read_fast.py requires CUDA. How to run on M1 Mac.

[kebwi]

do_tts.py and read.py work on my Mac Studio (M1 Max), but read_fast.py exhibits a common Mac PyTorch/tortoise complaint about being configured to require CUDA. I have encountered this in other PyTorch projects, but I'm unsure where in tortoise to alter just read_fast.py (not underlying tortoise, since do_tts.py and read.py work fine) to address this issue. I would prefer to configure it to use the MPS option, which I have found to be helpful on other PyTorch projects, but I'll settle for CPU if I must. Ultimately, I'm curious if read_fast.py is actually any faster than read.py. I suppose likewise, I would like to know if do_tts.py and read.py use MPS or just CPU, and if they are using CPU, can I ensure they use MPS when possible? I have found some resources on that, but I'm not sure I fully understand how to guarantee that tortoise uses MPS...but first, how do I get read_fast.py running?

Thanks.

$ PYTORCH_ENABLE_MPS_FALLBACK=1 python tortoise/read_fast.py --textfile input/some_text.txt --voice tom --preset fast
tortoise_tts_env/lib/python3.9/site-packages/urllib3/__init__.py:35: NotOpenSSLWarning: urllib3 v2 only supports OpenSSL 1.1.1+, currently the 'ssl' module is compiled with 'LibreSSL 2.8.3'. See: https://github.com/urllib3/urllib3/issues/3020
  warnings.warn(
Some weights of the model checkpoint at jbetker/wav2vec2-large-robust-ft-libritts-voxpopuli were not used when initializing Wav2Vec2ForCTC: ['wav2vec2.encoder.pos_conv_embed.conv.weight_v', 'wav2vec2.encoder.pos_conv_embed.conv.weight_g']
- This IS expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at jbetker/wav2vec2-large-robust-ft-libritts-voxpopuli and are newly initialized: ['wav2vec2.encoder.pos_conv_embed.conv.parametrizations.weight.original0', 'wav2vec2.encoder.pos_conv_embed.conv.parametrizations.weight.original1']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
tortoise_tts_env/lib/python3.9/site-packages/torch/nn/utils/weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.
  warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.")
hifidecoder.pth: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 71.5M/71.5M [00:19<00:00, 3.67MB/s]
Traceback (most recent call last):
  File "tortoise-tts/tortoise/read_fast.py", line 33, in <module>
    tts = TextToSpeech(models_dir=args.model_dir, use_deepspeed=args.use_deepspeed, kv_cache=args.kv_cache, half=args.half)
  File "tortoise-tts/tortoise/api_fast.py", line 222, in __init__
    hifi_model = torch.load(get_model_path('hifidecoder.pth'))
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 1025, in load
    return _load(opened_zipfile,
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 1445, in _load
    result = unpickler.load()
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 1415, in persistent_load
    typed_storage = load_tensor(dtype, nbytes, key, _maybe_decode_ascii(location))
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 1389, in load_tensor
    wrap_storage=restore_location(storage, location),
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 390, in default_restore_location
    result = fn(storage, location)
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 265, in _cuda_deserialize
    device = validate_cuda_device(location)
  File "tortoise_tts_env/lib/python3.9/site-packages/torch/serialization.py", line 249, in validate_cuda_device
    raise RuntimeError('Attempting to deserialize object on a CUDA '
RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.

[kebwi]

Oh, adding map_location=torch.device('cpu') on read_fast.py, line 223 in the call to torch.load, in accordance with the corresponding line in read.py, seems to have gotten things off the ground. The docs for map_location seem pretty clear on what is going on there. I'm still new to PyTorch. I'm waiting to see if the run goes to completion but it doesn't appear to have stalled out early as it did previously.