google

app.py

@@ -1,13 +1,16 @@
 import gradio as gr
 import re
+import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 from nltk.corpus import stopwords
-
-# Download the NLTK stopwords (only the first time you run)
+from spaces import GPU
 import nltk
+
+# Download stopwords if not already
 nltk.download('stopwords')
+stop_words = set(stopwords.words('english'))
 
-# Model choices ordered by accuracy
+# Model choices
 model_choices = {
     "Pegasus (google/pegasus-xsum)": "google/pegasus-xsum",
     "BigBird-Pegasus (google/bigbird-pegasus-large-arxiv)": "google/bigbird-pegasus-large-arxiv",
@@ -33,66 +36,54 @@ model_choices = {
 
 model_cache = {}
 
-# Get NLTK stopwords (common stop words)
-stop_words = set(stopwords.words('english'))
-
-# Function to clean input text by removing unnecessary words like stop words
 def clean_text(input_text):
-    # Replace special characters with a space
     cleaned_text = re.sub(r'[^A-Za-z0-9\s]', ' ', input_text)
-    
-    # Tokenize the input text and remove stop words
     words = cleaned_text.split()
     words = [word for word in words if word.lower() not in stop_words]
-    
-    # Rebuild the cleaned text
-    cleaned_text = " ".join(words)
-    
-    # Strip leading and trailing spaces
-    cleaned_text = cleaned_text.strip()
-    
-    return cleaned_text
+    return " ".join(words).strip()
 
-# Load model and tokenizer
 def load_model(model_name):
     if model_name not in model_cache:
         tokenizer = AutoTokenizer.from_pretrained(model_name)
-        model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+        model = AutoModelForSeq2SeqLM.from_pretrained(
+            model_name,
+            torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
+        )
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        model = model.to(device)
         model_cache[model_name] = (tokenizer, model)
+
+        # Warm up with dummy input
+        dummy_input = tokenizer("summarize: hello world", return_tensors="pt").input_ids.to(device)
+        model.generate(dummy_input, max_length=10)
+
     return model_cache[model_name]
 
-# Summarize the text using a selected model
+@GPU  # 👈 Required for ZeroGPU to allocate GPU when this is called
 def summarize_text(input_text, model_label, char_limit):
     if not input_text.strip():
         return "Please enter some text."
 
-    # Clean the input text by removing special characters and stop words
     input_text = clean_text(input_text)
-
     model_name = model_choices[model_label]
     tokenizer, model = load_model(model_name)
 
-    # Adjust the input format for T5 and FLAN models
     if "t5" in model_name.lower() or "flan" in model_name.lower():
         input_text = "summarize: " + input_text
 
+    device = model.device
     inputs = tokenizer(input_text, return_tensors="pt", truncation=True)
+    input_ids = inputs["input_ids"].to(device)
 
     summary_ids = model.generate(
-        inputs["input_ids"],
-        max_length=20,  # Still approximate; can be tuned per model
+        input_ids,
+        max_length=30,
         min_length=5,
         do_sample=False
     )
 
-    # Decode the summary while skipping special tokens and cleaning unwanted characters
     summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
-
-    # Remove unwanted characters like pipes or any unwanted symbols
-    summary = summary.replace("|", "")  # Remove pipes
-    summary = summary.strip()  # Remove leading/trailing whitespace
-
-    return summary[:char_limit]  # Enforce character limit
+    return summary[:char_limit].strip()
 
 # Gradio UI
 iface = gr.Interface(
@@ -100,11 +91,11 @@ iface = gr.Interface(
     inputs=[
         gr.Textbox(lines=6, label="Enter text to summarize"),
         gr.Dropdown(choices=list(model_choices.keys()), label="Choose summarization model", value="Pegasus (google/pegasus-xsum)"),
-        gr.Slider(minimum=30, maximum=200, value=65, step=1, label="Max Character Limit")
+        gr.Slider(minimum=30, maximum=200, value=80, step=1, label="Max Character Limit")
     ],
     outputs=gr.Textbox(lines=3, label="Summary (truncated to character limit)"),
-    title="Multi-Model Text Summarizer",
-    description="Summarize text using different Hugging Face models with a user-defined character limit."
+    title="Multi-Model Text Summarizer (GPU Ready)",
+    description="Summarize long or short texts using state-of-the-art Hugging Face models with GPU acceleration (ZeroGPU-compatible)."
 )
 
 iface.launch()