mirror of
https://github.com/storytold/torchscript-models.git
synced 2026-10-19 07:42:59 +00:00
Brandon Thomas
164 lines
7.9 KiB
Python
Executable File
164 lines
7.9 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
# From: https://github.com/pytorch/pytorch/issues/20356#issuecomment-545572400
|
|
import torch
|
|
|
|
# Import glow-tts code
|
|
import sys
|
|
sys.path.append("./glow-tts-multispeaker")
|
|
import utils
|
|
import models
|
|
import commons
|
|
from text.symbols import symbols
|
|
|
|
# Nvidia Apex
|
|
#from apex import amp
|
|
#from apex.parallel import DistributedDataParallel as DDP
|
|
|
|
# Data
|
|
import numpy as np
|
|
from text import text_to_sequence, cmudict
|
|
|
|
trace_model = False
|
|
if sys.argv[1] == 'true':
|
|
trace_model = True
|
|
elif sys.argv[1] == 'false':
|
|
trace_model = False
|
|
else:
|
|
raise Exception('Scripting or tracing?')
|
|
|
|
torch.set_grad_enabled(False)
|
|
|
|
def load(model_checkpoint_path):
|
|
model_dir = "./logs/base/"
|
|
hps = utils.get_hparams_from_dir(model_dir)
|
|
print('Checkpoint path: {}'.format(model_checkpoint_path))
|
|
|
|
with torch.no_grad():
|
|
# NB(bt): This should fix the following:
|
|
# RuntimeError: Cannot insert a Tensor that requires grad as a constant.
|
|
model = models.FlowGenerator(
|
|
len(symbols),
|
|
out_channels=hps.data.n_mel_channels,
|
|
**hps.model) #.to("cuda")
|
|
|
|
utils.load_checkpoint(model_checkpoint_path, model)
|
|
model.decoder.store_inverse() # do not calcuate jacobians for fast decoding
|
|
_ = model.eval()
|
|
|
|
return (hps, model)
|
|
|
|
|
|
#model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut/pretrained.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_998.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1063.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1086.pth'
|
|
|
|
# Training *EXPENSIVE* LJS base model on LambdaLabs Tesla GPU
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_135.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_336.pth'
|
|
|
|
# NB: Had to adjust num_speakers from 64 to 8
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_44.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_142.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_264.pth'
|
|
|
|
# Transfer learn from dense 8-speaker LJS base
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_309.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_325.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_337.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_541.pth'
|
|
# Transfer learning, but with less dense LJS (1000sec total across 4 channels)
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_596.pth' # All channels taking on speaker character
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_607.pth' # Same, but better fit + enunciated words
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_757.pth' # Trump great, but dominates other speakers
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_845.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_929.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1112.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1204.pth'
|
|
|
|
# Removed 75% of trump training data to make speakers more equal (started from 583)
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_604.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_619.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_640.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_854.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1027.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1227.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1319.pth'
|
|
|
|
# Restart from 1000, with more data for speakers + ronald reagan
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1048.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1153.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_1802.pth' # These spectrograms are amazing
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_2458.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_2898.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_3208.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_4244.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_4817.pth' # Run5: I just started "run 5", so maybe it went bad
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_4815.pth' # SAVE THIS. Run4: Last of "run 4" (best sound so far)
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_5092.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_5634.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_6045.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_6590.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_6805.pth'
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-multispeaker/G_7351.pth'
|
|
|
|
# This experiment is over.
|
|
model_checkpoint_path = '/home/bt/models/glow-tts-jaywalnut-8speaker-256gin/glow_tts_multispeaker_8-speaker_256-gin_ljs_txlearn_2020.06.19_experiment-5_chkpt-7466.pth'
|
|
|
|
hps, glow_model = load(model_checkpoint_path)
|
|
output_filename = 'glow_multispeaker.torchjit'
|
|
output_filename = 'glow_ljs_base.torchjit'
|
|
output_filename = 'glow_ljs_base_dense.torchjit'
|
|
output_filename = 'glow_ljs_base_dense2.torchjit'
|
|
output_filename = 'glow_ljs_base_dense3.torchjit'
|
|
output_filename = 'glow_ljs_multispeaker_1.torchjit'
|
|
output_filename = 'glow_tts_multispeaker_8-speaker_256-gin_ljs_txlearn_2020.06.19_experiment-5_chkpt-7466.torchjit'
|
|
|
|
|
|
cmu_dict = cmudict.CMUDict(hps.data.cmudict_path)
|
|
|
|
tst_stn = "This is a test of the machine learning model. How does this sound? I wonder if the length of this text has any importance whatsoever in the model evaluation. Perhaps it does. Perhaps not. Who knows?"
|
|
seq = text_to_sequence(tst_stn, ['english_cleaners'], cmu_dict)
|
|
sequence = np.array(seq)[None, :] # NB: None adds a new axis.
|
|
print("".join([symbols[c] for c in sequence[0]]))
|
|
|
|
torched = torch.from_numpy(sequence)
|
|
|
|
x_tst = torch.autograd.Variable(torched).long()
|
|
x_tst_lengths = torch.tensor([x_tst.shape[1]])
|
|
|
|
|
|
speaker_id = 1
|
|
speaker_g = torch.IntTensor([int(speaker_id)]).long()
|
|
|
|
print('\n tst_stn', len(tst_stn), tst_stn)
|
|
print('\n seq', len(seq), seq)
|
|
print('\n sequence', len(sequence), sequence.size, sequence)
|
|
print('\n torched', len(torched), torched.shape, torched)
|
|
|
|
print('\n x_tst', len(x_tst), x_tst)
|
|
print('\n x_tst_lengths', len(x_tst_lengths), x_tst_lengths)
|
|
|
|
print('\n speaker_g', len(speaker_g), speaker_g)
|
|
|
|
|
|
# NB: Tracing evaluates the model on input and unrolls and hardcodes branching
|
|
# and loops. Scripting allows these to remain by converting the entire program
|
|
# to TorchScript. Scripting seems MUCH faster, is also harder get working.
|
|
#trace_model = True # TODO: Try True.
|
|
if trace_model:
|
|
print('Tracing model and saving as: {}'.format(output_filename))
|
|
#mel_file = '/home/bt/dev/voder/data/mels/LJ002-0320.mel'
|
|
#mel = torch.load(mel_file, map_location=torch.device('cpu'))
|
|
# NB: Getting `Tracing failed sanity checks! Graphs differed across invocations!`
|
|
inputs = (x_tst, x_tst_lengths, speaker_g)
|
|
#traced_script_module = torch.jit.trace(glow_model, inputs, check_trace=True)
|
|
traced_script_module = torch.jit.trace(glow_model, inputs, check_trace=False)
|
|
traced_script_module.save(output_filename)
|
|
else:
|
|
print('Saving script model as: {}'.format(output_filename))
|
|
container = torch.jit.script(glow_model)
|
|
container.save(output_filename)
|
|
|
|
|