Update app.py

app.py

@@ -8,7 +8,7 @@ from threading import Thread
 
 
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
-MODEL = "freeCS-dot-org/Art-exp0-7B"
+MODEL = "AGI-0/Art-v0-3B"
 
 TITLE = """<h2>Link to the model: <a href="https://huggingface.co/AGI-0/Artificium-llama3.1-8B-001" title="Visit the model repository on Hugging Face">AGI-0/Artificium-llama3.1-8B-001</a> please leave a like to the repository if you liked it. THIS INFERENCE IS 4-Bit Quantized</h2>"""
 
@@ -110,7 +110,7 @@ with gr.Blocks(css=CSS, theme="soft") as demo:
                 minimum=0,
                 maximum=1,
                 step=0.1,
-                value=0.3,
+                value=0.2,
                 label="Temperature",
                 render=False,
             ),
@@ -118,7 +118,7 @@ with gr.Blocks(css=CSS, theme="soft") as demo:
                 minimum=128,
                 maximum=8192,
                 step=1,
-                value=2048,
+                value=4096,
                 label="Max new tokens",
                 render=False,
             ),