Update app.py

app.py

@@ -21,22 +21,44 @@ def transcribe(audio):
     text = model.stt_file(audio)[0]
     return text
 
-client1 = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
+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-v0.3")
+    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
 
 system_instructions1 = "<s>[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. The request asks you to provide friendly responses as if You are the character Jarvis, made by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]"
 
-def model(text):
+def models(text, model="Mixtral 8x7B", seed=42):
+
+    seed = int(randomize_seed_fn(seed))
+    generator = torch.Generator().manual_seed(seed)  
+    
+    client = client_fn(model)
+    
     generate_kwargs = dict(
         temperature=0.7,
         max_new_tokens=512,
         top_p=0.95,
         repetition_penalty=1,
         do_sample=True,
-        seed=42,
+        seed=seed,
     )
     
     formatted_prompt = system_instructions1 + text + "[JARVIS]"
-    stream = client1.text_generation(
+    stream = client.text_generation(
         formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
     output = ""
     for response in stream:
@@ -47,7 +69,7 @@ def model(text):
 
 async def respond(audio):
     user = transcribe(audio)
-    reply = model(user)
+    reply = models(user, model, seed)
     communicate = edge_tts.Communicate(reply)
     with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
         tmp_path = tmp_file.name
@@ -62,6 +84,22 @@ 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,
+        maximum=999999,
+        step=1,
+        value=0,
+        visible=False
+        )
         input = gr.Audio(label="User", sources="microphone", type="filepath", waveform_options=False)
         output = gr.Audio(label="AI", type="filepath",
                         interactive=False,
@@ -71,8 +109,8 @@ with gr.Blocks(css="style.css") as demo:
             batch=True,
             max_batch_size=10, 
             fn=respond, 
-            inputs=[input],
-            outputs=[output], live=False)  
+            inputs=[input, select, seed],
+            outputs=[output], live=True)  
 
 if __name__ == "__main__":
     demo.queue(max_size=200).launch()