2001muhammadumair/Generative_Ai_Foundation_in_Python
0
1import os2import gradio as gr3import logging4from groq import Groq5from sentence_transformers import SentenceTransformer6import faiss7import numpy as np8import PyPDF29from sklearn.metrics.pairwise import cosine_similarity10from collections import Counter11 12# -------------------- Setup -------------------13 14logging.basicConfig(15 filename='query_logs.log',16 level=logging.INFO,17 format='%(asctime)s:%(levelname)s:%(message)s'18)19 20GROQ_API_KEY = "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"21client = Groq(api_key=GROQ_API_KEY)22PDF_PATH = 'Generative_AI_Foundations_in_Python_Discover_key_techniques_and.pdf'23sentence_transformer_model = SentenceTransformer('all-MiniLM-L6-v2')24cache = {}25 26# --------------------- Vectorization Function ---------------------27 28def vectorize_text(sentences_with_pages):29 """Vectorize sentences using SentenceTransformer and create a FAISS index."""30 try:31 sentences = [item['sentence'] for item in sentences_with_pages]32 embeddings = sentence_transformer_model.encode(sentences, show_progress_bar=True)33 index = faiss.IndexFlatL2(embeddings.shape[1])34 index.add(np.array(embeddings))35 logging.info(f"Added {len(sentences)} sentences to the vector store.")36 return index, sentences_with_pages37 except Exception as e:38 logging.error(f"Error during vectorization: {str(e)}")39 return None, None40 41# --------------------- PDF Processing ---------------------42 43def read_pdf(file_path):44 if not os.path.exists(file_path):45 logging.error(f"PDF file not found at: {file_path}")46 return []47 48 sentences_with_pages = []49 with open(file_path, 'rb') as file:50 reader = PyPDF2.PdfReader(file)51 for page_num, page in enumerate(reader.pages):52 text = page.extract_text()53 if text:54 sentences = [sentence.strip() for sentence in text.split('\n') if sentence.strip()]55 for sentence in sentences:56 sentences_with_pages.append({'sentence': sentence, 'page_number': page_num + 1})57 return sentences_with_pages58 59# Read and Vectorize PDF Content60sentences_with_pages = read_pdf(PDF_PATH)61vector_index, sentences_with_pages = vectorize_text(sentences_with_pages)62 63# --------------------- Query Handling ---------------------64 65def generate_query_embedding(query):66 return sentence_transformer_model.encode([query])67 68def is_query_relevant(distances, threshold=1.0):69 return distances[0][0] <= threshold70 71def generate_diverse_responses(prompt, n=3):72 responses = []73 for i in range(n):74 temperature = 0.7 + (i * 0.1)75 top_p = 0.9 - (i * 0.1)76 try:77 chat_completion = client.chat.completions.create(78 messages=[{"role": "user", "content": prompt}],79 model="llama3-8b-8192",80 temperature=temperature,81 top_p=top_p82 )83 responses.append(chat_completion.choices[0].message.content.strip())84 except Exception as e:85 logging.error(f"Error generating response: {str(e)}")86 responses.append("Error generating this response.")87 return responses88 89def aggregate_responses(responses):90 response_counter = Counter(responses)91 most_common_response, count = response_counter.most_common(1)[0]92 if count > 1:93 return most_common_response94 else:95 embeddings = sentence_transformer_model.encode(responses)96 avg_embedding = np.mean(embeddings, axis=0)97 similarities = cosine_similarity([avg_embedding], embeddings)[0]98 return responses[np.argmax(similarities)]99 100def generate_answer(query):101 if query in cache:102 logging.info(f"Cache hit for query: {query}")103 return cache[query]104 105 try:106 query_embedding = generate_query_embedding(query)107 D, I = vector_index.search(np.array(query_embedding), k=5)108 109 if is_query_relevant(D):110 relevant_items = [sentences_with_pages[i] for i in I[0]]111 combined_text = " ".join([item['sentence'] for item in relevant_items])112 page_numbers = sorted(set([item['page_number'] for item in relevant_items]))113 page_numbers_str = ', '.join(map(str, page_numbers))114 115 # Construct primary prompt116 prompt = f"""117Use the following context from "Generative AI Foundations" to answer the question. If additional explanation is needed, provide an example.118 119**Context (Pages {page_numbers_str}):**120{combined_text}121 122**User's question:**123{query}124 125**Remember to indicate the specific page numbers.**126"""127 primary_responses = generate_diverse_responses(prompt)128 primary_answer = aggregate_responses(primary_responses)129 130 # Construct additional prompt for explanations131 explanation_prompt = f"""132The user has a question about a complex topic. Could you provide an explanation or example for better understanding?133 134**User's question:**135{query}136 137**Primary answer:**138{primary_answer}139"""140 explanation_responses = generate_diverse_responses(explanation_prompt)141 explanation_answer = aggregate_responses(explanation_responses)142 143 # Combine primary answer and explanation144 full_response = f"{primary_answer}\n\n{explanation_answer}\n\n_From 'Generative AI Foundations,' pages {page_numbers_str}_"145 cache[query] = full_response146 logging.info(f"Generated response for query: {query}")147 return full_response148 149 else:150 # General knowledge fallback151 prompt = f"""152The user asked a question that is not covered in "Generative AI Foundations." Please provide a helpful answer using general knowledge.153 154**User's question:**155{query}156"""157 fallback_responses = generate_diverse_responses(prompt)158 fallback_answer = aggregate_responses(fallback_responses)159 cache[query] = fallback_answer160 return fallback_answer161 162 except Exception as e:163 logging.error(f"Error generating answer: {str(e)}")164 return "Sorry, an error occurred while generating the answer."165 166# --------------------- Gradio Interface ---------------------167 168def gradio_interface(user_query, history):169 response = generate_answer(user_query)170 history = history or []171 history.append({"role": "user", "content": user_query})172 history.append({"role": "assistant", "content": response})173 return history, history174 175# Create the Gradio interface176with gr.Blocks(css=".gradio-container {background-color: #f0f0f0}") as iface:177 gr.Markdown("""178 # **Generative AI Foundations Assistant**179 *Explore insights and get explanations with real-life examples from "Generative AI Foundations in Python".*180 """)181 182 chatbot = gr.Chatbot(height=500, type='messages')183 state = gr.State([])184 185 with gr.Row():186 txt = gr.Textbox(187 show_label=False,188 placeholder="Type your message here and press Enter",189 container=False190 )191 submit_btn = gr.Button("Send")192 193 def submit_message(user_query, history):194 history = history or []195 history.append({"role": "user", "content": user_query})196 return "", history197 198 def bot_response(history):199 user_query = history[-1]['content']200 response = generate_answer(user_query)201 history.append({"role": "assistant", "content": response})202 return history203 204 txt.submit(submit_message, [txt, state], [txt, state], queue=False).then(205 bot_response, state, chatbot206 )207 submit_btn.click(submit_message, [txt, state], [txt, state], queue=False).then(208 bot_response, state, chatbot209 )210 211 reset_btn = gr.Button("Reset Chat")212 reset_btn.click(lambda: ([], []), outputs=[chatbot, state], queue=False)213 214# Launch the Gradio app215if __name__ == "__main__":216 iface.launch()217