Create app.py

app.py

@@ -0,0 +1,275 @@
+import torch
+import tensorflow as tf
+from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering, AutoModelForCausalLM
+import gradio as gr
+import re
+
+# Check if GPU is available and use it if possible
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+# Load Spanish models and tokenizers
+confli_model_spanish = 'salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA'
+confli_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(confli_model_spanish)
+confli_tokenizer_spanish = AutoTokenizer.from_pretrained(confli_model_spanish)
+
+beto_model_spanish = 'salsarra/Beto-Spanish-Cased-NewsQA'
+beto_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(beto_model_spanish)
+beto_tokenizer_spanish = AutoTokenizer.from_pretrained(beto_model_spanish)
+
+confli_sqac_model_spanish = 'salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC'
+confli_sqac_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(confli_sqac_model_spanish)
+confli_sqac_tokenizer_spanish = AutoTokenizer.from_pretrained(confli_sqac_model_spanish)
+
+beto_sqac_model_spanish = 'salsarra/Beto-Spanish-Cased-SQAC'
+beto_sqac_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(beto_sqac_model_spanish)
+beto_sqac_tokenizer_spanish = AutoTokenizer.from_pretrained(beto_sqac_model_spanish)
+
+# Load Spanish GPT-2 model and tokenizer
+gpt2_spanish_model_name = 'datificate/gpt2-small-spanish'
+gpt2_spanish_tokenizer = AutoTokenizer.from_pretrained(gpt2_spanish_model_name)
+gpt2_spanish_model = AutoModelForCausalLM.from_pretrained(gpt2_spanish_model_name).to(device)
+
+# Load BLOOM-1.7B model and tokenizer for Spanish
+bloom_model_name = 'bigscience/bloom-1b7'
+bloom_tokenizer = AutoTokenizer.from_pretrained(bloom_model_name)
+bloom_model = AutoModelForCausalLM.from_pretrained(bloom_model_name).to(device)
+
+# Preload models with a dummy pass to improve first-time loading
+def preload_models():
+    dummy_context = "Este es un contexto de prueba."
+    dummy_question = "¿Cuál es el propósito de este contexto?"
+
+    # Run each model with a dummy input to initialize them
+    inputs = confli_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
+    _ = confli_model_spanish_qa(inputs)
+    
+    inputs = beto_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
+    _ = beto_model_spanish_qa(inputs)
+
+    inputs = confli_sqac_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
+    _ = confli_sqac_model_spanish_qa(inputs)
+    
+    inputs = beto_sqac_tokenizer_spanish(dummy_question, dummy_context, return_tensors='tf')
+    _ = beto_sqac_model_spanish_qa(inputs)
+
+preload_models()  # Initialize models
+
+# Error handling function
+def handle_error_message(e, default_limit=512):
+    error_message = str(e)
+    pattern = re.compile(r"The size of tensor a \((\d+)\) must match the size of tensor b \((\d+)\)")
+    match = pattern.search(error_message)
+    if match:
+        number_1, number_2 = match.groups()
+        return f"<span style='color: red; font-weight: bold;'>Error: Text Input is over limit where inserted text size {number_1} is larger than model limits of {number_2}</span>"
+    return f"<span style='color: red; font-weight: bold;'>Error: Text Input is over limit where inserted text size is larger than model limits of {default_limit}</span>"
+
+# Spanish QA functions
+def question_answering_spanish(context, question):
+    try:
+        inputs = confli_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
+        outputs = confli_model_spanish_qa(inputs)
+        answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
+        answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
+        answer = confli_tokenizer_spanish.convert_tokens_to_string(confli_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
+        return f"<span style='color: green; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+def beto_question_answering_spanish(context, question):
+    try:
+        inputs = beto_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
+        outputs = beto_model_spanish_qa(inputs)
+        answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
+        answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
+        answer = beto_tokenizer_spanish.convert_tokens_to_string(beto_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
+        return f"<span style='color: blue; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+def confli_sqac_question_answering_spanish(context, question):
+    try:
+        inputs = confli_sqac_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
+        outputs = confli_sqac_model_spanish_qa(inputs)
+        answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
+        answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
+        answer = confli_sqac_tokenizer_spanish.convert_tokens_to_string(confli_sqac_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
+        return f"<span style='color: teal; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+def beto_sqac_question_answering_spanish(context, question):
+    try:
+        inputs = beto_sqac_tokenizer_spanish(question, context, return_tensors='tf', truncation=True)
+        outputs = beto_sqac_model_spanish_qa(inputs)
+        answer_start = tf.argmax(outputs.start_logits, axis=1).numpy()[0]
+        answer_end = tf.argmax(outputs.end_logits, axis=1).numpy()[0] + 1
+        answer = beto_sqac_tokenizer_spanish.convert_tokens_to_string(beto_sqac_tokenizer_spanish.convert_ids_to_tokens(inputs['input_ids'].numpy()[0][answer_start:answer_end]))
+        return f"<span style='color: brown; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+def gpt2_spanish_question_answering(context, question):
+    try:
+        prompt = f"Contexto:\n{context}\n\nPregunta:\n{question}\n\nRespuesta:"
+        inputs = gpt2_spanish_tokenizer(prompt, return_tensors='pt').to(device)
+        outputs = gpt2_spanish_model.generate(
+            inputs['input_ids'], 
+            max_length=inputs['input_ids'].shape[1] + 50, 
+            num_return_sequences=1, 
+            pad_token_id=gpt2_spanish_tokenizer.eos_token_id,
+            do_sample=True,
+            top_k=40,
+            temperature=0.8
+        )
+        answer = gpt2_spanish_tokenizer.decode(outputs[0], skip_special_tokens=True)
+        answer = answer.split("Respuesta:")[-1].strip()
+        return f"<span style='color: orange; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+def bloom_question_answering(context, question):
+    try:
+        prompt = f"Contexto:\n{context}\n\nPregunta:\n{question}\n\nRespuesta:"
+        inputs = bloom_tokenizer(prompt, return_tensors='pt').to(device)
+        outputs = bloom_model.generate(
+            inputs['input_ids'], 
+            max_length=inputs['input_ids'].shape[1] + 50, 
+            num_return_sequences=1, 
+            pad_token_id=bloom_tokenizer.eos_token_id,
+            do_sample=True,
+            top_k=40,
+            temperature=0.8
+        )
+        answer = bloom_tokenizer.decode(outputs[0], skip_special_tokens=True)
+        answer = answer.split("Respuesta:")[-1].strip()
+        return f"<span style='color: purple; font-weight: bold;'>{answer}</span>"
+    except Exception as e:
+        return handle_error_message(e)
+
+# Main function for Spanish QA
+def compare_question_answering_spanish(context, question):
+    confli_answer_spanish = question_answering_spanish(context, question)
+    beto_answer_spanish = beto_question_answering_spanish(context, question)
+    confli_sqac_answer_spanish = confli_sqac_question_answering_spanish(context, question)
+    beto_sqac_answer_spanish = beto_sqac_question_answering_spanish(context, question)
+    gpt2_answer_spanish = gpt2_spanish_question_answering(context, question)
+    bloom_answer = bloom_question_answering(context, question)
+    return f"""
+    <div>
+        <h2 style='color: #2e8b57; font-weight: bold;'>Respuestas:</h2>
+    </div><br>
+    <div>
+        <strong>ConfliBERT-Spanish-Beto-Cased-NewsQA:</strong><br>{confli_answer_spanish}</div><br>
+    <div>
+        <strong>Beto-Spanish-Cased-NewsQA:</strong><br>{beto_answer_spanish}
+    </div><br>
+    <div>
+        <strong>ConfliBERT-Spanish-Beto-Cased-SQAC:</strong><br>{confli_sqac_answer_spanish}
+    </div><br>
+    <div>
+        <strong>Beto-Spanish-Cased-SQAC:</strong><br>{beto_sqac_answer_spanish}
+    </div><br>
+    <div>
+        <strong>GPT-2-Small-Spanish:</strong><br>{gpt2_answer_spanish}
+    </div><br>
+    <div>
+        <strong>BLOOM-1.7B:</strong><br>{bloom_answer}
+    </div><br>
+    <div>
+        <strong>Información del modelo:</strong><br>
+        ConfliBERT-Spanish-Beto-Cased-NewsQA: <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA' target='_blank'>salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA</a><br>
+        Beto-Spanish-Cased-NewsQA: <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-NewsQA' target='_blank'>salsarra/Beto-Spanish-Cased-NewsQA</a><br>
+        ConfliBERT-Spanish-Beto-Cased-SQAC: <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC' target='_blank'>salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC</a><br>
+        Beto-Spanish-Cased-SQAC: <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-SQAC' target='_blank'>salsarra/Beto-Spanish-Cased-SQAC</a><br>
+        GPT-2-Small-Spanish: <a href='https://huggingface.co/datificate/gpt2-small-spanish' target='_blank'>datificate GPT-2 Small Spanish</a><br>
+        BLOOM-1.7B: <a href='https://huggingface.co/bigscience/bloom-1b7' target='_blank'>bigscience BLOOM-1.7B</a><br>
+    </div>
+    """
+
+# CSS for Gradio interface
+css_styles = """
+    body {
+        background-color: #f0f8ff;
+        font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;
+    }
+    h1 a {
+        color: #2e8b57;
+        text-align: center;
+        font-size: 2em;
+        text-decoration: none;
+    }
+    h1 a:hover {
+        color: #ff8c00;
+    }
+    h2 {
+        color: #ff8c00;
+        text-align: center;
+        font-size: 1.5em;
+    }
+    .description-light {
+        color: black;
+        display: block;
+        font-size: 1em;
+        text-align: center;
+    }
+    .description-dark {
+        color: white;
+        display: none;
+        font-size: 1em;
+        text-align: center;
+    }
+    @media (prefers-color-scheme: dark) {
+        .description-light {
+            display: none;
+        }
+        .description-dark {
+            display: block;
+        }
+    }
+    .footer {
+        text-align: center;
+        margin-top: 10px;
+        font-size: 0.9em;
+        color: #666;
+        width: 100%;
+    }
+    .footer a {
+        color: #2e8b57;
+        font-weight: bold;
+        text-decoration: none;
+    }
+    .footer a:hover {
+        text-decoration: underline;
+    }
+"""
+
+# Define the Gradio interface with footer directly in the layout
+demo = gr.Interface(
+    fn=compare_question_answering_spanish,
+    inputs=[
+        gr.Textbox(lines=5, placeholder="Ingrese el contexto aquí...", label="Contexto"),
+        gr.Textbox(lines=2, placeholder="Ingrese su pregunta aquí...", label="Pregunta")
+    ],
+    outputs=gr.HTML(label="Salida"),
+    title="<a href='https://eventdata.utdallas.edu/conflibert/' target='_blank'>ConfliBERT-Spanish-QA</a>",
+    description="""
+    <span class="description-light">Compare respuestas entre los modelos ConfliBERT, BETO, ConfliBERT SQAC, Beto SQAC, GPT-2 Small Spanish y BLOOM-1.7B para preguntas en español.</span>
+    <span class="description-dark">Compare respuestas entre los modelos ConfliBERT, BETO, ConfliBERT SQAC, Beto SQAC, GPT-2 Small Spanish y BLOOM-1.7B para preguntas en español.</span>
+    """,
+    css=css_styles,
+    allow_flagging="never",
+    # Footer HTML with centered, green links
+    article="""
+    <div class='footer'>
+        <a href='https://eventdata.utdallas.edu/' style='color: #2e8b57; font-weight: bold;'>UTD Event Data</a> | 
+        <a href='https://www.utdallas.edu/' style='color: #2e8b57; font-weight: bold;'>University of Texas at Dallas</a>
+    </div>
+    <div class='footer'>
+        Developed By: <a href='https://www.linkedin.com/in/sultan-alsarra-phd-56977a63/' target='_blank' style='color: #2e8b57; font-weight: bold;'>Sultan Alsarra</a>
+    </div>
+    """
+)
+
+# Launch the Gradio demo
+demo.launch(share=True)