Update app.py

app.py

@@ -1,56 +1,60 @@
 import gradio as gr
 from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 
-# Charger le modèle et le tokenizer
-tokenizer = AutoTokenizer.from_pretrained("NousResearch/Hermes-2-Pro-Llama-3-8B")
-model = AutoModelForCausalLM.from_pretrained("NousResearch/Hermes-2-Pro-Llama-3-8B")
+# Load the model and tokenizer
+tokenizer = AutoTokenizer.from_pretrained("unsloth/Llama-3.2-1B")
+model = AutoModelForCausalLM.from_pretrained("unsloth/Llama-3.2-1B")
 
-# Créer un pipeline pour la génération de texte
+# Use a pipeline for text generation
 text_gen_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-# Fonction pour générer du texte
-def generate_text(prompt, max_length=100, temperature=0.7, top_p=0.9, top_k=50, repetition_penalty=1.2):
+# Text generation function with stronger penalties and no_repeat_ngram_size
+def generate_text(prompt, max_length=40, temperature=0.8, top_p=0.9, top_k=40, repetition_penalty=1.5, no_repeat_ngram_size=4):
     generated_text = text_gen_pipeline(prompt, 
                                        max_length=max_length, 
                                        temperature=temperature, 
                                        top_p=top_p, 
                                        top_k=top_k, 
-                                       repetition_penalty=repetition_penalty,  # Pénalisation de répétition pour éviter les répétitions
+                                       repetition_penalty=repetition_penalty,  # Stronger repetition penalty
+                                       no_repeat_ngram_size=no_repeat_ngram_size,  # Higher n-gram size to reduce repetition
                                        num_return_sequences=1)
     return generated_text[0]['generated_text']
 
-# Interface Gradio
+# Gradio Interface
 with gr.Blocks() as demo:
-    gr.Markdown("## Text Generation with Hermes 2 - Llama 3 8B")
-
-    # Zone de saisie pour le prompt utilisateur
+    gr.Markdown("## Text Generation with Llama 3.2 - 1B")
+    
+    # Input box for user prompt
     prompt_input = gr.Textbox(label="Input (Prompt)", placeholder="Enter your prompt here...")
-
-    # Curseur pour la longueur maximale du texte
-    max_length_input = gr.Slider(minimum=10, maximum=200, value=100, step=10, label="Maximum Length")
-
-    # Curseur pour la température (contrôle de la créativité)
-    temperature_input = gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature (creativity)")
-
-    # Curseur pour top_p (échantillonnage par noyau)
+    
+    # Slider for maximum text length
+    max_length_input = gr.Slider(minimum=10, maximum=200, value=40, step=10, label="Maximum Length")
+    
+    # Slider for temperature (controls creativity)
+    temperature_input = gr.Slider(minimum=0.1, maximum=1.0, value=0.8, step=0.1, label="Temperature (creativity)")
+    
+    # Slider for top_p (nucleus sampling)
     top_p_input = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.1, label="Top-p (nucleus sampling)")
-
-    # Curseur pour top_k (contrôle de la diversité)
-    top_k_input = gr.Slider(minimum=1, maximum=100, value=50, step=1, label="Top-k (sampling diversity)")
-
-    # Curseur pour la pénalisation de répétition
-    repetition_penalty_input = gr.Slider(minimum=1.0, maximum=2.0, value=1.2, step=0.1, label="Repetition Penalty")
-
-    # Zone de sortie pour le texte généré
+    
+    # Slider for top_k (controls diversity)
+    top_k_input = gr.Slider(minimum=1, maximum=100, value=40, step=1, label="Top-k (sampling diversity)")
+    
+    # Slider for repetition penalty
+    repetition_penalty_input = gr.Slider(minimum=1.0, maximum=2.0, value=1.5, step=0.1, label="Repetition Penalty")
+    
+    # Slider for no_repeat_ngram_size
+    no_repeat_ngram_size_input = gr.Slider(minimum=1, maximum=10, value=4, step=1, label="No Repeat N-Gram Size")
+    
+    # Output box for the generated text
     output_text = gr.Textbox(label="Generated Text")
-
-    # Bouton pour générer du texte
+    
+    # Submit button
     generate_button = gr.Button("Generate")
-
-    # Action au clic sur le bouton
+    
+    # Action on button click
     generate_button.click(generate_text, 
-                          inputs=[prompt_input, max_length_input, temperature_input, top_p_input, top_k_input, repetition_penalty_input], 
+                          inputs=[prompt_input, max_length_input, temperature_input, top_p_input, top_k_input, repetition_penalty_input, no_repeat_ngram_size_input], 
                           outputs=output_text)
 
-# Lancer l'application
+# Launch the app
 demo.launch()