Update app.py

app.py

@@ -1,37 +1,40 @@
 import gradio as gr
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
-# Custom CSS for pastel gradient
+checkpoint = "mistralai/Mistral-7B-Instruct-v0.3"
+
+# Download tokenizer & model
+tokenizer = AutoTokenizer.from_pretrained(checkpoint, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    checkpoint,
+    device_map="auto",             # or "cpu" / "cuda"
+    trust_remote_code=True
+)
+# (Optional) set model to inference mode, etc.
+# model.eval()
+
+def inference_fn(prompt):
+    # Tokenize
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    # Generate
+    output_tokens = model.generate(**inputs, max_new_tokens=128)
+    # Decode
+    return tokenizer.decode(output_tokens[0], skip_special_tokens=True)
+
+# Pastel gradient CSS
 css = """
 .gradio-container {
     background: linear-gradient(to right, #FFDEE9, #B5FFFC);
 }
 """
 
-# Load the Mistral-7B-Instruct-v0.3 model via Gradio's load function
-model = gr.load("models/mistralai/Mistral-7B-Instruct-v0.3")
-
-def inference_fn(prompt):
-    """
-    This function calls the loaded model with the user's prompt.
-    gr.load(...) returns a Gradio interface object, so we can call it like a function.
-    """
-    # If the loaded model is a pipeline or interface, calling it directly returns the response.
-    response = model(prompt)
-    return response
-
 with gr.Blocks(css=css) as demo:
-    # Greeting at the top
     gr.Markdown("<h1 style='text-align: center;'>Bonjour Dans le chat du consentement</h1>")
-
-    # Create the input/output layout
-    with gr.Row():
-        user_input = gr.Textbox(label="Entrez votre message ici:", lines=3)
-        output = gr.Textbox(label="Réponse du Modèle Mistral-7B-Instruct:", lines=5)
+    user_input = gr.Textbox(label="Entrez votre message ici:", lines=3)
+    output = gr.Textbox(label="Réponse du Modèle:", lines=5)
     send_button = gr.Button("Envoyer")
 
-    # Link the button to inference_fn
     send_button.click(fn=inference_fn, inputs=user_input, outputs=output)
 
-# Launch the app
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()