Create app.py

app.py

@@ -0,0 +1,64 @@
+import streamlit as st
+from transformers import AutoTokenizer, AutoModel, T5Tokenizer, T5ForConditionalGeneration
+import faiss
+import numpy as np
+
+# Load model and tokenizer for sentence transformers
+tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
+model = AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
+
+# Prepare dataset (Wikipedia dataset can be used)
+# Example: [title, text] pairs
+corpus = ["Article text 1", "Article text 2", "Article text 3"]
+
+# Tokenize and encode
+encoded_texts = [model(**tokenizer(text, return_tensors='pt', padding=True)).last_hidden_state.mean(1).detach().numpy() for text in corpus]
+
+# Create FAISS index
+dimension = encoded_texts[0].shape[1]
+index = faiss.IndexFlatL2(dimension)
+index.add(np.vstack(encoded_texts))
+
+def retrieve(query, k=5):
+    query_vector = model(**tokenizer(query, return_tensors='pt')).last_hidden_state.mean(1).detach().numpy()
+    distances, indices = index.search(query_vector, k)
+    return [corpus[i] for i in indices[0]]
+
+def generate_response(query):
+    retrieved_docs = retrieve(query)
+    context = " ".join(retrieved_docs)
+    
+    # Use the retrieved context to generate a humanized response
+    flan_t5_tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-base")
+    flan_t5_model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-base")
+
+    input_text = f"Generate a human-like response: {query}. Context: {context}"
+    input_ids = flan_t5_tokenizer(input_text, return_tensors="pt").input_ids
+
+    # Generate text with length constraint
+    generated_ids = flan_t5_model.generate(input_ids, max_length=1500)
+    response = flan_t5_tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+    return response
+
+def trim_to_word_limit(text, word_limit=1500):
+    words = text.split()
+    if len(words) > word_limit:
+        return " ".join(words[:word_limit])
+    return text
+
+# Streamlit UI
+st.title("Humanized Text Generator with RAG")
+
+# Input for the query
+query = st.text_input("Enter your query:")
+
+# Generate button
+if st.button("Generate"):
+    with st.spinner("Generating response..."):
+        response = generate_response(query)
+        response = trim_to_word_limit(response)
+        st.write("### Generated Response:")
+        st.write(response)
+
+# Additional info or about section
+st.write("This app uses FAISS, sentence-transformers, and FLAN-T5 to generate contextually relevant human-like responses.")