Update app.py

app.py

@@ -8,7 +8,7 @@ from threading import Thread
 
 
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
-MODEL_ID = "ehristoforu/RQwen-v0.1"
+MODEL_ID = "ehristoforu/RQwen-v0.2"
 MODELS = os.environ.get("MODELS")
 MODEL_NAME = MODELS.split("/")[-1]
 
@@ -97,7 +97,7 @@ with gr.Blocks(css=CSS) as demo:
                 minimum=0,
                 maximum=1,
                 step=0.1,
-                value=0.8,
+                value=0.1,
                 label="Temperature",
                 render=False,
             ),
@@ -105,7 +105,7 @@ with gr.Blocks(css=CSS) as demo:
                 minimum=128,
                 maximum=4096,
                 step=1,
-                value=1024,
+                value=4096,
                 label="Max new tokens",
                 render=False,
             ),
@@ -113,7 +113,7 @@ with gr.Blocks(css=CSS) as demo:
                 minimum=0.0,
                 maximum=1.0,
                 step=0.1,
-                value=0.8,
+                value=1,
                 label="top_p",
                 render=False,
             ),
@@ -129,7 +129,7 @@ with gr.Blocks(css=CSS) as demo:
                 minimum=0.0,
                 maximum=2.0,
                 step=0.1,
-                value=1.0,
+                value=1.2,
                 label="Repetition penalty",
                 render=False,
             ),