Upload app.py

app.py

@@ -0,0 +1,112 @@
+import gradio as gr
+from transformers import pipeline, BertTokenizer, BertForSequenceClassification
+
+# Charger le modèle zéro-shot de Hugging Face
+zero_shot_classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+
+# Charger le modèle few-shot à partir du modèle sauvegardé
+tokenizer = BertTokenizer.from_pretrained('./animal_offense_model')
+model = BertForSequenceClassification.from_pretrained('./animal_offense_model')
+few_shot_classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)
+
+# Fonction pour classifier le texte avec le modèle zéro-shot
+def classify_with_zero_shot(input_text):
+    candidate_labels = [
+        "non-offensive, it's cute! 😇", 
+        "very slightly offensive, but not a big deal! 😅", 
+        "slightly offensive, just a little! 🤏", 
+        "a bit offensive, ouch! 🤭", 
+        "moderately offensive, getting there! 😬", 
+        "fairly offensive, watch out! 🚨", 
+        "offensive, that's a no-no! 🚫", 
+        "very offensive, you really shouldn't say that! 😳", 
+        "extremely offensive, seriously? 😡", 
+        "totally unacceptable and offensive, you are crazy! 🤯"
+    ]
+
+    result = zero_shot_classifier(input_text, candidate_labels)
+    labels_scores = dict(zip(result["labels"], result["scores"]))
+    return labels_scores
+
+# Fonction pour classifier le texte avec votre propre modèle few-shot
+def classify_with_few_shot(input_text):
+    result = few_shot_classifier(input_text)
+    label = result[0]["label"]
+    score = result[0]["score"]
+    
+    # Ajuster la logique pour garantir l'interprétation correcte des labels
+    if label == "LABEL_0":
+        return {"non-offensive, it's cute! 😇": score, "offensive": 1 - score}
+    elif label == "LABEL_1":
+        return {"offensive": score, "non-offensive, it's cute! 😇": 1 - score}
+    else:
+        return {"unknown": 1.0}  # Pour des fins de débogage si un label inattendu est trouvé
+
+# Fonction principale pour sélectionner le modèle
+def classify_text(input_text, model_choice):
+    if model_choice == "Zero-Shot Model":
+        return classify_with_zero_shot(input_text)
+    elif model_choice == "Few-Shot Model":
+        return classify_with_few_shot(input_text)
+    else:
+        return "Please select a valid model."
+
+# Liste de phrases exemples (chaque sous-liste est [texte])
+example_phrases = [
+    ["Your dog is the cutest ever!"],
+    ["I think your cat needs to lose some weight."],
+    ["Why would anyone like such an ugly fish?"],
+    ["Oh no, saying that about a rabbit is not okay at all!"],
+    ["That’s a bit harsh on a parrot."],
+    ["You should be more gentle when talking about horses."],
+    ["This kitten is just too adorable!"],
+    ["Wow, calling a bird annoying is really offensive!"],
+    ["That’s a lovely compliment for a hamster!"],
+    ["Saying that a dog smells bad is quite rude!"]
+]
+
+# Créer une interface Gradio Blocks pour plus de flexibilité
+with gr.Blocks() as iface:
+    gr.Markdown("# Animal Offense Detector")
+
+    with gr.Column():
+        gr.Markdown("## Enter Your Text Below:")
+        text_input = gr.Textbox(lines=5, placeholder="Enter your text here...")
+        model_choice = gr.Radio(choices=["Zero-Shot Model", "Few-Shot Model"], label="Choose Model")
+        label_output = gr.Label(label="Labels and Scores")
+
+        gr.Interface(fn=classify_text, inputs=[text_input, model_choice], outputs=label_output)
+        
+        gr.Examples(examples=example_phrases, inputs=[text_input])
+
+    # Ajouter de l'espacement et une taille de police plus grande pour la documentation
+    gr.Markdown("""
+    <div style="margin-top: 40px; font-size: 18px;">
+    
+    #### Documentation for `Animal Offense Detector`
+
+    This script classifies text to determine the level of offense towards animals using two natural language processing models from Hugging Face. Users can choose between a zero-shot model and a few-shot model to evaluate the input text.
+
+    #### Libraries Used
+    - **gradio**: Used to create web-based user interfaces for Python functions.
+    - **transformers**: Provides machine learning models for natural language processing tasks.
+
+    #### Features
+    1. **Zero-Shot Model Classification**: 
+       - Uses `facebook/bart-large-mnli` to classify text based on several predefined labels.
+       - This model can understand and classify text without needing specific training for each task.
+    
+    2. **Few-Shot Model Classification**:
+       - Uses a custom-trained BERT model to evaluate text as "non-offensive" or "offensive".
+       - This model provides a quick and accurate classification based on the training data.
+
+    3. **Model Selection**:
+       - The interface allows the user to choose between the zero-shot and few-shot models to classify the text.
+
+    4. **Example Phrases**:
+       - Provides example phrases that the user can select to test the models. Each example is designed to test different levels of potential offense.
+    </div>
+    """)
+
+if __name__ == "__main__":
+    iface.launch(share=True)