Update app.py

app.py

@@ -30,7 +30,7 @@ phi_base_model = AutoModelForCausalLM.from_pretrained(
 from peft import PeftModel
 phi_new_model = "models/phi_adapter"
 phi_model = PeftModel.from_pretrained(phi_base_model, phi_new_model)
-phi_model = phi_model.merge_and_unload()
+phi_model = phi_model.merge_and_unload().to(device)
 
 '''compute_type = 'float32'
 if device != 'cpu':
@@ -134,9 +134,9 @@ def bot(history):
     else:
         if context_type == 'image':
             query_ids = tokenizer.encode(query)
-            query_ids = torch.tensor(query_ids, dtype=torch.int32).unsqueeze(0)
+            query_ids = torch.tensor(query_ids, dtype=torch.int32).unsqueeze(0).to(device)
             query_embeds = phi_model.get_input_embeddings()(query_ids)
-            inputs_embeds = torch.cat([context, query_embeds], dim=1)
+            inputs_embeds = torch.cat([context.to(device), query_embeds], dim=1)
             out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50,
                                      bos_token_id=tokenizer.bos_token_id)
             response = tokenizer.decode(out[0], skip_special_tokens=True)
@@ -144,7 +144,7 @@ def bot(history):
             input_text = context + query
 
             input_tokens = tokenizer.encode(input_text)
-            input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0)
+            input_ids = torch.tensor(input_tokens, dtype=torch.int32).unsqueeze(0).to(device)
             inputs_embeds = phi_model.get_input_embeddings()(input_ids)
             out = phi_model.generate(inputs_embeds=inputs_embeds, min_new_tokens=10, max_new_tokens=50,
                                      bos_token_id=tokenizer.bos_token_id)
@@ -214,12 +214,12 @@ with gr.Blocks() as app:
     chatbot.like(print_like_dislike, None, None)
     clear.click(clear_fn, None, chatbot, queue=False)
 
-    aud.stop_recording(audio_file, [chatbot, aud], [chatbot], queue=False).then(
+    aud.change(audio_file, [chatbot, aud], [chatbot], queue=False).then(
         bot, chatbot, chatbot, api_name="bot_response"
     )
-    aud.upload(audio_file, [chatbot, aud], [chatbot], queue=False).then(
+    '''aud.upload(audio_file, [chatbot, aud], [chatbot], queue=False).then(
         bot, chatbot, chatbot, api_name="bot_response"
-    )
+    )'''
 
 app.queue()
 app.launch()