app.py

import gradio as gr
from transformers import pipeline, set_seed
import re
import numpy as np
import pandas as pd

# Set a seed for reproducibility
set_seed(42)

# Define five small models for generation (free, lightweight)
small_models = [
    "distilgpt2",                    # ~82M parameters
    "gpt2",                          # ~124M parameters
    "EleutherAI/gpt-neo-125M",         # ~125M parameters
    "sshleifer/tiny-gpt2",           # extremely small variant
    "microsoft/DialoGPT-small"       # DialoGPT small
]

# Define five languages: English, German, Spanish, French, Portuguese
languages = {
    "en": "English",
    "de": "German",
    "es": "Spanish",
    "fr": "French",
    "pt": "Portuguese"
}

# Define two cost-effective grammar evaluation models (unchanged)
grammar_model_names = [
    "vennify/t5-base-grammar-correction",
    "hassaanik/grammar-correction-model"
]

# Functions to load pipelines on demand
def load_generation_pipeline(model_name):
    try:
        # Using text-generation pipeline for causal LM models
        return pipeline("text-generation", model=model_name)
    except Exception as e:
        print(f"Error loading generation model {model_name}: {e}")
        return None

def load_grammar_pipeline(model_name):
    try:
        return pipeline("text2text-generation", model=model_name)
    except Exception as e:
        print(f"Error loading grammar model {model_name}: {e}")
        return None

# Pre-load grammar evaluator pipelines
rater_models = []
for model_name in grammar_model_names:
    p = load_grammar_pipeline(model_name)
    if p is not None:
        rater_models.append(p)

# Utility functions for checking palindromes and cleaning text
def clean_text(text):
    return re.sub(r'[^a-zA-Z0-9]', '', text.lower())

def is_palindrome(text):
    cleaned = clean_text(text)
    return cleaned == cleaned[::-1]

# Updated prompt that instructs the model to output ONLY the palindrome.
def build_prompt(lang):
    return (
        f"Instruction: Write the longest original palindrome you can in {lang}. "
        "The output should contain nothing else but the palindrome. "
        "Do not include any additional commentary or repeated instructions. "
        "Palindrome: "
    )

def grammar_prompt(pal, lang):
    return f'''Rate from 0 to 100 how grammatically correct this palindrome is in {lang}. Only return a number with no explanation:\n\n"{pal}"\n'''

def extract_score(text):
    match = re.search(r"\d{1,3}", text)
    if match:
        score = int(match.group())
        return min(max(score, 0), 100)
    return 0

# Main benchmark function that runs all tests at once
def run_benchmark_all():
    results = []
    # Iterate over each small model
    for model_name in small_models:
        gen_pipeline = load_generation_pipeline(model_name)
        if gen_pipeline is None:
            continue  # Skip this model if it fails to load

        # Iterate over each language
        for code, lang in languages.items():
            prompt = build_prompt(lang)
            try:
                # Generate text with a moderate max token limit
                gen_output = gen_pipeline(prompt, max_new_tokens=50, do_sample=True)[0]['generated_text'].strip()
            except Exception as e:
                gen_output = f"Error generating text: {e}"
            # Check if the generated output is a palindrome
            valid = is_palindrome(gen_output)
            cleaned_len = len(clean_text(gen_output))
            
            # Evaluate grammar using both grammar models
            scores = []
            for rater in rater_models:
                rprompt = grammar_prompt(gen_output, lang)
                try:
                    rtext = rater(rprompt, max_new_tokens=10)[0]['generated_text']
                    score = extract_score(rtext)
                    scores.append(score)
                except Exception as e:
                    scores.append(0)
            avg_score = np.mean(scores) if scores else 0
            # Penalize if the generated text is not a valid palindrome
            penalty = (avg_score / 100) if valid else (avg_score / 100) * 0.5
            final_score = round(cleaned_len * penalty, 2)
            
            results.append({
                "Model": model_name,
                "Language": lang,
                "Palindrome": gen_output,
                "Valid": "✅" if valid else "❌",
                "Length": cleaned_len,
                "Grammar Score": avg_score,
                "Final Score": final_score
            })
    
    df = pd.DataFrame(results).sort_values(by="Final Score", ascending=False).reset_index(drop=True)
    return gr.Dataframe(df)

# Build the Gradio UI using Blocks (canvas layout)
with gr.Blocks(title="Small Model Palindrome Benchmark") as demo:
    gr.Markdown("# Small Model Palindrome Benchmark")
    gr.Markdown("This benchmark automatically runs over 5 small text-generation models and 5 languages (English, German, Spanish, French, Portuguese).")
    
    with gr.Row():
        run_button = gr.Button("Run All Benchmarks")
    output_table = gr.Dataframe(label="Benchmark Results")
    
    run_button.click(fn=run_benchmark_all, inputs=[], outputs=output_table)

demo.launch()