take care of bfloat16 conversions

app.py

@@ -74,7 +74,10 @@ def get_device():
 
 def get_dtype(device: torch.device):
     if device == torch.device("cuda"):
-        dtype = torch.bfloat16
+        if torch.cuda.is_bf16_supported():
+            dtype = torch.bfloat16
+        else:
+            dtype = torch.float16
     elif device == torch.device("mps"):
         dtype = torch.float32
     else:
@@ -114,7 +117,7 @@ def embed_page_images(model, processor, page_images, batch_size=1):
             embeddings = model(**batch)
             page_embeddings.extend(list(torch.unbind(embeddings.to("cpu"))))
         pbar.progress((ibatch + 1) / len(page_images), text="embedding pages")
-    return np.array(page_embeddings)
+    return np.array([el.to(torch.float32) for el in page_embeddings])
 
 
 def embed_query_texts(model, processor, query_texts, batch_size=1):
@@ -132,7 +135,7 @@ def embed_query_texts(model, processor, query_texts, batch_size=1):
             batch = {k: v.to(model.device) for k, v in batch.items()}
             embeddings = model(**batch)
             query_embeddings.extend(list(torch.unbind(embeddings.to("cpu"))))
-    return np.array(query_embeddings)[0]
+    return np.array([el.to(torch.float32) for el in query_embeddings])[0]
 
 
 def get_pdf_page_images_from_bytes(