google

app.py

@@ -1,9 +1,8 @@
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
-# Supported summarization models
+# Model choices ordered by accuracy
 model_choices = {
-    # 🥇 High Accuracy Models
     "Pegasus (google/pegasus-xsum)": "google/pegasus-xsum",
     "BigBird-Pegasus (google/bigbird-pegasus-large-arxiv)": "google/bigbird-pegasus-large-arxiv",
     "LongT5 Large (google/long-t5-tglobal-large)": "google/long-t5-tglobal-large",
@@ -12,8 +11,6 @@ model_choices = {
     "LED (allenai/led-base-16384)": "allenai/led-base-16384",
     "T5 Large (t5-large)": "t5-large",
     "Flan-T5 Large (google/flan-t5-large)": "google/flan-t5-large",
-
-    # ⚖️ Balanced (Speed vs Accuracy)
     "DistilBART CNN (sshleifer/distilbart-cnn-12-6)": "sshleifer/distilbart-cnn-12-6",
     "DistilBART XSum (mrm8488/distilbart-xsum-12-6)": "mrm8488/distilbart-xsum-12-6",
     "T5 Base (t5-base)": "t5-base",
@@ -21,16 +18,13 @@ model_choices = {
     "BART CNN SamSum (philschmid/bart-large-cnn-samsum)": "philschmid/bart-large-cnn-samsum",
     "T5 SamSum (knkarthick/pegasus-samsum)": "knkarthick/pegasus-samsum",
     "LongT5 Base (google/long-t5-tglobal-base)": "google/long-t5-tglobal-base",
-
-    # ⚡ Lighter / Faster Models
     "T5 Small (t5-small)": "t5-small",
     "MBART (facebook/mbart-large-cc25)": "facebook/mbart-large-cc25",
-    "MarianMT (Helsinki-NLP/opus-mt-en-ro)": "Helsinki-NLP/opus-mt-en-ro",  # not trained for summarization, just as placeholder
-    "Falcon Instruct (tiiuae/falcon-7b-instruct)": "tiiuae/falcon-7b-instruct",  # general-purpose, not summarization-specific
-    "BART ELI5 (yjernite/bart_eli5)": "yjernite/bart_eli5"  # trained for explain-like-I'm-5
+    "MarianMT (Helsinki-NLP/opus-mt-en-ro)": "Helsinki-NLP/opus-mt-en-ro",
+    "Falcon Instruct (tiiuae/falcon-7b-instruct)": "tiiuae/falcon-7b-instruct",
+    "BART ELI5 (yjernite/bart_eli5)": "yjernite/bart_eli5"
 }
 
-# Cache for loaded models/tokenizers
 model_cache = {}
 
 def load_model(model_name):
@@ -40,38 +34,39 @@ def load_model(model_name):
         model_cache[model_name] = (tokenizer, model)
     return model_cache[model_name]
 
-# Summarization function
-def summarize_text(input_text, model_label):
+def summarize_text(input_text, model_label, char_limit):
     if not input_text.strip():
         return "Please enter some text."
 
     model_name = model_choices[model_label]
     tokenizer, model = load_model(model_name)
 
-    if "t5" in model_name.lower():
+    if "t5" in model_name.lower() or "flan" in model_name.lower():
         input_text = "summarize: " + input_text
 
     inputs = tokenizer(input_text, return_tensors="pt", truncation=True)
+
     summary_ids = model.generate(
         inputs["input_ids"],
-        max_length=20,  # Approximate for 65 characters
+        max_length=20,  # Still approximate; can be tuned per model
         min_length=5,
         do_sample=False
     )
-    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
 
-    return summary[:65]  # Ensure character limit
+    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
+    return summary[:char_limit]  # Enforce character limit
 
 # Gradio UI
 iface = gr.Interface(
     fn=summarize_text,
     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.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")
     ],
-    outputs=gr.Textbox(lines=2, label="Summary (max 65 characters)"),
-    title="Short Text Summarizer",
-    description="Summarizes input text to under 65 characters using a selected model."
+    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."
 )
 
 iface.launch()