TorchFlare is a simple, beginner-friendly and an easy-to-use PyTorch Framework to train your models with ease. It provides an almost Keras-like experience for training your models with all the callbacks, metrics, etc
- A high-level module for Keras-like training.
- Flexibility to write custom training and validation loops for advanced use cases.
- Off-the-shelf Pytorch style Datasets/Dataloaders for standard tasks such as Image classification, Image segmentation, Text Classification, etc
- Callbacks for model checkpoints, early stopping, and much more!
- TorchFlare uses powerful torchmetrics in the backend for metric computations!
- Reduction of the boiler plate code required for training your models.
- Create interactive UI for model prototyping and POC
Currently, TorchFlare supports CPU and GPU training. DDP and TPU support will be coming soon!
pip install torchflare
The Documentation is available here
The core idea around TorchFlare is the Experiment class. It handles all the internal stuff like boiler plate code for training, calling callbacks,metrics,etc. The only thing you need to focus on is creating you PyTorch Model.
Also, there are off-the-shelf pytorch style datasets/dataloaders available for standard tasks, so that you don't have to worry about creating Pytorch Datasets/Dataloaders.
Here is an easy-to-understand example to show how Experiment class works.
import torch
import torchmetrics
import torch.nn as nn
from torchflare.experiments import Experiment, ModelConfig
import torchflare.callbacks as cbs
# Some dummy dataloaders
train_dl = SomeTrainingDataloader()
valid_dl = SomeValidationDataloader()
test_dl = SomeTestingDataloader()Create a pytorch Model
class Net(nn.Module):
def __init__(self, n_classes, p_dropout):
super().__init__()
self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
self.conv2_drop = nn.Dropout2d(p=p_dropout)
self.fc1 = nn.Linear(320, 50)
self.fc2 = nn.Linear(50, n_classes)
def forward(self, x):
x = F.relu(F.max_pool2d(self.conv1(x), 2))
x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
x = x.view(-1, 320)
x = F.relu(self.fc1(x))
x = F.dropout(x, training=self.training)
x = self.fc2(x)
return xDefine callbacks and metrics
metric_list = [
torchmetrics.Accuracy(num_classes=num_classes)
]
callbacks = [
cbs.EarlyStopping(monitor="val_accuracy", mode="max"),
cbs.ModelCheckpoint(monitor="val_accuracy"),
cbs.ReduceLROnPlateau(mode="max", patience=2),
]Define Model Configuration
#Defining Model Config for experiment.
config = ModelConfig(
nn_module=Net,
module_params={"n_classes": 10, "p_dropout": 0.3},
optimizer="Adam",
optimizer_params={"lr": 3e-4},
criterion="cross_entropy",
)Define your experiment
# Set some constants for training
exp = Experiment(
num_epochs=5,
fp16=False,
device="cuda",
seed=42,
)
exp.compile_experiment(
model_config=config,
callbacks=callbacks,
metrics=metric_list,
main_metrics="accuracy",
)
# Run your experiment with training dataloader and validation dataloader.
exp.fit_loader(train_dl=train_dl, valid_dl=valid_dl)For inference, you can use infer method, which yields output per batch. You can use it as follows
outputs = []
for op in exp.predict_on_loader(
test_loader=test_dl, path_to_model="./models/model.bin", device="cuda"
):
op = some_post_process_function(op)
outputs.extend(op)If you want to access your experiments history or get as a dataframe. You can do it as follows.
history = exp.history # This will return a dict
exp.get_logs() # This will return a dataframe constructed from model-history.- Image Classification using TorchFlare on MNIST dataset.
- Text Classification using Tiny Bert on IMDB dataset
- Variational Auto-encoders to generate MNIST dataset.
- Train DCGANS to generate MNIST data.
- Self Supervised learning using Bootstrap your own latent(BYOL)
To contribute please refer to Contributing Guide
- Atharva Phatak - Atharva-Phatak
Please use this bibtex if you want to cite this repository in your publications:
@misc{TorchFlare,
author = {Atharva Phatak},
title = {TorchFlare: Easy model training and experimentation.},
year = {2020},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/Atharva-Phatak/torchflare}},
}