Directly initialize msg_enc tensor to torch.float32 to support mps

[dw61]

Trying out SesameAILabs/csmmps which imports this. MPS on my mac only supports float32. See this attempt PR SesameAILabs/csm#52. Initialize the tensor early on directly to float32 fixes the error on MPS. tensor.float() is float32. Test by manually editing this silentcipher in site-packages. I can run the model locally on my M3 Pro Mac with this fix.

before: MPS error

(base) [main][~/Documents/Python/csm]$ PYTORCH_ENABLE_MPS_FALLBACK=1 python csm.py
import error: No module named 'triton'
/Users/wlt/Documents/Python/csm/generator.py:175: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  state_dict = torch.load(ckpt_path)
ckpt path or config path does not exist! Downloading the model from the Hugging Face Hub...
Fetching 13 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████| 13/13 [00:00<00:00, 24429.19it/s]
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:465: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self.enc_c.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, "enc_c.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:466: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self.dec_c.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, "dec_c.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:468: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  m.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, f"dec_m_{i}.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/bitsandbytes/cextension.py:34: UserWarning: The installed version of bitsandbytes was compiled without GPU support. 8-bit optimizers, 8-bit multiplication, and GPU quantization are unavailable.
  warn("The installed version of bitsandbytes was compiled without GPU support. "
'NoneType' object has no attribute 'cadam32bit_grad_fp32'
Traceback (most recent call last):
  File "/Users/wlt/Documents/Python/csm/csm.py", line 7, in <module>
    audio = generator.generate(
            ^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/wlt/Documents/Python/csm/generator.py", line 160, in generate
    audio, wm_sample_rate = watermark(self._watermarker, audio, self.sample_rate, CSM_1B_GH_WATERMARK)
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/wlt/Documents/Python/csm/watermarking.py", line 36, in watermark
    encoded, _ = watermarker.encode_wav(audio_array_44khz, 44100, watermark_key, calc_sdr=False, message_sdr=36)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py", line 318, in encode_wav
    msg_enc = torch.tensor(msgs, device=self.device).unsqueeze(0).float()
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.

after: no MPS error and successful audio output

(base) [33.504s][1][main][~/Documents/Python/csm]$ PYTORCH_ENABLE_MPS_FALLBACK=1 python csm.py
import error: No module named 'triton'
/Users/wlt/Documents/Python/csm/generator.py:175: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  state_dict = torch.load(ckpt_path)
ckpt path or config path does not exist! Downloading the model from the Hugging Face Hub...
Fetching 13 files: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 13/13 [00:00<00:00, 245612.40it/s]
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:465: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self.enc_c.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, "enc_c.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:466: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self.dec_c.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, "dec_c.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/server.py:468: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  m.load_state_dict(self.convert_dataparallel_to_normal(torch.load(os.path.join(ckpt_dir, f"dec_m_{i}.ckpt"), map_location=self.device)))
/opt/anaconda3/lib/python3.12/site-packages/bitsandbytes/cextension.py:34: UserWarning: The installed version of bitsandbytes was compiled without GPU support. 8-bit optimizers, 8-bit multiplication, and GPU quantization are unavailable.
  warn("The installed version of bitsandbytes was compiled without GPU support. "
'NoneType' object has no attribute 'cadam32bit_grad_fp32'
/opt/anaconda3/lib/python3.12/site-packages/silentcipher/stft.py:36: UserWarning: The operator 'aten::unfold_backward' is not currently supported on the MPS backend and will fall back to run on the CPU. This may have performance implications. (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/aten/src/ATen/mps/MPSFallback.mm:13.)
  inverse_transform = torch.istft(recombine_magnitude_phase, self.filter_length, hop_length=self.hop_len, win_length=self.win_len, window=self.window.to(magnitude.device)).unsqueeze(1)  # , length=self.num_samples

[ZackHodari]

Thanks! I've tested this on linux cpu+gpu and works there too