Finetune Transformers Models with PyTorch Lightning¶
Author: PL team
License: CC BY-SA
Generated: 2021-08-31T13:56:12.832145
This notebook will use HuggingFace’s datasets library to get data, which will be wrapped in a LightningDataModule. Then, we write a class to perform text classification on any dataset from the GLUE Benchmark. (We just show CoLA and MRPC due to constraint on compute/disk)
Open in 
Give us a ⭐ on Github | Check out the documentation | Join us on Slack
Setup¶
This notebook requires some packages besides pytorch-lightning.
[1]:
! pip install --quiet "datasets" "scipy" "torchmetrics>=0.3" "transformers" "scikit-learn" "torch>=1.6, <1.9" "pytorch-lightning>=1.3"
[2]:
from datetime import datetime
from typing import Optional
import datasets
import torch
from pytorch_lightning import LightningDataModule, LightningModule, Trainer, seed_everything
from torch.utils.data import DataLoader
from transformers import (
AdamW,
AutoConfig,
AutoModelForSequenceClassification,
AutoTokenizer,
get_linear_schedule_with_warmup,
)
AVAIL_GPUS = min(1, torch.cuda.device_count())
Training BERT with Lightning¶
Lightning DataModule for GLUE¶
[3]:
class GLUEDataModule(LightningDataModule):
task_text_field_map = {
"cola": ["sentence"],
"sst2": ["sentence"],
"mrpc": ["sentence1", "sentence2"],
"qqp": ["question1", "question2"],
"stsb": ["sentence1", "sentence2"],
"mnli": ["premise", "hypothesis"],
"qnli": ["question", "sentence"],
"rte": ["sentence1", "sentence2"],
"wnli": ["sentence1", "sentence2"],
"ax": ["premise", "hypothesis"],
}
glue_task_num_labels = {
"cola": 2,
"sst2": 2,
"mrpc": 2,
"qqp": 2,
"stsb": 1,
"mnli": 3,
"qnli": 2,
"rte": 2,
"wnli": 2,
"ax": 3,
}
loader_columns = [
"datasets_idx",
"input_ids",
"token_type_ids",
"attention_mask",
"start_positions",
"end_positions",
"labels",
]
def __init__(
self,
model_name_or_path: str,
task_name: str = "mrpc",
max_seq_length: int = 128,
train_batch_size: int = 32,
eval_batch_size: int = 32,
**kwargs,
):
super().__init__()
self.model_name_or_path = model_name_or_path
self.task_name = task_name
self.max_seq_length = max_seq_length
self.train_batch_size = train_batch_size
self.eval_batch_size = eval_batch_size
self.text_fields = self.task_text_field_map[task_name]
self.num_labels = self.glue_task_num_labels[task_name]
self.tokenizer = AutoTokenizer.from_pretrained(self.model_name_or_path, use_fast=True)
def setup(self, stage: str):
self.dataset = datasets.load_dataset("glue", self.task_name)
for split in self.dataset.keys():
self.dataset[split] = self.dataset[split].map(
self.convert_to_features,
batched=True,
remove_columns=["label"],
)
self.columns = [c for c in self.dataset[split].column_names if c in self.loader_columns]
self.dataset[split].set_format(type="torch", columns=self.columns)
self.eval_splits = [x for x in self.dataset.keys() if "validation" in x]
def prepare_data(self):
datasets.load_dataset("glue", self.task_name)
AutoTokenizer.from_pretrained(self.model_name_or_path, use_fast=True)
def train_dataloader(self):
return DataLoader(self.dataset["train"], batch_size=self.train_batch_size)
def val_dataloader(self):
if len(self.eval_splits) == 1:
return DataLoader(self.dataset["validation"], batch_size=self.eval_batch_size)
elif len(self.eval_splits) > 1:
return [DataLoader(self.dataset[x], batch_size=self.eval_batch_size) for x in self.eval_splits]
def test_dataloader(self):
if len(self.eval_splits) == 1:
return DataLoader(self.dataset["test"], batch_size=self.eval_batch_size)
elif len(self.eval_splits) > 1:
return [DataLoader(self.dataset[x], batch_size=self.eval_batch_size) for x in self.eval_splits]
def convert_to_features(self, example_batch, indices=None):
# Either encode single sentence or sentence pairs
if len(self.text_fields) > 1:
texts_or_text_pairs = list(zip(example_batch[self.text_fields[0]], example_batch[self.text_fields[1]]))
else:
texts_or_text_pairs = example_batch[self.text_fields[0]]
# Tokenize the text/text pairs
features = self.tokenizer.batch_encode_plus(
texts_or_text_pairs, max_length=self.max_seq_length, pad_to_max_length=True, truncation=True
)
# Rename label to labels to make it easier to pass to model forward
features["labels"] = example_batch["label"]
return features
You could use this datamodule with standalone PyTorch if you wanted…
[4]:
dm = GLUEDataModule("distilbert-base-uncased")
dm.prepare_data()
dm.setup("fit")
next(iter(dm.train_dataloader()))
Downloading and preparing dataset glue/mrpc (download: 1.43 MiB, generated: 1.43 MiB, post-processed: Unknown size, total: 2.85 MiB) to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mrpc/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad...
Dataset glue downloaded and prepared to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mrpc/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad. Subsequent calls will reuse this data.
Reusing dataset glue (/home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mrpc/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2184: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).
warnings.warn(
[4]:
{'attention_mask': tensor([[1, 1, 1, ..., 0, 0, 0],
[1, 1, 1, ..., 0, 0, 0],
[1, 1, 1, ..., 0, 0, 0],
...,
[1, 1, 1, ..., 0, 0, 0],
[1, 1, 1, ..., 0, 0, 0],
[1, 1, 1, ..., 0, 0, 0]]),
'input_ids': tensor([[ 101, 2572, 3217, ..., 0, 0, 0],
[ 101, 9805, 3540, ..., 0, 0, 0],
[ 101, 2027, 2018, ..., 0, 0, 0],
...,
[ 101, 1996, 2922, ..., 0, 0, 0],
[ 101, 6202, 1999, ..., 0, 0, 0],
[ 101, 16565, 2566, ..., 0, 0, 0]]),
'labels': tensor([1, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 1, 1, 0, 1,
1, 1, 0, 0, 1, 1, 1, 0])}
Transformer LightningModule¶
[5]:
class GLUETransformer(LightningModule):
def __init__(
self,
model_name_or_path: str,
num_labels: int,
task_name: str,
learning_rate: float = 2e-5,
adam_epsilon: float = 1e-8,
warmup_steps: int = 0,
weight_decay: float = 0.0,
train_batch_size: int = 32,
eval_batch_size: int = 32,
eval_splits: Optional[list] = None,
**kwargs,
):
super().__init__()
self.save_hyperparameters()
self.config = AutoConfig.from_pretrained(model_name_or_path, num_labels=num_labels)
self.model = AutoModelForSequenceClassification.from_pretrained(model_name_or_path, config=self.config)
self.metric = datasets.load_metric(
"glue", self.hparams.task_name, experiment_id=datetime.now().strftime("%d-%m-%Y_%H-%M-%S")
)
def forward(self, **inputs):
return self.model(**inputs)
def training_step(self, batch, batch_idx):
outputs = self(**batch)
loss = outputs[0]
return loss
def validation_step(self, batch, batch_idx, dataloader_idx=0):
outputs = self(**batch)
val_loss, logits = outputs[:2]
if self.hparams.num_labels >= 1:
preds = torch.argmax(logits, axis=1)
elif self.hparams.num_labels == 1:
preds = logits.squeeze()
labels = batch["labels"]
return {"loss": val_loss, "preds": preds, "labels": labels}
def validation_epoch_end(self, outputs):
if self.hparams.task_name == "mnli":
for i, output in enumerate(outputs):
# matched or mismatched
split = self.hparams.eval_splits[i].split("_")[-1]
preds = torch.cat([x["preds"] for x in output]).detach().cpu().numpy()
labels = torch.cat([x["labels"] for x in output]).detach().cpu().numpy()
loss = torch.stack([x["loss"] for x in output]).mean()
self.log(f"val_loss_{split}", loss, prog_bar=True)
split_metrics = {
f"{k}_{split}": v for k, v in self.metric.compute(predictions=preds, references=labels).items()
}
self.log_dict(split_metrics, prog_bar=True)
return loss
preds = torch.cat([x["preds"] for x in outputs]).detach().cpu().numpy()
labels = torch.cat([x["labels"] for x in outputs]).detach().cpu().numpy()
loss = torch.stack([x["loss"] for x in outputs]).mean()
self.log("val_loss", loss, prog_bar=True)
self.log_dict(self.metric.compute(predictions=preds, references=labels), prog_bar=True)
return loss
def setup(self, stage=None) -> None:
if stage != "fit":
return
# Get dataloader by calling it - train_dataloader() is called after setup() by default
train_loader = self.train_dataloader()
# Calculate total steps
tb_size = self.hparams.train_batch_size * max(1, self.trainer.gpus)
ab_size = self.trainer.accumulate_grad_batches * float(self.trainer.max_epochs)
self.total_steps = (len(train_loader.dataset) // tb_size) // ab_size
def configure_optimizers(self):
"""Prepare optimizer and schedule (linear warmup and decay)"""
model = self.model
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
"weight_decay": self.hparams.weight_decay,
},
{
"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)],
"weight_decay": 0.0,
},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=self.hparams.learning_rate, eps=self.hparams.adam_epsilon)
scheduler = get_linear_schedule_with_warmup(
optimizer,
num_warmup_steps=self.hparams.warmup_steps,
num_training_steps=self.total_steps,
)
scheduler = {"scheduler": scheduler, "interval": "step", "frequency": 1}
return [optimizer], [scheduler]
Training¶
CoLA¶
See an interactive view of the CoLA dataset in NLP Viewer
[6]:
seed_everything(42)
dm = GLUEDataModule(model_name_or_path="albert-base-v2", task_name="cola")
dm.setup("fit")
model = GLUETransformer(
model_name_or_path="albert-base-v2",
num_labels=dm.num_labels,
eval_splits=dm.eval_splits,
task_name=dm.task_name,
)
trainer = Trainer(max_epochs=1, gpus=AVAIL_GPUS)
trainer.fit(model, dm)
Global seed set to 42
Downloading and preparing dataset glue/cola (download: 368.14 KiB, generated: 596.73 KiB, post-processed: Unknown size, total: 964.86 KiB) to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/cola/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad...
Dataset glue downloaded and prepared to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/cola/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad. Subsequent calls will reuse this data.
Some weights of the model checkpoint at albert-base-v2 were not used when initializing AlbertForSequenceClassification: ['predictions.LayerNorm.weight', 'predictions.dense.bias', 'predictions.LayerNorm.bias', 'predictions.bias', 'predictions.decoder.weight', 'predictions.decoder.bias', 'predictions.dense.weight']
- This IS expected if you are initializing AlbertForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing AlbertForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of AlbertForSequenceClassification were not initialized from the model checkpoint at albert-base-v2 and are newly initialized: ['classifier.weight', 'classifier.bias']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
Reusing dataset glue (/home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/cola/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/core/datamodule.py:423: LightningDeprecationWarning: DataModule.setup has already been called, so it will not be called again. In v1.6 this behavior will change to always call DataModule.setup.
rank_zero_deprecation(
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1]
| Name | Type | Params
----------------------------------------------------------
0 | model | AlbertForSequenceClassification | 11.7 M
----------------------------------------------------------
11.7 M Trainable params
0 Non-trainable params
11.7 M Total params
46.740 Total estimated model params size (MB)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, val dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
/usr/local/lib/python3.9/dist-packages/sklearn/metrics/_classification.py:873: RuntimeWarning: invalid value encountered in double_scalars
mcc = cov_ytyp / np.sqrt(cov_ytyt * cov_ypyp)
Global seed set to 42
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, train dataloader, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
/usr/local/lib/python3.9/dist-packages/sklearn/metrics/_classification.py:873: RuntimeWarning: invalid value encountered in double_scalars
mcc = cov_ytyp / np.sqrt(cov_ytyt * cov_ypyp)
MRPC¶
See an interactive view of the MRPC dataset in NLP Viewer
[7]:
seed_everything(42)
dm = GLUEDataModule(
model_name_or_path="distilbert-base-cased",
task_name="mrpc",
)
dm.setup("fit")
model = GLUETransformer(
model_name_or_path="distilbert-base-cased",
num_labels=dm.num_labels,
eval_splits=dm.eval_splits,
task_name=dm.task_name,
)
trainer = Trainer(max_epochs=3, gpus=AVAIL_GPUS)
trainer.fit(model, dm)
Global seed set to 42
Reusing dataset glue (/home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mrpc/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2184: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).
warnings.warn(
Some weights of the model checkpoint at distilbert-base-cased were not used when initializing DistilBertForSequenceClassification: ['vocab_layer_norm.weight', 'vocab_projector.bias', 'vocab_transform.bias', 'vocab_layer_norm.bias', 'vocab_projector.weight', 'vocab_transform.weight']
- This IS expected if you are initializing DistilBertForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing DistilBertForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of DistilBertForSequenceClassification were not initialized from the model checkpoint at distilbert-base-cased and are newly initialized: ['classifier.weight', 'pre_classifier.bias', 'classifier.bias', 'pre_classifier.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
Reusing dataset glue (/home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mrpc/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/core/datamodule.py:423: LightningDeprecationWarning: DataModule.setup has already been called, so it will not be called again. In v1.6 this behavior will change to always call DataModule.setup.
rank_zero_deprecation(
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1]
| Name | Type | Params
--------------------------------------------------------------
0 | model | DistilBertForSequenceClassification | 65.8 M
--------------------------------------------------------------
65.8 M Trainable params
0 Non-trainable params
65.8 M Total params
263.132 Total estimated model params size (MB)
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, val dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
Global seed set to 42
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, train dataloader, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
MNLI¶
The MNLI dataset is huge, so we aren’t going to bother trying to train on it here.
We will skip over training and go straight to validation.
See an interactive view of the MRPC dataset in NLP Viewer
[8]:
dm = GLUEDataModule(
model_name_or_path="distilbert-base-cased",
task_name="mnli",
)
dm.setup("fit")
model = GLUETransformer(
model_name_or_path="distilbert-base-cased",
num_labels=dm.num_labels,
eval_splits=dm.eval_splits,
task_name=dm.task_name,
)
trainer = Trainer(gpus=AVAIL_GPUS, progress_bar_refresh_rate=20)
trainer.validate(model, dm.val_dataloader())
Downloading and preparing dataset glue/mnli (download: 298.29 MiB, generated: 78.65 MiB, post-processed: Unknown size, total: 376.95 MiB) to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mnli/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad...
Dataset glue downloaded and prepared to /home/AzDevOps_azpcontainer/.cache/huggingface/datasets/glue/mnli/1.0.0/dacbe3125aa31d7f70367a07a8a9e72a5a0bfeb5fc42e75c9db75b96da6053ad. Subsequent calls will reuse this data.
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/transformers/tokenization_utils_base.py:2184: FutureWarning: The `pad_to_max_length` argument is deprecated and will be removed in a future version, use `padding=True` or `padding='longest'` to pad to the longest sequence in the batch, or use `padding='max_length'` to pad to a max length. In this case, you can give a specific length with `max_length` (e.g. `max_length=45`) or leave max_length to None to pad to the maximal input size of the model (e.g. 512 for Bert).
warnings.warn(
Some weights of the model checkpoint at distilbert-base-cased were not used when initializing DistilBertForSequenceClassification: ['vocab_layer_norm.weight', 'vocab_projector.bias', 'vocab_transform.bias', 'vocab_layer_norm.bias', 'vocab_projector.weight', 'vocab_transform.weight']
- This IS expected if you are initializing DistilBertForSequenceClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing DistilBertForSequenceClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
Some weights of DistilBertForSequenceClassification were not initialized from the model checkpoint at distilbert-base-cased and are newly initialized: ['classifier.weight', 'pre_classifier.bias', 'classifier.bias', 'pre_classifier.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
GPU available: True, used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1]
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, val dataloader 0, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
/home/AzDevOps_azpcontainer/.local/lib/python3.9/site-packages/pytorch_lightning/trainer/data_loading.py:105: UserWarning: The dataloader, val dataloader 1, does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` (try 12 which is the number of cpus on this machine) in the `DataLoader` init to improve performance.
rank_zero_warn(
--------------------------------------------------------------------------------
DATALOADER:0 VALIDATE RESULTS
{'accuracy_matched': 0.3237901031970978,
'accuracy_mismatched': 0.31794142723083496,
'val_loss_matched': 1.104950189590454,
'val_loss_mismatched': 1.1043992042541504}
--------------------------------------------------------------------------------
DATALOADER:1 VALIDATE RESULTS
{'accuracy_matched': 0.3237901031970978,
'accuracy_mismatched': 0.31794142723083496,
'val_loss_matched': 1.104950189590454,
'val_loss_mismatched': 1.1043992042541504}
--------------------------------------------------------------------------------
[8]:
[{'val_loss_matched': 1.104950189590454,
'accuracy_matched': 0.3237901031970978,
'val_loss_mismatched': 1.1043992042541504,
'accuracy_mismatched': 0.31794142723083496},
{'val_loss_matched': 1.104950189590454,
'accuracy_matched': 0.3237901031970978,
'val_loss_mismatched': 1.1043992042541504,
'accuracy_mismatched': 0.31794142723083496}]
Congratulations - Time to Join the Community!¶
Congratulations on completing this notebook tutorial! If you enjoyed this and would like to join the Lightning movement, you can do so in the following ways!
Star Lightning on GitHub¶
The easiest way to help our community is just by starring the GitHub repos! This helps raise awareness of the cool tools we’re building.
Join our Slack!¶
The best way to keep up to date on the latest advancements is to join our community! Make sure to introduce yourself and share your interests in #general channel
Contributions !¶
The best way to contribute to our community is to become a code contributor! At any time you can go to Lightning or Bolt GitHub Issues page and filter for “good first issue”.
You can also contribute your own notebooks with useful examples !
Great thanks from the entire Pytorch Lightning Team for your interest !¶
