Being unabe to transfer an allen.nlp.Elmo model and character_ids on the same device

Question

I am trying to measure the speed at which elmo model from the allen.nlp library performs on CPU and T4 GPU on google collab. The eval function accepts a function that makes sentence embedding with the given model

for m in elmo_models:
    print('\n\n\n', m, '\n==========================================')

    options_file = "/content/" + m + "/options.json"
    weight_file = "/content/" + m + "/model.hdf5"
    model = Elmo(options_file, weight_file, 2, dropout=0)

    speed_task_gpu.eval(lambda x: elmo_embed(x, model), m)
    speed_task_cpu.eval(lambda x: elmo_embed(x, model, 'cpu'), m)

And this is the function which is meant to get text embeddings:

def elmo_embed(text, model, name='cuda:0'):
  character_ids = batch_to_ids(text)

  device = torch.device(name)
  model = model.to(device)
  character_ids = character_ids.to(device)
  with torch.no_grad():
    embeddings = model(character_ids)['elmo_representations'][0]

  embeddings = embeddings.sum(dim=(0, 1), keepdim=True)
  embeddings = torch.nn.functional.normalize(embeddings)

  embeddings = embeddings[0][0].cpu().numpy()
  return {'cls': embeddings}

Even though I transfer both model and character_ids to the same device, I get this error when trying to evaluate model's performance on CPU: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper__index_select) What is the correct way of transfering model and character_ids to the same device?