Update app.py

app.py

@@ -322,10 +322,10 @@ def motionllm(
     sample = {"instruction": prompt, "input": input_video_path}
     
     prefix = generate_prompt_mlp(sample)
-    pre = torch.cat((tokenizer.encode(prefix.split('INPUT_VIDEO: ')[0] + "\n", bos=True, eos=False, device=model.device).view(1, -1), tokenizer.encode("INPUT_VIDEO: ", bos=False, eos=False, device=model.device).view(1, -1)), dim=1)
+    pre = torch.cat((tokenizer.encode(prefix.split('INPUT_VIDEO: ')[0] + "\n", bos=True, eos=False, device=device).view(1, -1), tokenizer.encode("INPUT_VIDEO: ", bos=False, eos=False, device=device).view(1, -1)), dim=1)
 
     prompt = (pre, ". ASSISTANT: ")
-    encoded = (prompt[0], video_feature[0], tokenizer.encode(prompt[1], bos=False, eos=False, device=model.device).view(1, -1))
+    encoded = (prompt[0], video_feature[0], tokenizer.encode(prompt[1], bos=False, eos=False, device=device).view(1, -1))
     
     t0 = time.perf_counter()
         
@@ -565,7 +565,7 @@ print('Load mlp model again from', mlp_path)
 print(f"Time to load model: {time.time() - t0:.02f} seconds.", file=sys.stderr)
 
 model.eval()
-model = fabric.setup_module(model)
+model = model.cuda()
 linear_proj.eval()
 
 tokenizer = Tokenizer(tokenizer_llm_path)