dropdown for both models

app.py

@@ -10,9 +10,16 @@ if torch.cuda.is_available():
 else:
     device = "cpu"
 
-tokenizer = AutoTokenizer.from_pretrained("PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct")
-model = AutoModelForCausalLM.from_pretrained("PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct", torch_dtype=torch.float16, device_map="auto").to(device)
-model.gradient_checkpointing_enable()
+def load_model_and_tokenizer(model_choice):
+    if model_choice == "Patronus Lynx 8B":
+        model_name = "PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct"
+    else:
+        model_name = "PatronusAI/Llama-3-Patronus-Lynx-70B-Instruct"
+    
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto").to(device)
+    model.gradient_checkpointing_enable()
+    return tokenizer, model
 
 PROMPT = """
 Given the following QUESTION, DOCUMENT and ANSWER you must analyze the provided answer and determine whether it is faithful to the contents of the DOCUMENT. The ANSWER must not offer new information beyond the context provided in the DOCUMENT. The ANSWER also must not contradict information provided in the DOCUMENT. Output your final verdict by strictly following this format: "PASS" if the answer is faithful to the DOCUMENT and "FAIL" if the answer is not faithful to the DOCUMENT. Show your reasoning.
@@ -66,7 +73,7 @@ HEADER = """
 """
 
 @spaces.GPU()
-def model_call(question, document, answer):
+def model_call(question, document, answer, tokenizer, model):
     device = next(model.parameters()).device
     NEW_FORMAT = PROMPT.format(question=question, document=document, answer=answer)
     print("ENTIRE NEW_FORMAT", NEW_FORMAT)
@@ -88,13 +95,31 @@ def model_call(question, document, answer):
     print(generated_text)
     return generated_text
 
+def update_model(model_choice, tokenizer_state, model_state):
+    new_tokenizer, new_model = load_model_and_tokenizer(model_choice)
+    print("UPDATED MODEL", new_tokenizer, new_model)
+    return new_tokenizer, new_model
+
 inputs = [
     gr.Textbox(label="Question"),
     gr.Textbox(label="Document"),
     gr.Textbox(label="Answer")
 ]
+output = gr.Textbox(label="Output")
+
+submit_button = gr.Button("Submit")
 
 with gr.Blocks() as demo:
     gr.Markdown(HEADER)
-    gr.Interface(fn=model_call, inputs=inputs, outputs="text")
+
+    tokenizer_state = gr.State()
+    model_state = gr.State()
+
+    model_dropdown = gr.Dropdown(choices=["Patronus Lynx 8B", "Patronus Lynx 70B"], value="Patronus Lynx 8B", label="Model")
+    model_dropdown.change(fn=update_model, inputs=[model_dropdown, tokenizer_state, model_state], outputs=[tokenizer_state, model_state])
+
+    submit_button.click(fn=model_call, inputs=inputs, outputs=output)
+
+initial_tokenizer, initial_model = load_model_and_tokenizer("Patronus Lynx 8B")
+demo.load(fn=lambda: (initial_tokenizer, initial_model), outputs=[tokenizer_state, model_state])
 demo.launch()