Update app.py

app.py

@@ -2,34 +2,33 @@ import gradio as gr
 from transformers import pipeline
 import re
 
-# Load the model. gpt2-large is highly capable but small enough to run incredibly fast on CPU.
-# We use device=-1 to force CPU, which is the default for free HF Spaces.
-generator = pipeline("text-generation", model="gpt2-large", device=-1)
+# Load the model. Qwen 2.5 (1.5B parameters) is vastly superior to GPT-2.
+# It is a modern, un-gated base model that excels at text continuation.
+# device=-1 forces CPU execution for free HF Spaces.
+generator = pipeline("text-generation", model="Qwen/Qwen2.5-1.5B", device=-1)
 
 def predict_next_word(context):
     if not context.strip():
         return ""
         
-    # We generate up to 5 tokens to ensure we get a full word, 
-    # since models sometimes split long words into multiple tokens.
+    # Generate a few tokens to ensure we get a complete next word.
     result = generator(
         context,
         max_new_tokens=5,
-        return_full_text=False,  # Returns ONLY the new text, not the prompt
-        do_sample=False,         # Greedy decoding (forces the absolute most likely next path)
+        return_full_text=False,  # Returns ONLY the new text
+        do_sample=False,         # Greedy decoding (forces absolute highest probability path)
         pad_token_id=generator.tokenizer.eos_token_id
     )
     
     generated_text = result[0]["generated_text"]
     
-    # Extract just the very first word from the newly generated text
-    # This ignores trailing spaces and grabs the first sequence of word characters
+    # Cleanly extract just the very first word (ignoring leading spaces)
     match = re.search(r'^\s*([a-zA-Z0-9\'-]+)', generated_text)
     
     if match:
         next_word = match.group(1)
     else:
-        # Fallback in case it generates punctuation first
+        # Fallback in case the model generates punctuation first
         next_word = generated_text.strip().split()[0] if generated_text.strip() else ""
         
     return next_word
@@ -37,11 +36,11 @@ def predict_next_word(context):
 # Create the Gradio Interface
 demo = gr.Interface(
     fn=predict_next_word,
-    inputs=gr.Textbox(lines=2, placeholder="e.g., 'Dogs are'", label="Context"),
+    inputs=gr.Textbox(lines=2, placeholder="e.g., 'How is it'", label="Context"),
     outputs=gr.Textbox(label="Most Likely Next Word"),
-    title="Next Word Predictor",
+    title="Next Word Predictor (Powered by Qwen 2.5)",
     description="Enter a string of context and the model will instantly return the most likely next word.",
-    flagging_mode="never"  # <--- THIS IS THE FIX FOR GRADIO 4.0+
+    flagging_mode="never"
 )
 
 if __name__ == "__main__":