Metadata-Version: 2.1
Name: sconce
Version: 0.0.28
Summary: sconce: torch helper
Home-page: https://github.com/satabios/sconce
Download-URL: https://pypi.org/project/sconce/
Author: Sathyaprakash Narayanan
Author-email: Sathyaprakash Narayanan <snaray17@ucsc.edu>
Project-URL: code, https://github.com/satabios/sconce
Description-Content-Type: text/markdown
License-File: LICENSE.txt
Requires-Dist: certifi (==2023.5.7)
Requires-Dist: charset-normalizer (==3.2.0)
Requires-Dist: cmake (==3.27.0)
Requires-Dist: contourpy (==1.1.0)
Requires-Dist: cycler (==0.11.0)
Requires-Dist: filelock (==3.12.2)
Requires-Dist: fonttools (==4.41.0)
Requires-Dist: idna (==3.4)
Requires-Dist: Jinja2 (==3.1.2)
Requires-Dist: kiwisolver (==1.4.4)
Requires-Dist: lit (==16.0.6)
Requires-Dist: MarkupSafe (==2.1.3)
Requires-Dist: matplotlib (==3.7.2)
Requires-Dist: mpmath (==1.3.0)
Requires-Dist: networkx (==3.1)
Requires-Dist: numpy (==1.25.1)
Requires-Dist: nvidia-cublas-cu11 (==11.10.3.66)
Requires-Dist: nvidia-cuda-cupti-cu11 (==11.7.101)
Requires-Dist: nvidia-cuda-nvrtc-cu11 (==11.7.99)
Requires-Dist: nvidia-cuda-runtime-cu11 (==11.7.99)
Requires-Dist: nvidia-cudnn-cu11 (==8.5.0.96)
Requires-Dist: nvidia-cufft-cu11 (==10.9.0.58)
Requires-Dist: nvidia-curand-cu11 (==10.2.10.91)
Requires-Dist: nvidia-cusolver-cu11 (==11.4.0.1)
Requires-Dist: nvidia-cusparse-cu11 (==11.7.4.91)
Requires-Dist: nvidia-nccl-cu11 (==2.14.3)
Requires-Dist: nvidia-nvtx-cu11 (==11.7.91)
Requires-Dist: packaging (==23.1)
Requires-Dist: Pillow (==10.0.0)
Requires-Dist: pyparsing (==3.0.9)
Requires-Dist: python-dateutil (==2.8.2)
Requires-Dist: requests (==2.31.0)
Requires-Dist: six (==1.16.0)
Requires-Dist: sympy (==1.12)
Requires-Dist: torch (==2.0.1)
Requires-Dist: torchprofile (==0.0.4)
Requires-Dist: torchvision (==0.15.2)
Requires-Dist: tqdm (==4.65.0)
Requires-Dist: triton (==2.0.0)
Requires-Dist: typing-extensions (==4.7.1)
Requires-Dist: urllib3 (==2.0.4)

# SCONCE (Make Pytorch Development and Deployment Efficient)

This is a Pytorch Helper package aimed to aid the workflow of deep learning model development and deployment. 


1. This packages has boiler plate defintions that can ease the development of torch model development
2. Pruning Techniques are being imported from Tomoco Package
3. Model Quantization and Deployment features are in the development pipeline which will be available for use soon.
## Package install:

```python

pip install git+https://github.com/satabios/sconce

```


## Define Network and Config's:

```python
import torch
import torchvision
import torch.nn as nn
import torch.nn.functional as F

# Define your network

class Net(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 8, 3)
        self.bn1 = nn.BatchNorm2d(8)
        self.pool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(8, 16, 3)
        self.bn2 = nn.BatchNorm2d(16)
        self.fc1 = nn.Linear(16*6*6, 32)
        self.fc2 = nn.Linear(32, 10)

    def forward(self, x):
        x = self.pool(self.bn1(F.relu(self.conv1(x))))
        x = self.pool(self.bn2(F.relu(self.conv2(x))))
        x = torch.flatten(x, 1)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return x
    

# Make a Dict of Dataloader

image_size = 32
transforms = {
    "train": Compose([
        RandomCrop(image_size, padding=4),
        RandomHorizontalFlip(),
        ToTensor(),
    ]),
    "test": ToTensor(),
}
dataset = {}
for split in ["train", "test"]:
  dataset[split] = CIFAR10(
    root="data/cifar10",
    train=(split == "train"),
    download=True,
    transform=transforms[split],
  )
dataloader = {}
for split in ['train', 'test']:
  dataloader[split] = DataLoader(
    dataset[split],
    batch_size=512,
    shuffle=(split == 'train'),
    num_workers=0,
    pin_memory=True,
  )
```

## Config:
```python
# Define a cofig of the below parameters
from sconce import sconce, FineGrainedPruner, config

config['sparsity_dict'] = {
    'conv1.weight' : 0.90,
    'conv2.weight' : 0.90,
    'fc1.weight' : 0.90,
    'fc2.weight' : 0.90
}


config['model']= Net() # Model Definition
config['criterion'] = nn.CrossEntropyLoss() # Loss
config['optimizer'] = optim.Adam(config['model'].parameters(), lr=1e-4)
config['scheduler'] = optim.lr_scheduler.CosineAnnealingLR(config['optimizer'], T_max=200)
config['dataloader'] = dataloader

```

## Pipeline using Sconce usage:
```python

sconces = sconce()
print(model_params['model'])

sconces.train()
print("Model Size:",sconces.get_model_size(count_nonzero_only=True))


pruner = FineGrainedPruner()
pruner.prune()
print("Model Size:",sconces.get_model_size(count_nonzero_only=True))

config['callbacks'] = [lambda: pruner.apply()]
if(config['fine_tune']):
   sconces.train()
print("Model Size:",sconces.get_model_size(count_nonzero_only=True))
sconces.evaluate()

```





### To-Do

- [x] Universal Channel-Wise Pruning
- [x] Update Tutorials
- [+] Fine Grained Purning (In-Progress)
- [ ] Quantisation
- [ ] Universal AutoML package
- [ ] Introduction of Sparsification in Pipeline


