Pytorch-lightning: empty meta_tags.csv

[soomiles]

Before asking:

1. search the issues.
2. search the docs.

If you still can't find what you need:

What is your question?

For loading data, I should write down path for 'meta_tags.csv'.
However, in the 'meta_tags.csv', there is nothing.

Code

Please paste a code snippet if your question requires it!
Same get_started code with checkpoint callbacks

import os
import torch
from torch.nn import functional as F
from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
import torchvision.transforms as transforms

import pytorch_lightning as pl

class CoolSystem(pl.LightningModule):

    def __init__(self):
        super(CoolSystem, self).__init__()
        # not the best model...
        self.l1 = torch.nn.Linear(28 * 28, 10)

    def forward(self, x):
        return torch.relu(self.l1(x.view(x.size(0), -1)))

    def training_step(self, batch, batch_nb):
        # REQUIRED
        x, y = batch
        y_hat = self.forward(x)
        return {'loss': F.cross_entropy(y_hat, y)}

    def validation_step(self, batch, batch_nb):
        # OPTIONAL
        x, y = batch
        y_hat = self.forward(x)
        return {'val_loss': F.cross_entropy(y_hat, y)}

    def validation_end(self, outputs):
        # OPTIONAL
        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
        return {'avg_val_loss': avg_loss}

    def configure_optimizers(self):
        # REQUIRED
        # can return multiple optimizers and learning_rate schedulers
        return torch.optim.Adam(self.parameters(), lr=0.02)

    @pl.data_loader
    def tng_dataloader(self):
        # REQUIRED
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def val_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def test_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

from pytorch_lightning import Trainer
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping
from test_tube import Experiment

model = CoolSystem()

checkpoint_callback = ModelCheckpoint(
    filepath='./model_ckpt/weights6.ckpt',
    save_best_only=True,
    verbose=True,
    monitor='val_loss',
    mode='auto'
)

early_stopping = EarlyStopping(
    monitor='val_loss',
    patience=5,
    verbose=True,
    mode='auto'
)

exp = Experiment(save_dir=os.getcwd(), version=24)
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, gpus=[1], checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

What have you tried?

I tried on my model and this base tutorial code. But both of them are failed.

I just started using this API today, but I've faced a lot of difficulties..
Sometimes, I need to import exit from sys.
Sometimes, loss doesn't decrease. When I restart jupyter notebook multiple times, suddenly loss gets decreasing.
Sometimes, memory can't be allocated during training (about 81 epochs).
And I can't tag something like 'Losses/train_loss', 'Losses/valid_loss' for tensorboard by using test-tube... The error said 'loss' is referenced before assignments.

I can tolerate these difficulties because I can still progress my training.
But I don't have any solution with loading models from empty file.

What's your environment?

• conda version (no venv)
  no conda

• PyTorch version
  torch==1.2.0+cu92
  torchvision==0.4.0

• Lightning version
  pytorch-lightning==0.4.7

• Test-tube version
  test-tube==0.6.9

[williamFalcon]

filepath='./model_ckpt/weights6.ckpt',

should be a path

    filepath='./model_ckpt/',

[soomiles]

filepath='./model_ckpt/weights6.ckpt',

should be a path

    filepath='./model_ckpt/',

Oh, I see
Thanks!

[soomiles]

Hello, I tested today from your reply!
But I had a same error when I load the pretrained model.

model = CountingSystem()

checkpoint_callback = ModelCheckpoint(
    filepath='./ckpt/',
    save_best_only=True,
    verbose=True,
    monitor='avg_val_score',
    mode='max'
)

early_stopping = EarlyStopping(
    monitor='avg_val_score',
    patience=20,
    verbose=True,
    mode='max'
)

exp = Experiment(save_dir=os.getcwd(), version=25)
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, gpus=[0], checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

trainer.fit(model)

Then output is

Epoch 00089: avg_val_score did not improve
12it [00:17,  1.26it/s, avg_val_loss=11, avg_val_score=0.732, batch_nb=3, epoch=88, gpu=0, loss=2.839, v_nb=25]
Epoch 00089: early stopping

Then I loaded this pretrained model

pretrained_model = pl.LightningModule.load_from_metrics(
    weights_path='./ckpt/',
    tags_csv='./default/version_25/meta_tags.csv',
    on_gpu=True,
    map_location=None
)

But, I got this error.

IsADirectoryError: [Errno 21] Is a directory: './ckpt/'

And then I tried with including file name

pretrained_model = pl.LightningModule.load_from_metrics(
    weights_path='./ckpt/_ckpt_epoch_69.ckpt',
    tags_csv='./default/version_25/meta_tags.csv',
    on_gpu=True,
    map_location=None
)

But I got this error.

---

TypeError Traceback (most recent call last)
in
3 tags_csv='./default/version_25/meta_tags.csv',
4 on_gpu=True,
----> 5 map_location=None
6 )

/opt/conda/lib/python3.6/site-packages/pytorch_lightning/root_module/root_module.py in load_from_metrics(cls, weights_path, tags_csv, on_gpu, map_location)
132
133 # load the state_dict on the model automatically
--> 134 model = cls(hparams)
135 model.load_state_dict(checkpoint['state_dict'])
136

/opt/conda/lib/python3.6/site-packages/pytorch_lightning/root_module/root_module.py in __init__(self, args, *kwargs)
11
12 def __init__(self, args, *kwargs):
---> 13 super(LightningModule, self).__init__(args, *kwargs)
14
15 self.dtype = torch.FloatTensor

TypeError: __init__() takes 1 positional argument but 2 were given

Would you give me some advice?

[williamFalcon]

You need to load your module, not the general LightningModule.

Instead of

pl.LightningModule.load_from_metrics(...

Do this:

CountingSystem.load_from_metrics( 

[soomiles]

I changed code, but I got the same error message.

pretrained_model = CountingSystem.load_from_metrics(
    weights_path='./ckpt/_ckpt_epoch_69.ckpt',
    tags_csv='./default/version_25/meta_tags.csv',
    on_gpu=True,
    map_location={'cuda:0':'cuda:1'}
)

---

TypeError Traceback (most recent call last)
in
3 tags_csv='./default/version_25/meta_tags.csv',
4 on_gpu=True,
----> 5 map_location={'cuda:0':'cuda:1'}
6 )

/opt/conda/lib/python3.6/site-packages/pytorch_lightning/root_module/root_module.py in load_from_metrics(cls, weights_path, tags_csv, on_gpu, map_location)
132
133 # load the state_dict on the model automatically
--> 134 model = cls(hparams)
135 model.load_state_dict(checkpoint['state_dict'])
136

TypeError: __init__() takes 1 positional argument but 2 were given

BTW, is this correct way to map gpu location?
I couldn't find an example which uses this load_from_metrics, so I'm confused

[williamFalcon]

https://discuss.pytorch.org/t/on-a-cpu-device-how-to-load-checkpoint-saved-on-gpu-device/349

[soomiles]

pretrained_model = CountingSystem.load_from_metrics(
    weights_path='./ckpt/_ckpt_epoch_69.ckpt',
    tags_csv='./default/version_25/meta_tags.csv',
    on_gpu=True,
    map_location='cuda:1'
)

---

TypeError Traceback (most recent call last)
in
3 tags_csv='./default/version_25/meta_tags.csv',
4 on_gpu=True,
----> 5 map_location='cuda:1'
6 )

/opt/conda/lib/python3.6/site-packages/pytorch_lightning/root_module/root_module.py in load_from_metrics(cls, weights_path, tags_csv, on_gpu, map_location)
132
133 # load the state_dict on the model automatically
--> 134 model = cls(hparams)
135 model.load_state_dict(checkpoint['state_dict'])
136

TypeError: __init__() takes 1 positional argument but 2 were given

Still same error..

And this is meta_tags.csv file.
Is it normal that this file has nothing?

In metric.csv, there are training logs

[williamFalcon]

no. the file should have something. it might have to do with you using ipynb. i believe the logger doesn’t support that atm

[williamFalcon]

fit and load without ipynb

[williamFalcon]

we have plenty of tests that save and load models. so it’s likely the ipynb

[soomiles]

I see.. then I can't save model with jupyter notebook..
I'll test with .py and reopen this issue if there is a problem.

[soomiles]

I'm sad to reopen this issue..
I got the same error from .py execution.

And in this time, I only used tutorial code + (checkpoint & early stopping callbacks)

Path setting is also okay I think

meta_tags.csv is still empty.

This is changed part from tutorial code.

model = CoolSystem()

checkpoint_callback = ModelCheckpoint(
    filepath='base_ckpt',
    save_best_only=True,
    verbose=True,
    monitor='avg_val_loss',
    mode='auto'
)

early_stopping = EarlyStopping(
    monitor='avg_val_loss',
    patience=5,
    verbose=True,
    mode='auto'
)

exp = Experiment(save_dir=os.getcwd(), version=0)
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, gpus=[0], checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

trainer.fit(model)

pretrained_model = CoolSystem.load_from_metrics(
    weights_path='./base_ckpt/_ckpt_epoch_6.ckpt',
    tags_csv='./default/version_0/meta_tags.csv',
    on_gpu=True,
    map_location=None
)

Traceback (most recent call last):
File "/tmp/pycharm_project_164/lightning_basecode.py", line 89, in
map_location=None
File "/opt/conda/lib/python3.6/site-packages/pytorch_lightning/root_module/root_module.py", line 134, in load_from_metrics
model = cls(hparams)
TypeError: __init__() takes 1 positional argument but 2 were given

[williamFalcon]

enable auto_save to true or do

exp.save()

otherwise, tags won’t be saved

so...

exp = Experiment(save_dir=os.getcwd(), version=0)
exp.save()  # <------- here
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, gpus=[0], checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

trainer.fit(model)

[williamFalcon]

maybe auto_save makes sense to default to true?

[soomiles]

exp.save()
or
exp = Experiment(save_dir=os.getcwd(), autosave=True, version=0)
both of them don't work for me.

I dig into Experiment class in test-tube to figure out where a tag is not saved, but I couldn't get it..

meta_tags.csv is still empty

It'd be nice if meta tags are saved automatically as default.

[williamFalcon]

ok. is this exactly the code you’re using? i’ll run it locally to see what’s happening.

[soomiles]

import os
import torch
from torch.nn import functional as F
from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
import torchvision.transforms as transforms
import pytorch_lightning as pl

class CoolSystem(pl.LightningModule):

    def __init__(self):
        super(CoolSystem, self).__init__()
        # not the best model...
        self.l1 = torch.nn.Linear(28 * 28, 10)

    def forward(self, x):
        return torch.relu(self.l1(x.view(x.size(0), -1)))

    def training_step(self, batch, batch_nb):
        # REQUIRED
        x, y = batch
        y_hat = self.forward(x)
        return {'loss': F.cross_entropy(y_hat, y)}

    def validation_step(self, batch, batch_nb):
        # OPTIONAL
        x, y = batch
        y_hat = self.forward(x)
        return {'val_loss': F.cross_entropy(y_hat, y)}

    def validation_end(self, outputs):
        # OPTIONAL
        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
        return {'avg_val_loss': avg_loss}

    def configure_optimizers(self):
        # REQUIRED
        # can return multiple optimizers and learning_rate schedulers
        return torch.optim.Adam(self.parameters(), lr=0.02)

    @pl.data_loader
    def tng_dataloader(self):
        # REQUIRED
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def val_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def test_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

from pytorch_lightning import Trainer
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping
from test_tube import Experiment

model = CoolSystem()

checkpoint_callback = ModelCheckpoint(
    filepath='base_ckpt',
    save_best_only=True,
    verbose=True,
    monitor='avg_val_loss',
    mode='auto'
)

early_stopping = EarlyStopping(
    monitor='avg_val_loss',
    patience=5,
    verbose=True,
    mode='auto'
)

exp = Experiment(save_dir=os.getcwd(), version=1)
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, gpus=[0], checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

trainer.fit(model)
exp.save()
pretrained_model = CoolSystem.load_from_metrics(
    weights_path='./base_ckpt/_ckpt_epoch_6.ckpt',
    tags_csv='./default/version_0/meta_tags.csv',
    on_gpu=True,
    map_location=None
)

Yes this is the full code which I'm using.
I just added callbacks based on the tutorial code

[williamFalcon]

I see it now...
meta_tags.csv is not empty

(ddt) mbp:$ cat test/version_0/meta_tags.csv
key,value

A few things are wrong with your code:

1. you load tags path version_0 but set the experiment to version 1.
2. to load from metrics the module expects and hparams argument. Otherwise there are no tags to use for loading.

this works for me:

import os
import torch
from torch.nn import functional as F
from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
import torchvision.transforms as transforms
import pytorch_lightning as pl

class CoolSystem(pl.LightningModule):

    def __init__(self, hparams=None):
        super(CoolSystem, self).__init__()
        # not the best model...
        self.l1 = torch.nn.Linear(28 * 28, 10)

    def forward(self, x):
        return torch.relu(self.l1(x.view(x.size(0), -1)))

    def training_step(self, batch, batch_nb):
        # REQUIRED
        x, y = batch
        y_hat = self.forward(x)
        return {'loss': F.cross_entropy(y_hat, y)}

    def validation_step(self, batch, batch_nb):
        # OPTIONAL
        x, y = batch
        y_hat = self.forward(x)
        return {'val_loss': F.cross_entropy(y_hat, y)}

    def validation_end(self, outputs):
        # OPTIONAL
        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
        return {'avg_val_loss': avg_loss}

    def configure_optimizers(self):
        # REQUIRED
        # can return multiple optimizers and learning_rate schedulers
        return torch.optim.Adam(self.parameters(), lr=0.02)

    @pl.data_loader
    def tng_dataloader(self):
        # REQUIRED
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def val_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

    @pl.data_loader
    def test_dataloader(self):
        # OPTIONAL
        return DataLoader(MNIST(os.getcwd(), train=True, download=True, transform=transforms.ToTensor()), batch_size=32)

from pytorch_lightning import Trainer
from pytorch_lightning.callbacks import ModelCheckpoint, EarlyStopping
from test_tube import Experiment

model = CoolSystem()

checkpoint_callback = ModelCheckpoint(
    filepath='base_ckpt',
    save_best_only=True,
    verbose=True,
    monitor='avg_val_loss',
    mode='auto'
)

early_stopping = EarlyStopping(
    monitor='avg_val_loss',
    patience=5,
    verbose=True,
    mode='auto'
)

exp = Experiment(save_dir=os.getcwd(), version=0)
trainer = Trainer(experiment=exp, max_nb_epochs=1000, train_percent_check=0.1, checkpoint_callback=checkpoint_callback, early_stop_callback=early_stopping)

# trainer.fit(model)
exp.save()
pretrained_model = CoolSystem.load_from_metrics(
    weights_path='./base_ckpt/_ckpt_epoch_2.ckpt',
    tags_csv='./default/version_0/meta_tags.csv',
    on_gpu=True,
    map_location=None
)
print('model loaded...')

[williamFalcon]

@Soo95 but your user experience highlights a good thing we need to fix.

1. We need to make load_from_metrics use an OPTIONAL tags_csv in the case where your model has no hparam arguments.

2. Expand on the model loading documentation

[dehoyosb]

Hello! Good Day.
It appears im having a similar issue. I have the next model:

class Predictor(pl.LightningModule):
    def __init__(self, 
                nb_layers,
                 nb_lstm_units = 100, 
                 input_dim = 10, 
                 batch_size = 256, 
                 bilstm = False, 
                 dropout = 0.2, 
                 hidden_size = 100,
                 encoder_discrete = [],
                 embedding_dim = 10):

        super(Predictor, self).__init__()


        self.input_dim = input_dim
        self.nb_lstm_layers = nb_layers
        self.nb_lstm_units = nb_lstm_units
        self.batch_size = batch_size
        self.bilstm = bilstm
        self.dropout = dropout
        self.hidden_size = hidden_size
        self.unique_length_encoders = encoder_discrete
        self.embedding_dim = embedding_dim
        self.loss = RMSELoss()
        self.leaky_relu = nn.LeakyReLU(negative_slope=0.1)

        self.__build_model()

    def __build_model(self):

        self.embedding = nn.ModuleList([nn.Embedding(num_embeddings = length + 1, 
                                         embedding_dim = self.embedding_dim,
                                         padding_idx = 0) for length in self.unique_length_encoders])

        # Design LSTM
        self.lstm = torch.nn.LSTM(input_size = self.input_dim,
                                  hidden_size = self.nb_lstm_units,
                                  num_layers = self.nb_lstm_layers,
                                  batch_first=True,
                                  dropout = self.dropout if self.dropout and self.nb_lstm_layers > 1 else 0,
                                  bidirectional = self.bilstm).to(DEVICE)

        self.dense1 = torch.nn.Linear(self.nb_lstm_units*2 if self.bilstm else self.nb_lstm_units, 
                                      1).to(DEVICE)

        self.init_weigths()

    def init_hidden(self):
        hidden = torch.randn(self.nb_lstm_layers*2 if self.bilstm else self.nb_lstm_layers, 
                               self.batch_size, self.nb_lstm_units).to(DEVICE)
        cell = torch.randn(self.nb_lstm_layers*2 if self.bilstm else self.nb_lstm_layers, 
                               self.batch_size, self.nb_lstm_units).to(DEVICE)

        return (hidden, cell)

    def init_weigths(self):

        for name, param in self.lstm.named_parameters():
            if 'weight_hh' in name:
                torch.nn.init.orthogonal_(param)
            elif 'weight_ih' in name:
                torch.nn.init.xavier_uniform_(param)
            elif 'bias' in name:
                torch.nn.init.zeros_(param)
        torch.nn.init.xavier_uniform_(self.dense1.weight)
        print('weigths initializer: done!')

    def forward(self, X_discrete, X_continuous, X_lengths):

        (self.hidden, self.cell) = self.init_hidden()

        for i, emb in enumerate(self.embedding):
            if i == 0:
                X_discrete_emb = emb(X_discrete[:,:,i])
            else:
                X_discrete_emb = torch.cat((X_discrete_emb, emb(X_discrete[:,:,i])), 2)

        X = torch.cat((X_discrete_emb,X_continuous), 2)

        X = pack_padded_sequence(X, X_lengths, batch_first=True, enforce_sorted=False)

        X, (self.hidden, self.cell) = self.lstm(X, (self.hidden, self.cell))

        X, _ = pad_packed_sequence(X, batch_first=True, padding_value=0)

        hidden = self.hidden.view(self.nb_lstm_layers, 2, -1,
                                  self.nb_lstm_units)[-1] if self.bilstm else self.hidden[-1]
        hidden = hidden.contiguous()

        hidden = hidden.view(-1, self.nb_lstm_units*2 if self.bilstm else self.nb_lstm_units)

        hidden = self.dense1(hidden)

        return hidden

    def my_loss(self, y_hat, y):
        return self.loss(y_hat, y)

    def training_step(self, batch, batch_nb):
        (x_discrete, x_continuous, lengths, y) = batch
        y_hat = self.forward(x_discrete, x_continuous, lengths)
        loss = self.my_loss(y_hat, y)
        tensorboard_logs = {'train_loss': loss}
        return {'loss': loss, 'log': tensorboard_logs}

    def validation_step(self, batch, batch_nb):
        (x_discrete, x_continuous, lengths, y) = batch
        y_hat = self.forward(x_discrete, x_continuous, lengths)

        return {'val_loss': self.my_loss(y_hat, y)}

    def validation_end(self, outputs):
        avg_loss = torch.stack([x['val_loss'] for x in outputs]).mean()
        tensorboard_logs = {'val_loss': avg_loss}
        return {'avg_val_loss': avg_loss, 'log': tensorboard_logs}

    def configure_optimizers(self):
        return [torch.optim.Adam(self.parameters(), lr=0.0001, weight_decay=5e-4, amsgrad = True)]

    @pl.data_loader
    def train_dataloader(self):
        return DataLoader(train_dataset, batch_size = self.batch_size, shuffle = True, collate_fn = my_collate)
    @pl.data_loader
    def val_dataloader(self):
        return DataLoader(val_dataset, batch_size = self.batch_size, shuffle = True, collate_fn = my_collate)

I have already trained the model, and i wasnt able to make inference because of not enough GPU memory.

Therefore a restarted my kernel, and decided to load the model again for inference with the next few lines of code:

model = Predictor.load_from_metrics(
                                weights_path='/lightning_logs/version_50/checkpoints/_ckpt_epoch_50.ckpt',
                                tags_csv="/lightning_logs/version_50/meta_tags.csv",
                            )

Im getting the next error:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-68-aa919fee611f> in <module>
      1 model = HitsPredictor.load_from_metrics(
      2                                 weights_path='/home/daniel/trivago_challenge2/lightning_logs/version_50/checkpoints/_ckpt_epoch_50.ckpt',
----> 3                                 tags_csv="/home/daniel/trivago_challenge2/lightning_logs/version_50/meta_tags_old.csv",
      4                             )
      5 

/opt/conda/envs/nlp/lib/python3.7/site-packages/pytorch_lightning/root_module/root_module.py in load_from_metrics(cls, weights_path, tags_csv)
    152 
    153         # load the state_dict on the model automatically
--> 154         model = cls(hparams)
    155         model.load_state_dict(checkpoint['state_dict'])
    156 

<ipython-input-67-161a6c83168a> in __init__(self, nb_layers, nb_lstm_units, input_dim, batch_size, bilstm, dropout, hidden_size, encoder_discrete, embedding_dim)
     46 
     47         # Function that builds the actual model
---> 48         self.__build_model()
     49 
     50     def __build_model(self):

<ipython-input-67-161a6c83168a> in __build_model(self)
     67                                   batch_first=True,
     68                                   # Appliying dropout only if we have more than 1 layer in the network
---> 69                                   dropout = self.dropout if self.dropout and self.nb_lstm_layers > 1 else 0,
     70                                   # Bidirectional if i choose so in the input.
     71                                   bidirectional = self.bilstm).to(DEVICE)

TypeError: '>' not supported between instances of 'Namespace' and 'int'

From what is have been testing, it seams that my file meta_data.csv file only has the headers key and value, but it didnt save any thing regarding the input parameters that i chose at the beggining of training. So, the first question is why it didnt save them? and the second one is: is there a way in which i can load those parameters in any way, so i can use the model?. This model training took almost 18 hours, and i dont have time to train it again right now.

Any help would be much appreciated