Update app.py

app.py

@@ -1,105 +1,37 @@
-# -*- coding: utf-8 -*-
-"""Flan-t5-xl_with_GPU.ipynb
-
-Automatically generated by Colaboratory.
-
-Original file is located at
-    https://colab.research.google.com/drive/15P4GWaUNFBqJf5_I58DxSiusjPiHUeU6
-"""
-
-#import gradio as gr
-
-#def greet(name)
-#    return "Hello" + name + "!"
-
-#iface = gr.Interface(fn=greet, inputs="text", outputs="text")
-#iface.launch()
-
-#theme = gr.themes.Soft().set(
-#    body_background_fill='*background_fill_secondary',
-#    body_text_color_subdued='*body_text_color',
-#    body_text_color_subdued_dark='*chatbot_code_background_color'
-#)
-
-#app = gr.Interface(
-#    fn=qa_result,
-#    btn=gr.UploadButton("📁", file_types=[".pdf", ".csv", ".doc"], ),
-#    inputs=['textbox', 'text', 'file'],
-#    outputs='textbox',
-#    title='Բարև՛, ինչպե՞ս ես։',
-#    theme=theme,
-#    description='Ի՞նչ հարցեր ունես։'
-#)
-
-import os
-os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:512"
-
-from IPython.display import HTML, display
-
-def set_css():
-  display(HTML('''
-  <style>
-    pre {
-      white-space: pre-wrap;
-    }
-  </style>
-  '''))
-#get_ipython().events.register('pre_run_cell', set_css)
-
-import multiprocessing
-import torch
-
-torch.cuda.empty_cache()
-
-from deep_translator import GoogleTranslator
-
-# Use any translator you like, in this example GoogleTranslator
-#translated = GoogleTranslator(source='hy', target='en').translate("Բարև, ո՞նց ես։")  # output -> Hello, how are you?
-
-#device = "cuda:0" if torch.cuda.is_available() else "cpu"
-#device
-
 import streamlit as st
-
-#x = st.slider('Select a value')
-#st.write(x, 'squared is', x * x)
-
 import torch
 from transformers import T5Tokenizer, T5ForConditionalGeneration
+from deep_translator import GoogleTranslator
 
+# Initialize tokenizer and model
 tokenizer = T5Tokenizer.from_pretrained("google/flan-t5-xl")
-model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-xl", device_map = "auto")
-
-# We are running FP32!
-
-#my_text = "Summarize: \
-#Science can ignite new discoveries for society, \
-#Society has the tendency to refer to old, familiar ways of doing. \
-#Chaos is also a part of our society, although increasingly often so.\
-#Innovative ways lead to new growthin businesses and under certain conditions all of society participates."
-
-#my_text = "Write an essay with 100 words about Quantum Physics and it's problems regarding our understanding of laws of Physics."
-
-#my_text = "Q: Can Geoffrey Hinton have a conversation with George Washington? Give the rationale before answering."
-# my_text = "A short explanation of machine learning for medical applications."
-
-#def process():
-translated = st.text_input("Գրեք ձեր հարցը: ")
-my_text = GoogleTranslator(source='hy', target='en').translate(translated)
-
-input_ids = tokenizer(my_text, return_tensors = "pt").input_ids.to("cuda")
-
-outputs = model.generate(input_ids,
-                         min_length = 20,
-                         max_new_tokens = 600,
-                         length_penalty = 1.0, # Set to values < 1.0 in order to encourage the model to generate shorter answers.
-                         num_beams = 2,
-                         no_repeat_ngram_size = 3,
-                         temperature = 1.0,
-                         top_k = 150, # default 50
-                         top_p = 0.92,
-                         repetition_penalty = 2.1)
-
-st.write(GoogleTranslator(source='en', target='hy').translate(tokenizer.decode(outputs[0], skip_special_tokens = True)))
-    #process()
-
+model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-xl", device="cuda:0")
+
+# Function to generate text
+def generate_text(input_text):
+    input_ids = tokenizer(input_text, return_tensors="pt").input_ids.to("cuda:0")
+    
+    outputs = model.generate(input_ids,
+                             min_length=20,
+                             max_new_tokens=600,
+                             length_penalty=1.0,
+                             num_beams=2,
+                             no_repeat_ngram_size=3,
+                             temperature=1.0,
+                             top_k=150,
+                             top_p=0.92,
+                             repetition_penalty=2.1)
+    
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+# Streamlit app
+st.title("Text Generation with Flan-t5-xl")
+translated_input = st.text_input("Enter your question in Armenian:")
+if translated_input:
+    input_text = GoogleTranslator(source='hy', target='en').translate(translated_input)
+    generated_text = generate_text(input_text)
+    st.write("Generated Answer (Armenian):")
+    st.write(GoogleTranslator(source='en', target='hy').translate(generated_text))
+    torch.cuda.empty_cache()  # Release GPU memory
+
+# Note: The 'torch.cuda.empty_cache()' line releases GPU memory after generating each text.