Update app.py

app.py

@@ -1,12 +1,26 @@
+import os
+import openai
 import torch
 import tensorflow as tf
-from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering, AutoModelForCausalLM
+from transformers import AutoTokenizer, TFAutoModelForQuestionAnswering
 import gradio as gr
 import re
 
+# Set your OpenAI API key here temporarily for testing
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
 # Check if GPU is available and use it if possible
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
+# Load the English models and tokenizers
+qa_model_name_v1 = 'salsarra/ConfliBERT-QA'
+qa_model_v1 = TFAutoModelForQuestionAnswering.from_pretrained(qa_model_name_v1)
+qa_tokenizer_v1 = AutoTokenizer.from_pretrained(qa_model_name_v1)
+
+bert_model_name_v1 = 'salsarra/BERT-base-cased-SQuAD-v1'
+bert_qa_model_v1 = TFAutoModelForQuestionAnswering.from_pretrained(bert_model_name_v1)
+bert_qa_tokenizer_v1 = AutoTokenizer.from_pretrained(bert_model_name_v1)
+
 # Load Spanish models and tokenizers
 confli_model_spanish = 'salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA'
 confli_model_spanish_qa = TFAutoModelForQuestionAnswering.from_pretrained(confli_model_spanish)
@@ -16,6 +30,7 @@ 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)
 
+# Load the newly added models for Spanish (Beto and ConfliBERT SQAC)
 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)
@@ -24,175 +39,89 @@ 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?"
-
-    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
+# Define error handling to separate input size errors from other issues
 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 \((\d+)\)")
+    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; text-align: center;'>Respuestas:</h2>
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>ConfliBERT-Spanish-Beto-Cased-NewsQA:</strong><br>{confli_answer_spanish}</div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>Beto-Spanish-Cased-NewsQA:</strong><br>{beto_answer_spanish}
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>ConfliBERT-Spanish-Beto-Cased-SQAC:</strong><br>{confli_sqac_answer_spanish}
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>Beto-Spanish-Cased-SQAC:</strong><br>{beto_sqac_answer_spanish}
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>GPT-2-Small-Spanish:</strong><br>{gpt2_answer_spanish}
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>BLOOM-1.7B:</strong><br>{bloom_answer}
-    </div><br>
-    <div style='color: #2e8b57; font-weight: bold;'>
-        <strong>Información del modelo:</strong><br>
-        <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-NewsQA' target='_blank'>ConfliBERT-Spanish-Beto-Cased-NewsQA</a><br>
-        <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-NewsQA' target='_blank'>Beto-Spanish-Cased-NewsQA</a><br>
-        <a href='https://huggingface.co/salsarra/ConfliBERT-Spanish-Beto-Cased-SQAC' target='_blank'>ConfliBERT-Spanish-Beto-Cased-SQAC</a><br>
-        <a href='https://huggingface.co/salsarra/Beto-Spanish-Cased-SQAC' target='_blank'>Beto-Spanish-Cased-SQAC</a><br>
-        <a href='https://huggingface.co/datificate/gpt2-small-spanish' target='_blank'>GPT-2-Small-Spanish</a><br>
-        <a href='https://huggingface.co/bigscience/bloom-1b7' target='_blank'>BLOOM-1.7B</a><br>
-    </div>
-    """
+    pattern_qa = re.compile(r"indices\\[0,(\\d+)\\] = \\d+ is not in \\[0, (\\d+)\\)")
+    match_qa = pattern_qa.search(error_message)
+    if match_qa:
+        number_1, number_2 = match_qa.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>"
 
-# CSS for Gradio interface
-css_styles = """
+    return f"<span style='color: red; font-weight: bold;'>Error: {error_message}</span>"
+
+# Main comparison function with language selection
+def compare_question_answering(language, context, question):
+    if language == "English":
+        confli_answer_v1 = question_answering_v1(context, question)
+        bert_answer_v1 = bert_question_answering_v1(context, question)
+        chatgpt_answer = chatgpt_question_answering(context, question)
+        return f"""
+        <div>
+            <h2 style='color: #2e8b57; font-weight: bold;'>Answers:</h2>
+        </div><br>
+        <div>
+            <strong>ConfliBERT-cont-cased-SQuAD-v1:</strong><br>{confli_answer_v1}</div><br>
+        <div>
+            <strong>BERT-base-cased-SQuAD-v1:</strong><br>{bert_answer_v1}
+        </div><br>
+        <div>
+            <strong>ChatGPT:</strong><br>{chatgpt_answer}
+        </div><br>
+        <div>
+            <strong>Model Information:</strong><br>
+            ConfliBERT-cont-cased-SQuAD-v1: <a href='https://huggingface.co/salsarra/ConfliBERT-QA' target='_blank'>salsarra/ConfliBERT-QA</a><br>
+            BERT-base-cased-SQuAD-v1: <a href='https://huggingface.co/salsarra/BERT-base-cased-SQuAD-v1' target='_blank'>salsarra/BERT-base-cased-SQuAD-v1</a><br>
+            ChatGPT (GPT-3.5 Turbo): <a href='https://platform.openai.com/docs/models/gpt-3-5' target='_blank'>OpenAI API</a><br>
+        </div>
+        """
+    elif language == "Spanish":
+        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)
+        chatgpt_answer_spanish = chatgpt_question_answering_spanish(context, question)
+        return f"""
+        <div>
+            <h2 style='color: #2e8b57; font-weight: bold;'>Answers:</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>ChatGPT:</strong><br>{chatgpt_answer_spanish}
+        </div><br>
+        <div>
+            <strong>Model Information:</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>
+            ChatGPT (GPT-3.5 Turbo): <a href='https://platform.openai.com/docs/models/gpt-3-5' target='_blank'>OpenAI API</a><br>
+        </div>
+        """
+
+# Setting up Gradio Blocks interface with footer
+with gr.Blocks(css="""
     body {
         background-color: #f0f8ff;
         font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;
     }
-    h1 a {
+    h1, h1 a {
         color: #2e8b57;
         text-align: center;
         font-size: 2em;
@@ -202,54 +131,44 @@ css_styles = """
         color: #ff8c00;
     }
     h2 {
-        color: #2e8b57;
+        color: #ff8c00;
         text-align: center;
         font-size: 1.5em;
     }
-    .description {
-        color: #2e8b57;
-        display: block;
-        font-size: 1.2em;
-        text-align: center;
-    }
-    .footer {
-        text-align: center;
-        margin-top: 10px;
-        font-size: 0.9em;
-        width: 100%;
-    }
-    .footer a {
-        font-weight: bold;
-        text-decoration: none;
+    .gradio-container {
+        max-width: 100%;
+        margin: 10px auto;
+        padding: 10px;
+        background-color: #ffffff;
+        border-radius: 10px;
+        box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
     }
-    .footer a:hover {
-        text-decoration: underline;
+    .button-row {
+        display: flex;
+        justify-content: center;
+        gap: 10px;
     }
-"""
-
-# 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' style='color: #2e8b57; text-align: center;'>ConfliBERT-Spanish-QA</a>",
-    description="<span class='description'>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 green "Developed By" text
-    article="""
-    <div class='footer'>
-        <a href='https://eventdata.utdallas.edu/'>UTD Event Data</a> | 
-        <a href='https://www.utdallas.edu/'>University of Texas at Dallas</a>
-    </div>
-    <div class='footer' style='color: #2e8b57; font-weight: bold;'>
-        Developed By: <a href='https://www.linkedin.com/in/sultan-alsarra-phd-56977a63/' target='_blank'>Sultan Alsarra</a>
-    </div>
-    """
-)
+""") as demo:
+    
+    gr.Markdown("# [ConfliBERT-QA](https://eventdata.utdallas.edu/conflibert/)", elem_id="title")
+    gr.Markdown("Compare answers between ConfliBERT, BERT, and ChatGPT for English, and ConfliBERT, BETO, ConfliBERT-SQAC, Beto-SQAC, and ChatGPT for Spanish.")
+    
+    language = gr.Dropdown(choices=["English", "Spanish"], label="Select Language")
+    context = gr.Textbox(lines=5, placeholder="Enter the context here...", label="Context")
+    question = gr.Textbox(lines=2, placeholder="Enter your question here...", label="Question")
+    output = gr.HTML(label="Output")
+    
+    with gr.Row(elem_id="button-row"):
+        clear_btn = gr.Button("Clear")
+        submit_btn = gr.Button("Submit")
+    
+    submit_btn.click(fn=compare_question_answering, inputs=[language, context, question], outputs=output)
+    clear_btn.click(fn=lambda: ("", "", "", ""), inputs=[], outputs=[language, context, question, output])
+    
+    gr.Markdown("""
+        <div style="text-align: center; margin-top: 20px;">
+            Built by: <a href="https://www.linkedin.com/in/sultan-alsarra-phd-56977a63/" target="_blank">Sultan Alsarra</a>
+        </div>
+    """)
 
-# Launch the Gradio demo
 demo.launch(share=True)