when encoding text embeds, force the move to accelerator on round one

[bghira]

Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)
Traceback (most recent call last):
  File "/mnt/Linux/Code/SimpleTuner/train.py", line 50, in <module>
    trainer.init_data_backend()
  File "/mnt/Linux/Code/SimpleTuner/helpers/training/trainer.py", line 500, in init_data_backend
    raise e
  File "/mnt/Linux/Code/SimpleTuner/helpers/training/trainer.py", line 473, in init_data_backend
    configure_multi_databackend(
  File "/mnt/Linux/Code/SimpleTuner/helpers/data_backend/factory.py", line 1224, in configure_multi_databackend
    init_backend["text_embed_cache"].compute_embeddings_for_prompts(
  File "/mnt/Linux/Code/SimpleTuner/helpers/caching/text_embeds.py", line 294, in compute_embeddings_for_prompts
    output = self.compute_prompt_embeddings_with_model(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/helpers/caching/text_embeds.py", line 409, in compute_prompt_embeddings_with_model
    text_encoder_output = self.model.encode_text_batch(
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/helpers/models/common.py", line 1123, in encode_text_batch
    encoded_text = self._encode_prompts(text_batch, is_negative_prompt)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/helpers/models/hidream/model.py", line 255, in _encode_prompts
    ]._encode_prompt(
      ^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/helpers/models/hidream/pipeline.py", line 752, in _encode_prompt
    t5_prompt_embeds = self._get_t5_prompt_embeds(
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/helpers/models/hidream/pipeline.py", line 244, in _get_t5_prompt_embeds
    prompt_embeds = self.text_encoder_3(
                    ^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/transformers/models/t5/modeling_t5.py", line 2100, in forward
    encoder_outputs = self.encoder(
                      ^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/transformers/models/t5/modeling_t5.py", line 1009, in forward
    inputs_embeds = self.embed_tokens(input_ids)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1739, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1750, in _call_impl
    return forward_call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/modules/sparse.py", line 190, in forward
    return F.embedding(
           ^^^^^^^^^^^^
  File "/mnt/Linux/Code/SimpleTuner/venv/lib/python3.12/site-packages/torch/nn/functional.py", line 2551, in embedding
    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

(venv) rmbp@linux:/mnt/Linux/Code/SimpleTuner$ 

resolves issue reported on Discord.