Type error when trying run trainer.fit with tft

[jzicker]

• PyTorch-Forecasting version: 1.0
• PyTorch version: 2.0
• Python version:
• Operating System: running on google colab

Expected behavior

I executed code trainer.fit. It used to work and now I get a type error.

Actual behavior

I think it has to do with the april 10 release of pytorch-forecasting moving to pytorch 2.0

Code to reproduce the problem

trainer.fit(
tft,
train_dataloaders=train_dataloader,
val_dataloaders=val_dataloader
)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
[<ipython-input-57-127e7bdaac70>](https://localhost:8080/#) in <cell line: 1>()
----> 1 trainer.fit(
      2     tft,
      3     train_dataloaders=train_dataloader,
      4     val_dataloaders=val_dataloader
      5 )

1 frames
[/usr/local/lib/python3.9/dist-packages/pytorch_lightning/utilities/compile.py](https://localhost:8080/#) in _maybe_unwrap_optimized(model)
    123     if isinstance(model, pl.LightningModule):
    124         return model
--> 125     raise TypeError(
    126         f"`model` must be a `LightningModule` or `torch._dynamo.OptimizedModule`, got `{type(model).__qualname__}`"
    127     )

TypeError: `model` must be a `LightningModule` or `torch._dynamo.OptimizedModule`, got `TemporalFusionTransformer`

Paste the command(s) you ran and the output. Including a link to a colab notebook will speed up issue resolution.
If there was a crash, please include the traceback here.
The code used to initialize the TimeSeriesDataSet and model should be also included.

[MorrisHsieh3059]

same problem!!!!!

[jzicker]

I can't post all the code or give access to the notebook. But here are relevant portions.

early_stop_callback = EarlyStopping(monitor="val_loss", min_delta=MIN_DELTA,
patience=PATIENCE, verbose=False, mode="min")
checkpoint_callback = ModelCheckpoint(monitor="val_loss", mode="min")
lr_logger = LearningRateMonitor()
logger = TensorBoardLogger("lightning_logs")

trainer = pl.Trainer(
max_epochs=MAX_EPOCHS,
accelerator='gpu',
devices="auto",
enable_model_summary=True,
gradient_clip_val=GRADIENT_CLIP_VAL,
callbacks=[lr_logger, early_stop_callback, checkpoint_callback],
logger=logger,
)

tft = TemporalFusionTransformer.from_dataset(
training,
learning_rate=LR,
hidden_size=HIDDEN_SIZE,
attention_head_size=ATTENTION_HEAD_SIZE,
dropout=DROPOUT,
hidden_continuous_size=HIDDEN_CONTINUOUS_SIZE,
output_size=7, # there are 7 quantiles by default: [0.02, 0.1, 0.25, 0.5, 0.75, 0.9, 0.98]
loss=QuantileLoss(),
log_interval=10,
log_val_interval = 10,
reduce_on_plateau_patience=REDUCE_ON_PLATEAU_PATIENCE,
lstm_layers = 2
)

[rmagesh148]

same error!

[pjwu1997]

Same error here

[vikolss]

I get the same error with a custom model subclassed from TFT

[shuya-li-wmg]

Had the same issue but change pytorch_lighting to lightning.pytorch and resolved.

[MariaSky7]

still have this problem also with lightning.pytorch

[Smendowski]

I suggest you to perform the following changes:

import lightning.pytorch as pl # Instead of import pytorch_lightning as pl

# [...]
# Also the way callbacks are imported should be slightly modified.
from lightning.pytorch.callbacks.early_stopping import EarlyStopping
from lightning.pytorch.callbacks import LearningRateMonitor

Everything is up to date in the usage examples section:
https://github.com/jdb78/pytorch-forecasting

[jzicker]

@Smendowski Thank you so much! you unblocked me

[DevSoftChuck]

Hey guys! I'm still having the same issue, what I am doing wrong?

from pytorch_forecasting import TimeSeriesDataSet
import torch
from pytorch_forecasting import Baseline, TemporalFusionTransformer, TimeSeriesDataSet
from pytorch_forecasting.data import GroupNormalizer
from pytorch_forecasting.metrics import MAE, SMAPE, PoissonLoss, QuantileLoss
from pytorch_forecasting.models.temporal_fusion_transformer.tuning import optimize_hyperparameters
import lightning.pytorch as pl
from lightning.pytorch.callbacks.early_stopping import EarlyStopping
from lightning.pytorch.callbacks import LearningRateMonitor
from lightning.pytorch.loggers import TensorBoardLogger
from lightning.pytorch.tuner import Tuner

.
.
.

early_stop_callback = EarlyStopping(monitor="val_loss", min_delta=1e-4, patience=1, verbose=False, mode="min")
lr_logger = LearningRateMonitor()
trainer = pl.Trainer(
    max_epochs=100,
    accelerator="auto", 
    gradient_clip_val=0.1,
    limit_train_batches=30, 
    callbacks=[lr_logger, early_stop_callback],
    logger=TensorBoardLogger("lightning_logs")
)

tft = TemporalFusionTransformer.from_dataset(
    training,
    hidden_size=32,
    attention_head_size=1,
    dropout=0.1,
    hidden_continuous_size=16,
    loss=QuantileLoss(),
    log_interval=2,
    learning_rate=0.03,
    reduce_on_plateau_patience=4
)

res = Tuner(trainer).lr_find(
    tft, 
    train_dataloaders=train_dataloader, 
    val_dataloaders=val_dataloader, 
    early_stop_threshold=1000.0, 
    max_lr=10.0,
    min_lr=1e-6,
)

This is the error message:

TypeError: `model` must be a `LightningModule`, got `TemporalFusionTransformer`

[grosestq]

@DevSoftChuck Your packages may be out of date (torch etc.)?

[MBristle]

I had the same problem. It must have something to do with the breaking changes of torch to version 2.0.0 and pytorch-forecasting to version 1.0.0 (as mentioned by @Smendowski above).

As an intermediate solution, I stuck with pytorch-forecasting==0.10.3 torch==1.13.1, which works fine.

Further adjustments might be necessary for the upgrade. E.g., the "gpus" argument of the trainer is not valid anymore (https://lightning.ai/docs/pytorch/latest/upgrade/from_1_9.html). So when considering the upgrade, one must most likely adapt parts of the code.

[DevSoftChuck]

@grosestq I was using an outdated version of torch.

pytorch-forecasting==0.10.3
pytorch-lightning==1.9.4
torch==1.13.1

then I upgraded some packages to the following version and that problem was solved:

pytorch-forecasting==1.0.0
pytorch-lightning==2.0.1.post0
torch==2.0.0

Thanks for your help guys!!

[chaoss16]

Changing all the lightning.pytorch to pytorch_lightning works for me! Like this:

import pytorch_lightning as pl
from pytorch_lightning.callbacks import EarlyStopping, LearningRateMonitor
from pytorch_lightning.loggers import TensorBoardLogger