Update app.py

app.py

@@ -21,6 +21,18 @@ def transcribe(audio):
 
 HF_TOKEN = os.environ.get("HF_TOKEN", None)
 
+def client_fn(model):
+    if "Mixtral" in model:
+        return InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
+    elif "Llama" in model:
+        return InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
+    elif "Mistral" in model:
+        return InferenceClient("mistralai/Mistral-7B-Instruct-v0.2")
+    elif "Phi" in model:
+        return InferenceClient("microsoft/Phi-3-mini-4k-instruct")
+    else: 
+        return InferenceClient("microsoft/Phi-3-mini-4k-instruct")
+
 def randomize_seed_fn(seed: int) -> int:
     seed = random.randint(0, 999999)
     return seed
@@ -33,18 +45,17 @@ Respond in a normal, conversational manner while being friendly and helpful.
 [USER]
 """
 
-def models(text, seed=42):
+def models(text, model="Mixtral 8x7B", seed=42):
 
     seed = int(randomize_seed_fn(seed))
     generator = torch.Generator().manual_seed(seed)  
     
-    client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
+    client = client_fn(model)
     
     generate_kwargs = dict(
         max_new_tokens=300,
         seed=seed
-    )
-    
+    )    
     formatted_prompt = system_instructions1 + text + "[JARVIS]"
     stream = client.text_generation(
         formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
@@ -52,7 +63,6 @@ def models(text, seed=42):
     for response in stream:
         if not response.token.text == "</s>":
             output += response.token.text
-
     return output
 
 async def respond(audio, model, seed):
@@ -72,6 +82,14 @@ DESCRIPTION = """ # <center><b>JARVIS⚡</b></center>
 with gr.Blocks(css="style.css") as demo:    
     gr.Markdown(DESCRIPTION)
     with gr.Row():
+        select = gr.Dropdown([ 'Mixtral 8x7B',
+        'Llama 3 8B',
+        'Mistral 7B v0.3',
+        'Phi 3 mini',
+    ],
+    value="Mistral 7B v0.3",
+    label="Model"
+    )
         seed = gr.Slider(
         label="Seed",
         minimum=0,
@@ -89,8 +107,8 @@ with gr.Blocks(css="style.css") as demo:
             batch=True,
             max_batch_size=10, 
             fn=respond, 
-            inputs=[input, seed],
+            inputs=[input, select, seed],
             outputs=[output], live=True)  
-        
+
 if __name__ == "__main__":
     demo.queue(max_size=200).launch()