https://github.com/notedance/pytorch-pcgrad

PyTorch Implementation of "Gradient Surgery for Multi-Task Learning" using multiprocessing
https://github.com/notedance/pytorch-pcgrad

deep-learning deep-reinforcement-learning multi-task-learning multi-task-reinforcement-learning multi-task-rl optimizer pytorch reinforcement-learning

Last synced: 8 months ago
JSON representation

PyTorch Implementation of "Gradient Surgery for Multi-Task Learning" using multiprocessing

• Host: GitHub
• URL: https://github.com/notedance/pytorch-pcgrad
• Owner: NoteDance
• License: apache-2.0
• Created: 2025-05-17T14:51:36.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-05-17T14:53:48.000Z (9 months ago)
• Last Synced: 2025-06-10T20:54:15.339Z (8 months ago)
• Topics: deep-learning, deep-reinforcement-learning, multi-task-learning, multi-task-reinforcement-learning, multi-task-rl, optimizer, pytorch, reinforcement-learning
• Language: Python
• Homepage:
• Size: 8.79 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Pytorch-PCGrad

PyTorch Implementation of "Gradient Surgery for Multi-Task Learning" using multiprocessing

# Usage

```python

import torch

import torch.nn as nn

import torch.optim as optim

from ppcgrad import PPCGrad

# wrap your favorite optimizer

optimizer = PPCGrad(optim.Adam(net.parameters())) 

losses = [...] # a list of per-task losses

assert len(losses) == num_tasks

optimizer.pc_backward(losses) # calculate the gradient can apply gradient modification

optimizer.step()  # apply gradient step

```