Update app.py
Browse files
    	
        app.py
    CHANGED
    
    | @@ -15,6 +15,8 @@ llm = None | |
| 15 | 
             
            llm_model = None
         | 
| 16 | 
             
            hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.2-Q8_0-GGUF", filename="napoleon_24b_v0.2-q8_0.gguf",  local_dir = "./models")
         | 
| 17 | 
             
            hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.1-Q8_0-GGUF", filename="napoleon_24b_v0.1-q8_0.gguf",  local_dir = "./models")
         | 
|  | |
|  | |
| 18 |  | 
| 19 | 
             
            @spaces.GPU(duration=120)
         | 
| 20 | 
             
            def respond(
         | 
| @@ -81,9 +83,10 @@ demo = gr.ChatInterface( | |
| 81 | 
             
                respond,
         | 
| 82 | 
             
                additional_inputs=[
         | 
| 83 | 
             
                    gr.Dropdown([
         | 
|  | |
| 84 | 
             
                        'napoleon_24b_v0.2-q8_0.gguf',
         | 
| 85 | 
             
                        'napoleon_24b_v0.1-q8_0.gguf'
         | 
| 86 | 
            -
                    ], value=" | 
| 87 | 
             
                    gr.Slider(minimum=1, maximum=8192, value=8192, step=1, label="Max tokens"),
         | 
| 88 | 
             
                    gr.Slider(minimum=0.05, maximum=4.0, value=0.6, step=0.1, label="Temperature"),
         | 
| 89 | 
             
                    gr.Slider(
         | 
|  | |
| 15 | 
             
            llm_model = None
         | 
| 16 | 
             
            hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.2-Q8_0-GGUF", filename="napoleon_24b_v0.2-q8_0.gguf",  local_dir = "./models")
         | 
| 17 | 
             
            hf_hub_download(repo_id="baconnier/Napoleon_24B_V0.1-Q8_0-GGUF", filename="napoleon_24b_v0.1-q8_0.gguf",  local_dir = "./models")
         | 
| 18 | 
            +
            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")
         | 
| 19 | 
            +
             | 
| 20 |  | 
| 21 | 
             
            @spaces.GPU(duration=120)
         | 
| 22 | 
             
            def respond(
         | 
|  | |
| 83 | 
             
                respond,
         | 
| 84 | 
             
                additional_inputs=[
         | 
| 85 | 
             
                    gr.Dropdown([
         | 
| 86 | 
            +
                        'napoleon_24b_r1_v0.0-q8_0.gguf'
         | 
| 87 | 
             
                        'napoleon_24b_v0.2-q8_0.gguf',
         | 
| 88 | 
             
                        'napoleon_24b_v0.1-q8_0.gguf'
         | 
| 89 | 
            +
                    ], value="napoleon_24b_r1_v0.0-q8_0.gguf", label="Model"),
         | 
| 90 | 
             
                    gr.Slider(minimum=1, maximum=8192, value=8192, step=1, label="Max tokens"),
         | 
| 91 | 
             
                    gr.Slider(minimum=0.05, maximum=4.0, value=0.6, step=0.1, label="Temperature"),
         | 
| 92 | 
             
                    gr.Slider(
         | 
 
			
