Skip to content

fixed trt inference #471

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
1 commit merged into from Apr 27, 2020
+44 −38
Merged

fixed trt inference #471

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
70 changes: 36 additions & 34 deletions PyTorch/SpeechSynthesis/Tacotron2/trt/inference_trt.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,28 +130,30 @@ def init_decoder_outputs(memory, memory_lengths):
def init_decoder_tensors(decoder_inputs, decoder_outputs):

decoder_tensors = {
# inputs
'decoder_input': decoder_inputs[0],
'attention_hidden': decoder_inputs[1],
'attention_cell': decoder_inputs[2],
'decoder_hidden': decoder_inputs[3],
'decoder_cell': decoder_inputs[4],
'attention_weights': decoder_inputs[5],
'attention_weights_cum': decoder_inputs[6],
'attention_context': decoder_inputs[7],
'memory': decoder_inputs[8],
'processed_memory': decoder_inputs[9],
'mask': decoder_inputs[10],
# outputs
'out_attention_hidden': decoder_outputs[0],
'out_attention_cell': decoder_outputs[1],
'out_decoder_hidden': decoder_outputs[2],
'out_decoder_cell': decoder_outputs[3],
'out_attention_weights': decoder_outputs[4],
'out_attention_weights_cum': decoder_outputs[5],
'out_attention_context': decoder_outputs[6],
'decoder_output': decoder_outputs[7],
'gate_prediction': decoder_outputs[8],
"inputs" : {
'decoder_input': decoder_inputs[0],
'attention_hidden': decoder_inputs[1],
'attention_cell': decoder_inputs[2],
'decoder_hidden': decoder_inputs[3],
'decoder_cell': decoder_inputs[4],
'attention_weights': decoder_inputs[5],
'attention_weights_cum': decoder_inputs[6],
'attention_context': decoder_inputs[7],
'memory': decoder_inputs[8],
'processed_memory': decoder_inputs[9],
'mask': decoder_inputs[10]
},
"outputs" : {
'out_attention_hidden': decoder_outputs[0],
'out_attention_cell': decoder_outputs[1],
'out_decoder_hidden': decoder_outputs[2],
'out_decoder_cell': decoder_outputs[3],
'out_attention_weights': decoder_outputs[4],
'out_attention_weights_cum': decoder_outputs[5],
'out_attention_context': decoder_outputs[6],
'decoder_output': decoder_outputs[7],
'gate_prediction': decoder_outputs[8]
}
}
return decoder_tensors

Expand Down Expand Up @@ -196,10 +198,10 @@ def infer_tacotron2_trt(encoder, decoder_iter, postnet,
lens = torch.zeros_like(sequence_lengths)

encoder_tensors = {
# inputs
'sequences': sequences, 'sequence_lengths': sequence_lengths,
# outputs
'memory': memory, 'lens': lens, 'processed_memory': processed_memory
"inputs" :
{'sequences': sequences, 'sequence_lengths': sequence_lengths},
"outputs" :
{'memory': memory, 'lens': lens, 'processed_memory': processed_memory}
}

print("Running Tacotron2 Encoder")
Expand Down Expand Up @@ -252,10 +254,10 @@ def infer_tacotron2_trt(encoder, decoder_iter, postnet,
mel_outputs_postnet = torch.zeros_like(mel_outputs, device=device, dtype=dtype)

postnet_tensors = {
# inputs
'mel_outputs': mel_outputs,
# outputs
'mel_outputs_postnet': mel_outputs_postnet
"inputs" :
{'mel_outputs': mel_outputs},
"outputs" :
{'mel_outputs_postnet': mel_outputs_postnet}
}
print("Running Tacotron2 Postnet")
with MeasureTime(measurements, "tacotron2_postnet_time"):
Expand Down Expand Up @@ -284,10 +286,10 @@ def infer_waveglow_trt(waveglow, waveglow_context, mel, measurements, fp16):
audios = audios.half()

waveglow_tensors = {
# inputs
'mel': mel, 'z': z,
# outputs
'audio': audios
"inputs" :
{'mel': mel, 'z': z},
"outputs" :
{'audio': audios}
}
print("Running WaveGlow")
with MeasureTime(measurements, "waveglow_time"):
Expand Down
2 changes: 1 addition & 1 deletion PyTorch/SpeechSynthesis/Tacotron2/trt/test_infer_trt.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,7 @@
from inference import checkpoint_from_distributed, unwrap_distributed, MeasureTime, prepare_input_sequence
from inference_trt import infer_tacotron2_trt, infer_waveglow_trt

from trt.trt_utils import load_engine, run_trt_engine
from trt.trt_utils import load_engine
import tensorrt as trt

import time
Expand Down
10 changes: 7 additions & 3 deletions PyTorch/SpeechSynthesis/Tacotron2/trt/trt_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@

import tensorrt as trt


def is_dimension_dynamic(dim):
return dim is None or dim <= 0

Expand All @@ -38,15 +39,18 @@ def is_shape_dynamic(shape):
def run_trt_engine(context, engine, tensors):

bindings = [None]*engine.num_bindings
for name in tensors.keys():
for name,tensor in tensors['inputs'].items():
idx = engine.get_binding_index(name)
tensor = tensors.get(name)
bindings[idx] = tensor.data_ptr()
if engine.is_shape_binding(idx) and is_shape_dynamic(context.get_shape(idx)):
context.set_shape_input(idx, tensor)
elif is_shape_dynamic(context.get_binding_shape(idx)):
elif is_shape_dynamic(engine.get_binding_shape(idx)):
context.set_binding_shape(idx, tensor.shape)

for name,tensor in tensors['outputs'].items():
idx = engine.get_binding_index(name)
bindings[idx] = tensor.data_ptr()

context.execute_v2(bindings=bindings)


Expand Down