Update app.py

app.py

@@ -15,10 +15,10 @@ llm = None
 llm_model = None
 hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.2-Q8_0-GGUF", filename="napoleon_24b_v0.2-q8_0.gguf",  local_dir = "./models")
 hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.1-Q8_0-GGUF", filename="napoleon_24b_v0.1-q8_0.gguf",  local_dir = "./models")
-hf_hub_download(repo_id="baconnier/Napoleon_24B_R1_v0.0-Q8_0-GGUF", filename="napoleon_24b_r1_v0.0-q8_0.gguf",  local_dir = "./models")
+#hf_hub_download(repo_id="baconnier/Napoleon_24B_R1_v0.0-Q8_0-GGUF", filename="napoleon_24b_r1_v0.0-q8_0.gguf",  local_dir = "./models")
 
 
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=60)
 def respond(
     message,
     history: list[tuple[str, str]],
@@ -83,10 +83,10 @@ demo = gr.ChatInterface(
     respond,
     additional_inputs=[
         gr.Dropdown([
-            'napoleon_24b_r1_v0.0-q8_0.gguf',
             'napoleon_24b_v0.2-q8_0.gguf',
             'napoleon_24b_v0.1-q8_0.gguf', 
-        ], value="napoleon_24b_r1_v0.0-q8_0.gguf", label="Model"),
+            'napoleon_24b_r1_v0.0-q8_0.gguf',            
+        ], value="napoleon_24b_v0.2-q8_0.gguf", label="Model"),
         gr.Slider(minimum=1, maximum=8192, value=8192, step=1, label="Max tokens"),
         gr.Slider(minimum=0.05, maximum=4.0, value=0.6, step=0.1, label="Temperature"),
         gr.Slider(