maybe this works??

app.py

@@ -33,9 +33,9 @@ model.load_state_dict(torch.load(hf_hub_download(repo_id="nickgardner/chatbot",
                                       filename="alpaca_train_400_epoch.pt"), map_location=device))
 model.eval()
 
-def respond(custom_string):
+def respond(input):
     model.eval()
-    src = torch.tensor(text_pipeline(custom_string), dtype=torch.int64).unsqueeze(0).to(device)
+    src = torch.tensor(text_pipeline(input), dtype=torch.int64).unsqueeze(0).to(device)
     src_mask = ((src != pad_token) & (src != unknown_token)).unsqueeze(-2).to(device)
     e_outputs = model.encoder(src, src_mask)
 
@@ -47,8 +47,7 @@ def respond(custom_string):
 
         out = model.out(model.decoder(outputs[:i].unsqueeze(0), e_outputs, src_mask, trg_mask))
         out = torch.nn.functional.softmax(out, dim=-1).detach()
-        print(out[:, -1].data.shape)
-        ix = np.random.choice(np.arange(out[:, -1].data), 1, p=out[:, -1].data)
+        ix = np.random.choice(np.arange(out[:, -1].data[0]), 1, p=out[:, -1].data[0])
         # val, ix = out[:, -1].data.topk(1)
 
         # outputs[i] = ix[0][0]