Update app.py

app.py

@@ -1,4 +1,4 @@
-from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
+from transformers import pipeline
 from datasets import load_dataset
 import gradio as gr
 import torch
@@ -11,12 +11,10 @@ Generate a poetry in Arabic.
 pipe_ar = pipeline('text-generation', framework='pt', model='akhooli/ap2023', tokenizer='akhooli/ap2023')
 
 """### **English: Text-Generation:**
-Generate a poetry in English using GPT-2 Poet model with AutoTokenizer and AutoModelForCausalLM.
+Generate a poetry in English.
 """
 
-# Load the tokenizer and model for the GPT-2 poetry model
-tokenizer_en = AutoTokenizer.from_pretrained("ashiqabdulkhader/GPT2-Poet")
-model_en = AutoModelForCausalLM.from_pretrained("ashiqabdulkhader/GPT2-Poet")
+pipe_en = pipeline("text-generation", model="ashiqabdulkhader/GPT2-Poet", from_tf=True)
 
 """### **Arabic and English: Text-To-Speech:**
 Convert the Arabic/English poetry to speech.
@@ -56,14 +54,14 @@ This function will receive 2 inputs from the Gradio interface, and execute the f
 def generate_poem(selected_language, text):
     try:
         if selected_language == "English":
-            poem = generate_poem_english(text)  # Return the generated poem from the generate_poem_english function
-            sampling_rate, audio_data = text_to_speech_english(poem)  # Return the audio from the text_to_speech_english function
-            image = generate_image_from_poem(poem)  # Return the image from the generate_image_from_poem function
+            poem = generate_poem_english(text) # Return the generated poem from the generate_poem_english function
+            sampling_rate, audio_data = text_to_speech_english(poem) # Return the audio from the text_to_speech_english function
+            image = generate_image_from_poem(text) # Return the image from the generate_image_from_poem function
         elif selected_language == "Arabic":
-            poem = generate_poem_arabic(text)  # Return the generated poem from the generate_poem_arabic function
-            sampling_rate, audio_data = text_to_speech_arabic(poem)  # Return the audio from the text_to_speech_arabic function
-            translated_text = translate_arabic_to_english(poem)  # Return the translated poem from Arabic to English
-            image = generate_image_from_poem(translated_text)  # Return the image from the generate_image_from_poem function
+            poem = generate_poem_arabic(text) # Return the generated poem from the generate_poem_arabic function
+            sampling_rate, audio_data = text_to_speech_arabic(poem) # Return the audio from the text_to_speech_arabic function
+            translated_text = translate_arabic_to_english(text) # Return the translated poem from Arabic to English
+            image = generate_image_from_poem(translated_text) # Return the image from the generate_image_from_poem function
 
         return poem, (sampling_rate, audio_data), image
     except Exception as e:
@@ -75,18 +73,20 @@ This function is responsible for generating a poem (text) in Arabic or English,
 
 # Poem generation for Arabic
 def generate_poem_arabic(text):
-    generated_text = pipe_ar(text, max_length=96, do_sample=True, temperature=1.0, top_k=50, top_p=0.9,
-                             repetition_penalty=1.2, min_length=64, no_repeat_ngram_size=3, num_beams=5,
-                             num_return_sequences=1)[0]["generated_text"]
-    clean_text = generated_text.replace("-", "")  # To get rid of the dashes generated by the model.
+    temp = 1.0
+    topk = 50
+    topp = 0.9
+    penalty = 1.2
+    generated_text = pipe_ar(text, max_length=96, do_sample=True, temperature=temp, top_k=topk, top_p=topp, repetition_penalty=penalty,
+                             min_length=64, no_repeat_ngram_size=3, return_full_text=True,
+                             num_beams=5, num_return_sequences=1)[0]["generated_text"]
+    clean_text = generated_text.replace("-", "") # To get rid of the dashes generated by the model.
     return clean_text
 
-# Poem generation for English using AutoTokenizer and AutoModelForCausalLM
+# Poem generation for English
 def generate_poem_english(text):
-    inputs = tokenizer_en(text, return_tensors="pt")
-    outputs = model_en.generate(**inputs, max_length=100, do_sample=True, top_k=50, top_p=0.9, temperature=1.0)
-    generated_text = tokenizer_en.decode(outputs[0], skip_special_tokens=True)
-    clean_text = generated_text.replace("</s>", "")  # Clean any unwanted tokens
+    generated_text = pipe_en(text, do_sample=True, max_length=100, top_k=50, top_p=0.9, temperature=1.0, num_return_sequences=3)[0]['generated_text']
+    clean_text = generated_text.replace("</s>", "") # To get rid of the </s> generated by the model.
     return clean_text
 
 """### **Audio Function:**
@@ -172,6 +172,7 @@ Provide 4 predefined inputs to demonstrate how the interface works.
 """
 
 examples = [
+    # First parameter is for the dropdown menu, and the second parameter is for the starter of the poem
     ["English", "The shining sun rises over the calm ocean"],
     ["Arabic", "الورود تتفتح في الربيع"],
     ["English", "The night sky is filled with stars and dreams"],
@@ -185,15 +186,17 @@ Creating a Gradio interface to generate a poem, read the poem, and generate an i
 my_model = gr.Interface(
     fn=generate_poem,  # The primary function that will receive the inputs (language and the starter of the poem)
     inputs=[
-        gr.Dropdown(["English", "Arabic"], label="Select Language"),  # Dropdown menu to select the language
+        gr.Dropdown(["English", "Arabic"], label="Select Language"), # Dropdown menu to select the language, either "English" or "Arabic"
         gr.Textbox(label="Enter a sentence")  # Textbox where the user will input a sentence or phrase to generate the poem
     ],
+
     outputs=[
         gr.Textbox(label="Generated Poem", lines=10),  # Textbox to display the generated poem
         gr.Audio(label="Generated Audio", type="numpy"),  # Audio output for the generated poem
         gr.Image(label="Generated Image")  # Image output for the generated image
     ],
+
     examples=examples,  # Predefined examples to guide the user
     css=custom_css  # Applying custom CSS
 )
-my_model.launch()
+my_model.launch()