case['content']}\n\n=== End of Case Study ===\n\n"
|
170 |
+
return result
|
171 |
+
except Exception as e:
|
172 |
+
logger.error(f"Error retrieving past case studies: {str(e)}")
|
173 |
+
return "Error retrieving past case studies"
|
174 |
+
|
175 |
+
# Gradio interface
|
176 |
+
with gr.Blocks() as app:
|
177 |
+
gr.Markdown("# AI Case Study Generator (Optimized for GPU-T4 & CPU)")
|
178 |
+
with gr.Row():
|
179 |
+
topic = gr.Textbox(label="Enter Topic")
|
180 |
+
pdf = gr.File(label="Upload PDF", type="binary")
|
181 |
+
with gr.Row():
|
182 |
+
generate_btn = gr.Button("Generate Case Study")
|
183 |
+
retrieve_btn = gr.Button("Retrieve Past Case Studies")
|
184 |
+
output = gr.Textbox(label="Generated Case Study", lines=20)
|
185 |
+
past_cases = gr.Textbox(label="Past Case Studies", lines=20)
|
186 |
+
|
187 |
+
generator = CaseStudyGenerator()
|
188 |
+
generate_btn.click(generator.generate_case_study, inputs=[topic, pdf], outputs=output)
|
189 |
+
retrieve_btn.click(generator.retrieve_past_case_studies, outputs=past_cases)
|
190 |
+
|
191 |
+
# Launch the application
|
192 |
+
if __name__ == "__main__":
|
193 |
+
app.launch(share=True) # Remove enable_queue
|
194 |
+
# or, If using Gradio 3.x or later, use:
|
195 |
+
# app.queue().launch(share=True)
|