Update readme to reflect that values are no longer stored in uint8

README.md

@@ -18,7 +18,8 @@ def load_feat_acts(fname, only_active_docs=False):
   # Convert them back before constructing the matrix.
   csr_kwargs['crow_indices'] = csr_kwargs['crow_indices'].int()
   csr_kwargs['col_indices'] = csr_kwargs['col_indices'].int()
-  csr_kwargs['values'] = (csr_kwargs['values'].float()/255)
+  values = csr_kwargs['values'].float()
+  csr_kwargs['values'] = values/values.max()
   
   feat_acts = torch.sparse_csr_tensor(**csr_kwargs)
   return feat_acts