ConvTasNet Saving / Loading Model #563

cemox35 · 2021-10-10T11:20:05Z

cemox35
Oct 10, 2021

Hi,

I use the ConvTasNet model in Asteroid and Trainer automatically saves the model's checkpoints in PyTorch-lightning file, however, I cannot use "load_from_checkpoint" or "from_pretrained".

What is the correct solution of the saving and loading models use for the test, not training?

Answered by cemox35

Oct 13, 2021

Oh, I think, I need a use System instead ConvTasNet. However, I got another error.

This is my code:
`MODEL_PATH = ["lightning_logs/version_53/checkpoints/epoch=4-step=1444.ckpt"] # write model path

i = 5

train_loader, val_loader = load_data()

model = ConvTasNet(
    n_src=SPEAKER,
    n_filters=N[i % 3],
    kernel_size=L[i % 3],
    bn_chan=B[i % 3],
    skip_chan=Sc[i % 3],
    hid_chan=H[i % 3],
    conv_kernel_size=P[i % 3],
    n_blocks=X[i % 3],
    n_repeats=R[i % 3],
    sample_rate=SAMPLE_RATE,
    fb_name=ENCODER[i]
).double()

if torch.cuda.is_available():
    device = "cuda"
else:
    device = "cpu"
print(f"Using {device}")

model.to(device, non_blocking=True)

loss = PITLos…

View full answer

mpariente · 2021-10-10T12:52:03Z

mpariente
Oct 10, 2021
Maintainer

Can you show how you do the loading and the error message? Le dim. 10 oct. 2021 à 13:20, Cem Ayar ***@***.***> a écrit :

…

Hi, I use the ConvTasNet model in Asteroid and Trainer automatically saves the model's checkpoints in PyTorch-lightning file, however, I cannot use *"load_from_checkpoint"* or *"from_pretrained".* What is the correct solution of the saving and loading models use for the test, not training? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#563>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEND2HBBLGFPXQO3YNMPRNLUGFZHFANCNFSM5FWLT2DQ> .

5 replies

cemox35 Oct 10, 2021
Author

First try:

ENCODER = ["stft", "stft", "stft", "free", "free", "free"]
MODEL_PATH = ["lightning_logs/version_53/checkpoints/epoch=4-step=1444.ckpt"]  # write model path
i = 0

N = [512, 512, 512, 256]
L = [40, 40, 16, ]  # 2.5ms
B = [128, 128, 128, 256]
H = [256, 512, 512, 512]
Sc = [128, 512, 128, ]  # ??
P = [3, 3, 3, 3]
X = [7, 7, 8, 8]
R = [2, 2, 3, 4]

SAMPLE_RATE = 8000

model = ConvTasNet(
    n_src=SPEAKER,
    n_filters=N[i % 3],
    kernel_size=L[i % 3],
    bn_chan=B[i % 3],
    skip_chan=Sc[i % 3],
    hid_chan=H[i % 3],
    conv_kernel_size=P[i % 3],
    n_blocks=X[i % 3],
    n_repeats=R[i % 3],
    sample_rate=SAMPLE_RATE,
    fb_name=ENCODER[i]
).double()

model.from_pretrained(pretrained_model_conf_or_path=MODEL_PATH[i])`

ValueError: Expected config dictionary to have field model_name`. Found only: dict_keys(['epoch', 'global_step', 'pytorch-lightning_version', 'state_dict', 'callbacks', 'optimizer_states', 'lr_schedulers', 'native_amp_scaling_state', 'training_config'])

When I tried to "load_from_checkpoint":

model = ConvTasNet.load_from_checkpoint(MODEL_PATH[i])

AttributeError: type object 'ConvTasNet' has no attribute 'load_from_checkpoint'

Last try:

model = model.load_state_dict(torch.load(MODEL_PATH[i]))

RuntimeError: Error(s) in loading state_dict for ConvTasNet:
Missing key(s) in state_dict: "encoder.filterbank._filters", "masker.bottleneck.0.gamma", "masker.bottleneck.0.beta", "masker.bottleneck.1.weight", "masker.bottleneck.1.bias", "masker.TCN.0.shared_block.0.weight", "masker.TCN.0.shared_block.0.bias", "masker.TCN.0.shared_block.1.weight", "masker.TCN.0.shared_block.2.gamma", "masker.TCN.0.shared_block.2.beta", "masker.TCN.0.shared_block.3.weight", "masker.TCN.0.shared_block.3.bias", "masker.TCN.0.shared_block.4.weight", "masker.TCN.0.shared_block.5.gamma", "masker.TCN.0.shared_block.5.beta", "masker.TCN.0.res_conv.weight", "masker.TCN.0.res_conv.bias", "masker.TCN.0.skip_conv.weight", "masker.TCN.0.skip_conv.bias", "masker.TCN.1.shared_block.0.weight", "masker.TCN.1.shared_block.0.bias", "masker.TCN.1.shared_block.1.weight", "masker.TCN.1.shared_block.2.gamma", "masker.TCN.1.shared_block.2.beta", "masker.TCN.1.shared_block.3.weight", "masker.TCN.1.shared_block.3.bias", "masker.TCN.1.shared_block.4.weight", "masker.TCN.1.shared_block.5.gamma", "masker.TCN.1.shared_block.5.beta", "masker.TCN.1.res_conv.weight", "masker.TCN.1.res_conv.bias", "masker.TCN.1.skip_conv.weight", "masker.TCN.1.skip_conv.bias", "masker.TCN.2.shared_block.0.weight", "masker.TCN.2.shared_block.0.bias", "masker.TCN.2.shared_block.1.weight", "masker.TCN.2.shared_block.2.gamma", "masker.TCN.2.shared_block.2.beta", "masker.TCN.2.shared_block.3.weight", "masker.TCN.2.shared_block.3.bias", "masker.TCN.2.shared_block.4.weight", "masker.TCN.2.shared_block.5.gamma", "masker.TCN.2.shared_block.5.beta", "masker.TCN.2.res_conv.weight", "masker.TCN.2.res_conv.bias", "masker.TCN.2.skip_conv.weight", "masker.TCN.2.skip_conv.bias", "masker.TCN.3.shared_block.0.weight", "masker.TCN.3.shared_block.0.bias", "masker.TCN.3.shared_block.1.weight", "masker.TCN.3.shared_block.2.gamma", "masker.TCN.3.shared_block.2.beta", "masker.TCN.3.shared_block.3.weight", "masker.TCN.3.shared_block.3.bias", "masker.TCN.3.shared_block.4.weight", "masker.TCN.3.shared_block.5.gamma", "masker.TCN.3.shared_block.5.beta", "masker.TCN.3.res_conv.weight", "masker.TCN.3.res_conv.bias", "masker.TCN.3.skip_conv.weight", "masker.TCN.3.skip_conv.bias", "masker.TCN.4.shared_block.0.weight", "masker.TCN.4.shared_block.0.bias", "masker.TCN.4.shared_block.1.weight", "masker.TCN.4.shared_block.2.gamma", "masker.TCN.4.shared_block.2.beta", "masker.TCN.4.shared_block.3.weight", "masker.TCN.4.shared_block.3.bias", "masker.TCN.4.shared_block.4.weight", "masker.TCN.4.shared_block.5.gamma", "masker.TCN.4.shared_block.5.beta", "masker.TCN.4.res_conv.weight", "masker.TCN.4.res_conv.bias", "masker.TCN.4.skip_conv.weight", "masker.TCN.4.skip_conv.bias", "masker.TCN.5.shared_block.0.weight", "masker.TCN.5.shared_block.0.bias", "masker.TCN.5.shared_block.1.weight", "masker.TCN.5.shared_block.2.gamma", "masker.TCN.5.shared_block.2.beta", "masker.TCN.5.shared_block.3.weight", "masker.TCN.5.shared_block.3.bias", "masker.TCN.5.shared_block.4.weight", "masker.TCN.5.shared_block.5.gamma", "masker.TCN.5.shared_block.5.beta", "masker.TCN.5.res_conv.weight", "masker.TCN.5.res_conv.bias", "masker.TCN.5.skip_conv.weight", "masker.TCN.5.skip_conv.bias", "masker.TCN.6.shared_block.0.weight", "masker.TCN.6.shared_block.0.bias", "masker.TCN.6.shared_block.1.weight", "masker.TCN.6.shared_block.2.gamma", "masker.TCN.6.shared_block.2.beta", "masker.TCN.6.shared_block.3.weight", "masker.TCN.6.shared_block.3.bias", "masker.TCN.6.shared_block.4.weight", "masker.TCN.6.shared_block.5.gamma", "masker.TCN.6.shared_block.5.beta", "masker.TCN.6.res_conv.weight", "masker.TCN.6.res_conv.bias", "masker.TCN.6.skip_conv.weight", "masker.TCN.6.skip_conv.bias", "masker.TCN.7.shared_block.0.weight", "masker.TCN.7.shared_block.0.bias", "masker.TCN.7.shared_block.1.weight", "masker.TCN.7.shared_block.2.gamma", "masker.TCN.7.shared_block.2.beta", "masker.TCN.7.shared_block.3.weight", "masker.TCN.7.shared_block.3.bias", "masker.TCN.7.shared_block.4.weight", "masker.TCN.7.shared_block.5.gamma", "masker.TCN.7.shared_block.5.beta", "masker.TCN.7.res_conv.weight", "masker.TCN.7.res_conv.bias", "masker.TCN.7.skip_conv.weight", "masker.TCN.7.skip_conv.bias", "masker.TCN.8.shared_block.0.weight", "masker.TCN.8.shared_block.0.bias", "masker.TCN.8.shared_block.1.weight", "masker.TCN.8.shared_block.2.gamma", "masker.TCN.8.shared_block.2.beta", "masker.TCN.8.shared_block.3.weight", "masker.TCN.8.shared_block.3.bias", "masker.TCN.8.shared_block.4.weight", "masker.TCN.8.shared_block.5.gamma", "masker.TCN.8.shared_block.5.beta", "masker.TCN.8.res_conv.weight", "masker.TCN.8.res_conv.bias", "masker.TCN.8.skip_conv.weight", "masker.TCN.8.skip_conv.bias", "masker.TCN.9.shared_block.0.weight", "masker.TCN.9.shared_block.0.bias", "masker.TCN.9.shared_block.1.weight", "masker.TCN.9.shared_block.2.gamma", "masker.TCN.9.shared_block.2.beta", "masker.TCN.9.shared_block.3.weight", "masker.TCN.9.shared_block.3.bias", "masker.TCN.9.shared_block.4.weight", "masker.TCN.9.shared_block.5.gamma", "masker.TCN.9.shared_block.5.beta", "masker.TCN.9.res_conv.weight", "masker.TCN.9.res_conv.bias", "masker.TCN.9.skip_conv.weight", "masker.TCN.9.skip_conv.bias", "masker.TCN.10.shared_block.0.weight", "masker.TCN.10.shared_block.0.bias", "masker.TCN.10.shared_block.1.weight", "masker.TCN.10.shared_block.2.gamma", "masker.TCN.10.shared_block.2.beta", "masker.TCN.10.shared_block.3.weight", "masker.TCN.10.shared_block.3.bias", "masker.TCN.10.shared_block.4.weight", "masker.TCN.10.shared_block.5.gamma", "masker.TCN.10.shared_block.5.beta", "masker.TCN.10.res_conv.weight", "masker.TCN.10.res_conv.bias", "masker.TCN.10.skip_conv.weight", "masker.TCN.10.skip_conv.bias", "masker.TCN.11.shared_block.0.weight", "masker.TCN.11.shared_block.0.bias", "masker.TCN.11.shared_block.1.weight", "masker.TCN.11.shared_block.2.gamma", "masker.TCN.11.shared_block.2.beta", "masker.TCN.11.shared_block.3.weight", "masker.TCN.11.shared_block.3.bias", "masker.TCN.11.shared_block.4.weight", "masker.TCN.11.shared_block.5.gamma", "masker.TCN.11.shared_block.5.beta", "masker.TCN.11.res_conv.weight", "masker.TCN.11.res_conv.bias", "masker.TCN.11.skip_conv.weight", "masker.TCN.11.skip_conv.bias", "masker.TCN.12.shared_block.0.weight", "masker.TCN.12.shared_block.0.bias", "masker.TCN.12.shared_block.1.weight", "masker.TCN.12.shared_block.2.gamma", "masker.TCN.12.shared_block.2.beta", "masker.TCN.12.shared_block.3.weight", "masker.TCN.12.shared_block.3.bias", "masker.TCN.12.shared_block.4.weight", "masker.TCN.12.shared_block.5.gamma", "masker.TCN.12.shared_block.5.beta", "masker.TCN.12.res_conv.weight", "masker.TCN.12.res_conv.bias", "masker.TCN.12.skip_conv.weight", "masker.TCN.12.skip_conv.bias", "masker.TCN.13.shared_block.0.weight", "masker.TCN.13.shared_block.0.bias", "masker.TCN.13.shared_block.1.weight", "masker.TCN.13.shared_block.2.gamma", "masker.TCN.13.shared_block.2.beta", "masker.TCN.13.shared_block.3.weight", "masker.TCN.13.shared_block.3.bias", "masker.TCN.13.shared_block.4.weight", "masker.TCN.13.shared_block.5.gamma", "masker.TCN.13.shared_block.5.beta", "masker.TCN.13.res_conv.weight", "masker.TCN.13.res_conv.bias", "masker.TCN.13.skip_conv.weight", "masker.TCN.13.skip_conv.bias", "masker.mask_net.0.weight", "masker.mask_net.1.weight", "masker.mask_net.1.bias", "decoder.filterbank._filters".
Unexpected key(s) in state_dict: "epoch", "global_step", "pytorch-lightning_version", "state_dict", "callbacks", "optimizer_states", "lr_schedulers", "native_amp_scaling_state", "training_config".

mpariente Oct 11, 2021
Maintainer

You should probably load the state_dict only.

Can you try to reproduce the problem with a minimal script, I'll help from there.

cemox35 Oct 12, 2021
Author

This is my test code:

import torch
from asteroid.models import ConvTasNet

def test_models():

MODEL_PATH = ["lightning_logs/version_53/checkpoints/epoch=4-step=1444.ckpt"]  # write model path

i = 0
SPEAKER = 1

ENCODER = ["stft", "stft", "stft", "free", "free", "free"]

N = [512, 512, 512, 256]
L = [40, 40, 16, ]  # 2.5ms
B = [128, 128, 128, 256]
H = [256, 512, 512, 512]
Sc = [128, 512, 128, ]  # ??
P = [3, 3, 3, 3]
X = [7, 7, 8, 8]
R = [2, 2, 3, 4]

SAMPLE_RATE = 8000

model = ConvTasNet(
    n_src=SPEAKER,
    n_filters=N[i % 3],
    kernel_size=L[i % 3],
    bn_chan=B[i % 3],
    skip_chan=Sc[i % 3],
    hid_chan=H[i % 3],
    conv_kernel_size=P[i % 3],
    n_blocks=X[i % 3],
    n_repeats=R[i % 3],
    sample_rate=SAMPLE_RATE,
    fb_name=ENCODER[i]
).double()

model = model.load_state_dict(torch.load(MODEL_PATH[i]))

model.separate("s1.wav", resample=True)

if name == 'main':
test_models()
`

Its give that error:

Traceback (most recent call last):
File "F:\ALL PROJECTS\Pycharm Projects\Thesis\train.py", line 179, in
test_models()
File "F:\ALL PROJECTS\Pycharm Projects\Thesis\train.py", line 170, in test_models
model = model.load_state_dict(torch.load(MODEL_PATH[i]))
File "C:\conda\envs\thesis\lib\site-packages\torch\nn\modules\module.py", line 1406, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for ConvTasNet:
Missing key(s) in state_dict: "encoder.filterbank._filters", "masker.bottleneck.0.gamma", "masker.bottleneck.0.beta", "masker.bottleneck.1.weight", "masker.bottleneck.1.bias", "masker.TCN.0.shared_block.0.weight", "masker.TCN.0.shared_block.0.bias", "masker.TCN.0.shared_block.1.weight", "masker.TCN.0.shared_block.2.gamma", "masker.TCN.0.shared_block.2.beta", "masker.TCN.0.shared_block.3.weight", "masker.TCN.0.shared_block.3.bias", "masker.TCN.0.shared_block.4.weight", "masker.TCN.0.shared_block.5.gamma", "masker.TCN.0.shared_block.5.beta", "masker.TCN.0.res_conv.weight", "masker.TCN.0.res_conv.bias", "masker.TCN.0.skip_conv.weight", "masker.TCN.0.skip_conv.bias", "masker.TCN.1.shared_block.0.weight", "masker.TCN.1.shared_block.0.bias", "masker.TCN.1.shared_block.1.weight", "masker.TCN.1.shared_block.2.gamma", "masker.TCN.1.shared_block.2.beta", "masker.TCN.1.shared_block.3.weight", "masker.TCN.1.shared_block.3.bias", "masker.TCN.1.shared_block.4.weight", "masker.TCN.1.shared_block.5.gamma", "masker.TCN.1.shared_block.5.beta", "masker.TCN.1.res_conv.weight", "masker.TCN.1.res_conv.bias", "masker.TCN.1.skip_conv.weight", "masker.TCN.1.skip_conv.bias", "masker.TCN.2.shared_block.0.weight", "masker.TCN.2.shared_block.0.bias", "masker.TCN.2.shared_block.1.weight", "masker.TCN.2.shared_block.2.gamma", "masker.TCN.2.shared_block.2.beta", "masker.TCN.2.shared_block.3.weight", "masker.TCN.2.shared_block.3.bias", "masker.TCN.2.shared_block.4.weight", "masker.TCN.2.shared_block.5.gamma", "masker.TCN.2.shared_block.5.beta", "masker.TCN.2.res_conv.weight", "masker.TCN.2.res_conv.bias", "masker.TCN.2.skip_conv.weight", "masker.TCN.2.skip_conv.bias", "masker.TCN.3.shared_block.0.weight", "masker.TCN.3.shared_block.0.bias", "masker.TCN.3.shared_block.1.weight", "masker.TCN.3.shared_block.2.gamma", "masker.TCN.3.shared_block.2.beta", "masker.TCN.3.shared_block.3.weight", "masker.TCN.3.shared_block.3.bias", "masker.TCN.3.shared_block.4.weight", "masker.TCN.3.shared_block.5.gamma", "masker.TCN.3.shared_block.5.beta", "masker.TCN.3.res_conv.weight", "masker.TCN.3.res_conv.bias", "masker.TCN.3.skip_conv.weight", "masker.TCN.3.skip_conv.bias", "masker.TCN.4.shared_block.0.weight", "masker.TCN.4.shared_block.0.bias", "masker.TCN.4.shared_block.1.weight", "masker.TCN.4.shared_block.2.gamma", "masker.TCN.4.shared_block.2.beta", "masker.TCN.4.shared_block.3.weight", "masker.TCN.4.shared_block.3.bias", "masker.TCN.4.shared_block.4.weight", "masker.TCN.4.shared_block.5.gamma", "masker.TCN.4.shared_block.5.beta", "masker.TCN.4.res_conv.weight", "masker.TCN.4.res_conv.bias", "masker.TCN.4.skip_conv.weight", "masker.TCN.4.skip_conv.bias", "masker.TCN.5.shared_block.0.weight", "masker.TCN.5.shared_block.0.bias", "masker.TCN.5.shared_block.1.weight", "masker.TCN.5.shared_block.2.gamma", "masker.TCN.5.shared_block.2.beta", "masker.TCN.5.shared_block.3.weight", "masker.TCN.5.shared_block.3.bias", "masker.TCN.5.shared_block.4.weight", "masker.TCN.5.shared_block.5.gamma", "masker.TCN.5.shared_block.5.beta", "masker.TCN.5.res_conv.weight", "masker.TCN.5.res_conv.bias", "masker.TCN.5.skip_conv.weight", "masker.TCN.5.skip_conv.bias", "masker.TCN.6.shared_block.0.weight", "masker.TCN.6.shared_block.0.bias", "masker.TCN.6.shared_block.1.weight", "masker.TCN.6.shared_block.2.gamma", "masker.TCN.6.shared_block.2.beta", "masker.TCN.6.shared_block.3.weight", "masker.TCN.6.shared_block.3.bias", "masker.TCN.6.shared_block.4.weight", "masker.TCN.6.shared_block.5.gamma", "masker.TCN.6.shared_block.5.beta", "masker.TCN.6.res_conv.weight", "masker.TCN.6.res_conv.bias", "masker.TCN.6.skip_conv.weight", "masker.TCN.6.skip_conv.bias", "masker.TCN.7.shared_block.0.weight", "masker.TCN.7.shared_block.0.bias", "masker.TCN.7.shared_block.1.weight", "masker.TCN.7.shared_block.2.gamma", "masker.TCN.7.shared_block.2.beta", "masker.TCN.7.shared_block.3.weight", "masker.TCN.7.shared_block.3.bias", "masker.TCN.7.shared_block.4.weight", "masker.TCN.7.shared_block.5.gamma", "masker.TCN.7.shared_block.5.beta", "masker.TCN.7.res_conv.weight", "masker.TCN.7.res_conv.bias", "masker.TCN.7.skip_conv.weight", "masker.TCN.7.skip_conv.bias", "masker.TCN.8.shared_block.0.weight", "masker.TCN.8.shared_block.0.bias", "masker.TCN.8.shared_block.1.weight", "masker.TCN.8.shared_block.2.gamma", "masker.TCN.8.shared_block.2.beta", "masker.TCN.8.shared_block.3.weight", "masker.TCN.8.shared_block.3.bias", "masker.TCN.8.shared_block.4.weight", "masker.TCN.8.shared_block.5.gamma", "masker.TCN.8.shared_block.5.beta", "masker.TCN.8.res_conv.weight", "masker.TCN.8.res_conv.bias", "masker.TCN.8.skip_conv.weight", "masker.TCN.8.skip_conv.bias", "masker.TCN.9.shared_block.0.weight", "masker.TCN.9.shared_block.0.bias", "masker.TCN.9.shared_block.1.weight", "masker.TCN.9.shared_block.2.gamma", "masker.TCN.9.shared_block.2.beta", "masker.TCN.9.shared_block.3.weight", "masker.TCN.9.shared_block.3.bias", "masker.TCN.9.shared_block.4.weight", "masker.TCN.9.shared_block.5.gamma", "masker.TCN.9.shared_block.5.beta", "masker.TCN.9.res_conv.weight", "masker.TCN.9.res_conv.bias", "masker.TCN.9.skip_conv.weight", "masker.TCN.9.skip_conv.bias", "masker.TCN.10.shared_block.0.weight", "masker.TCN.10.shared_block.0.bias", "masker.TCN.10.shared_block.1.weight", "masker.TCN.10.shared_block.2.gamma", "masker.TCN.10.shared_block.2.beta", "masker.TCN.10.shared_block.3.weight", "masker.TCN.10.shared_block.3.bias", "masker.TCN.10.shared_block.4.weight", "masker.TCN.10.shared_block.5.gamma", "masker.TCN.10.shared_block.5.beta", "masker.TCN.10.res_conv.weight", "masker.TCN.10.res_conv.bias", "masker.TCN.10.skip_conv.weight", "masker.TCN.10.skip_conv.bias", "masker.TCN.11.shared_block.0.weight", "masker.TCN.11.shared_block.0.bias", "masker.TCN.11.shared_block.1.weight", "masker.TCN.11.shared_block.2.gamma", "masker.TCN.11.shared_block.2.beta", "masker.TCN.11.shared_block.3.weight", "masker.TCN.11.shared_block.3.bias", "masker.TCN.11.shared_block.4.weight", "masker.TCN.11.shared_block.5.gamma", "masker.TCN.11.shared_block.5.beta", "masker.TCN.11.res_conv.weight", "masker.TCN.11.res_conv.bias", "masker.TCN.11.skip_conv.weight", "masker.TCN.11.skip_conv.bias", "masker.TCN.12.shared_block.0.weight", "masker.TCN.12.shared_block.0.bias", "masker.TCN.12.shared_block.1.weight", "masker.TCN.12.shared_block.2.gamma", "masker.TCN.12.shared_block.2.beta", "masker.TCN.12.shared_block.3.weight", "masker.TCN.12.shared_block.3.bias", "masker.TCN.12.shared_block.4.weight", "masker.TCN.12.shared_block.5.gamma", "masker.TCN.12.shared_block.5.beta", "masker.TCN.12.res_conv.weight", "masker.TCN.12.res_conv.bias", "masker.TCN.12.skip_conv.weight", "masker.TCN.12.skip_conv.bias", "masker.TCN.13.shared_block.0.weight", "masker.TCN.13.shared_block.0.bias", "masker.TCN.13.shared_block.1.weight", "masker.TCN.13.shared_block.2.gamma", "masker.TCN.13.shared_block.2.beta", "masker.TCN.13.shared_block.3.weight", "masker.TCN.13.shared_block.3.bias", "masker.TCN.13.shared_block.4.weight", "masker.TCN.13.shared_block.5.gamma", "masker.TCN.13.shared_block.5.beta", "masker.TCN.13.res_conv.weight", "masker.TCN.13.res_conv.bias", "masker.TCN.13.skip_conv.weight", "masker.TCN.13.skip_conv.bias", "masker.mask_net.0.weight", "masker.mask_net.1.weight", "masker.mask_net.1.bias", "decoder.filterbank._filters".
Unexpected key(s) in state_dict: "epoch", "global_step", "pytorch-lightning_version", "state_dict", "callbacks", "optimizer_states", "lr_schedulers", "native_amp_scaling_state", "training_config".

mpariente Oct 12, 2021
Maintainer

I meant with the model saving part on the same code example. So that the example is standalone.

cemox35 Oct 12, 2021
Author

I don't use any extra model saving code. Trainer is automatically logging these. Should I write my own model saving code? If yes, how can I do that?

cemox35 · 2021-10-13T14:42:57Z

cemox35
Oct 13, 2021
Author

Oh, I think, I need a use System instead ConvTasNet. However, I got another error.

This is my code:
`MODEL_PATH = ["lightning_logs/version_53/checkpoints/epoch=4-step=1444.ckpt"] # write model path

i = 5

train_loader, val_loader = load_data()

model = ConvTasNet(
    n_src=SPEAKER,
    n_filters=N[i % 3],
    kernel_size=L[i % 3],
    bn_chan=B[i % 3],
    skip_chan=Sc[i % 3],
    hid_chan=H[i % 3],
    conv_kernel_size=P[i % 3],
    n_blocks=X[i % 3],
    n_repeats=R[i % 3],
    sample_rate=SAMPLE_RATE,
    fb_name=ENCODER[i]
).double()

if torch.cuda.is_available():
    device = "cuda"
else:
    device = "cpu"
print(f"Using {device}")

model.to(device, non_blocking=True)

loss = PITLossWrapper(singlesrc_neg_sisdr, pit_from="pw_pt")
# loss = PITLossWrapper(pairwise_neg_sisdr, pit_from="pt_mtx")

optimizer = optim.Adam(model.parameters(), lr=1e-3)

system = System(model, optimizer, loss, train_loader, val_loader)

system = system.load_from_checkpoint(MODEL_PATH[0])

print(system)`

the error is: TypeError: __init__() missing 4 required positional arguments: 'model', 'optimizer', 'loss_func', and 'train_loader'

1 reply

cemox35 Oct 13, 2021
Author

Well, I solved my problem with using Trainer.

Full example code:

  MODEL_PATH = ["lightning_logs/version_53/checkpoints/epoch=4-step=1444.ckpt"] # write model path
  i = 5
  
  train_loader, val_loader = load_data()
  
  model = ConvTasNet(
    n_src=SPEAKER,
    n_filters=N[i % 3],
    kernel_size=L[i % 3],
    bn_chan=B[i % 3],
    skip_chan=Sc[i % 3],
    hid_chan=H[i % 3],
    conv_kernel_size=P[i % 3],
    n_blocks=X[i % 3],
    n_repeats=R[i % 3],
    sample_rate=SAMPLE_RATE,
    fb_name=ENCODER[i]
  ).double()
  
  if torch.cuda.is_available():
    device = "cuda"
  else:
    device = "cpu"
  print(f"Using {device}")
  
  model.to(device, non_blocking=True)
  
  loss = PITLossWrapper(singlesrc_neg_sisdr, pit_from="pw_pt")
  
  optimizer = optim.Adam(model.parameters(), lr=1e-3)
  
  system = System(model, optimizer, loss, train_loader, val_loader)
  
  trainer = Trainer(max_epochs=EPOCHS,
  resume_from_checkpoints=MODEL_PATH[0],
  gpus=1)
  
  trainer.fit(system)
  
  model.float().separate("s1.wav", resample=True)

After all, I can get the s1_est.wav file.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ConvTasNet Saving / Loading Model #563

{{title}}

Replies: 2 comments 6 replies

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

{{title}}

Select a reply

ConvTasNet Saving / Loading Model #563

cemox35 Oct 10, 2021

Replies: 2 comments · 6 replies

mpariente Oct 10, 2021 Maintainer

cemox35 Oct 10, 2021 Author

mpariente Oct 11, 2021 Maintainer

cemox35 Oct 12, 2021 Author

mpariente Oct 12, 2021 Maintainer

cemox35 Oct 12, 2021 Author

cemox35 Oct 13, 2021 Author

cemox35 Oct 13, 2021 Author

cemox35
Oct 10, 2021

Replies: 2 comments 6 replies

mpariente
Oct 10, 2021
Maintainer

cemox35 Oct 10, 2021
Author

mpariente Oct 11, 2021
Maintainer

cemox35 Oct 12, 2021
Author

mpariente Oct 12, 2021
Maintainer

cemox35 Oct 12, 2021
Author

cemox35
Oct 13, 2021
Author

cemox35 Oct 13, 2021
Author