Update app.py

app.py

@@ -1,36 +1,30 @@
 import streamlit as st
 from PIL import Image
 import io
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, AutoProcessor, AutoModelForSeq2SeqLM
-from easyocr import Reader
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
-# Load the OCR model and text generation model
+# Load the OCR model and text explanation model
 ocr_reader = Reader(['en'])
-text_generator = AutoModelForCausalLM.from_pretrained("gpt2")
-text_tokenizer = AutoTokenizer.from_pretrained("gpt2")
 
-# Load the image captioning model
-processor = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
-caption_model = AutoModelForSeq2SeqLM.from_pretrained("Salesforce/blip-image-captioning-large")
+# Load the text explanation model
+text_generator = AutoModelForCausalLM.from_pretrained("HuggingFaceH4/zephyr-7b-alpha")
+text_tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-alpha")
 
 # Define a function to extract text from an image using OCR
 def extract_text(image):
     return ocr_reader.readtext(image)
 
 # Define a function to explain the extracted text using text generation
-def explain_text(text, caption_model, processor):
+def explain_text(text, text_generator, text_tokenizer):
     # Extracted text
     extracted_text = " ".join([res[1] for res in text])
 
-    # Generate an image caption using the image captioning model
-    inputs = processor(extracted_text, return_tensors="pt", padding="max_length", max_length=100, truncation=True)
-    input_ids = inputs["input_ids"]
-    caption = caption_model.generate(input_ids, max_length=50, num_return_sequences=1, no_repeat_ngram_size=2)
-
-    # Decode and return the generated caption
-    generated_caption = processor.decode(caption[0], skip_special_tokens=True)
-    return generated_caption
+    # Generate an explanation using the text explanation model
+    input_ids = text_tokenizer.encode(extracted_text, return_tensors="pt")
+    explanation_ids = text_generator.generate(input_ids, max_length=100, num_return_sequences=1)
+    explanation = text_tokenizer.decode(explanation_ids[0], skip_special_tokens=True)
+    
+    return explanation
 
 # Create a Streamlit layout
 st.title("Text Extraction and Explanation")
@@ -42,12 +36,12 @@ uploaded_file = st.file_uploader("Upload an image:")
 if uploaded_file is not None:
     image = Image.open(uploaded_file)
     ocr_results = extract_text(image)
-    explanation = explain_text(ocr_results, caption_model, processor)
+    explanation = explain_text(ocr_results, text_generator, text_tokenizer)
 
     st.markdown("**Extracted text:**")
     st.markdown(" ".join([res[1] for res in ocr_results]))
 
-    st.markdown("**Explanation (Image Caption):**")
+    st.markdown("**Explanation:**")
     st.markdown(explanation)
 
 else: