Update app.py
Browse files
app.py
CHANGED
@@ -19,9 +19,8 @@ except Exception as e:
|
|
19 |
print(f"Error detecting GPU. Using CPU. Error: {e}")
|
20 |
torch_device = torch.device("cpu")
|
21 |
|
22 |
-
# Load the SemantiCodec model
|
23 |
semanticodec = SemantiCodec(token_rate=100, semantic_vocab_size=32768)
|
24 |
-
semanticodec.to(torch_device)
|
25 |
|
26 |
# Global variables for cancellation
|
27 |
cancel_encode = False
|
@@ -105,18 +104,16 @@ def decode_audio(encoded_file_path):
|
|
105 |
tokens_numpy = np.frombuffer(tokens_numpy_bytes, dtype=np.int64).reshape(shape)
|
106 |
|
107 |
# Move the tensor to the same device as the model
|
108 |
-
tokens = torch.from_numpy(tokens_numpy)
|
109 |
-
|
110 |
-
print(f"
|
111 |
-
print(f"Model device: {semanticodec.device}")
|
112 |
-
|
113 |
-
# Ensure all model parameters are on the correct device
|
114 |
-
semanticodec.to(semanticodec.device)
|
115 |
|
116 |
# Decode the audio
|
117 |
with torch.no_grad():
|
118 |
waveform = semanticodec.decode(tokens)
|
119 |
|
|
|
|
|
120 |
# Move waveform to CPU for saving
|
121 |
waveform_cpu = waveform.cpu()
|
122 |
|
@@ -149,13 +146,9 @@ async def stream_decode_audio(encoded_file_path) -> Generator[tuple, None, None]
|
|
149 |
tokens_numpy = np.frombuffer(tokens_numpy_bytes, dtype=np.int64).reshape(shape)
|
150 |
|
151 |
# Move the tensor to the same device as the model
|
152 |
-
tokens = torch.from_numpy(tokens_numpy)
|
153 |
-
|
154 |
-
print(f"
|
155 |
-
print(f"Model device: {semanticodec.device}")
|
156 |
-
|
157 |
-
# Ensure all model parameters are on the correct device
|
158 |
-
semanticodec.to(semanticodec.device)
|
159 |
|
160 |
# Decode the audio in chunks
|
161 |
chunk_size = sample_rate * 2 # Adjust chunk size as needed
|
|
|
19 |
print(f"Error detecting GPU. Using CPU. Error: {e}")
|
20 |
torch_device = torch.device("cpu")
|
21 |
|
22 |
+
# Load the SemantiCodec model without specifying a device
|
23 |
semanticodec = SemantiCodec(token_rate=100, semantic_vocab_size=32768)
|
|
|
24 |
|
25 |
# Global variables for cancellation
|
26 |
cancel_encode = False
|
|
|
104 |
tokens_numpy = np.frombuffer(tokens_numpy_bytes, dtype=np.int64).reshape(shape)
|
105 |
|
106 |
# Move the tensor to the same device as the model
|
107 |
+
tokens = torch.from_numpy(tokens_numpy)
|
108 |
+
print(f"Tokens device: {tokens.device}")
|
109 |
+
print(f"Model device: {next(semanticodec.parameters()).device}")
|
|
|
|
|
|
|
|
|
110 |
|
111 |
# Decode the audio
|
112 |
with torch.no_grad():
|
113 |
waveform = semanticodec.decode(tokens)
|
114 |
|
115 |
+
print(f"Waveform device: {waveform.device}")
|
116 |
+
|
117 |
# Move waveform to CPU for saving
|
118 |
waveform_cpu = waveform.cpu()
|
119 |
|
|
|
146 |
tokens_numpy = np.frombuffer(tokens_numpy_bytes, dtype=np.int64).reshape(shape)
|
147 |
|
148 |
# Move the tensor to the same device as the model
|
149 |
+
tokens = torch.from_numpy(tokens_numpy)
|
150 |
+
print(f"Streaming tokens device: {tokens.device}")
|
151 |
+
print(f"Model device: {next(semanticodec.parameters()).device}")
|
|
|
|
|
|
|
|
|
152 |
|
153 |
# Decode the audio in chunks
|
154 |
chunk_size = sample_rate * 2 # Adjust chunk size as needed
|