Metadata-Version: 2.1
Name: rockmate
Version: 2.0.0
Summary: Efficient and Automatic Rematerialization for Pytorch training
Author: Théotime Le Hellard, Xunyi Zhao, Julia Gusak, Li Zhe, Olivier Beaumont, Lionel Eyraud-Dubois
Maintainer-email: Lionel Eyraud-Dubois <lionel.eyraud-dubois@inria.fr>
License: MIT License
        
        Copyright (c) 2021-present Inria 
        
        Permission is hereby granted, free of charge, to any person obtaining a copy
        of this software and associated documentation files (the "Software"), to deal
        in the Software without restriction, including without limitation the rights
        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
        copies of the Software, and to permit persons to whom the Software is
        furnished to do so, subject to the following conditions:
        
        The above copyright notice and this permission notice shall be included in all
        copies or substantial portions of the Software.
        
        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
        SOFTWARE.
        
Project-URL: Repository, https://github.com/topal-team/rockmate.git
Keywords: rematerialization,training,pytorch,memory
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: COPYING
Requires-Dist: numpy
Requires-Dist: torch>=1.7
Requires-Dist: rkgb>=2.0.0
Requires-Dist: pulp==2.7.0
Requires-Dist: configmypy>=0.2.0

# Rockmate

The `Rockmate` framework is designed for training a PyTorch neural network within a given GPU budget
constraint using automatic re-materialization (activation checkpointing) technique.

Given a PyTorch model, a sample input, and a GPU memory budget, `Rockmate` builds a new
`torch.nn.Module`, which performs forward and backward pass keeping activations under the given
budget.

- The new model produces the same outputs and gradients as the original one.
- Model training with a budget constraint, which is lower than the one required by PyTorch Autodiff,
  is achieved by re-computing some of the activations instead of storing them for gradient
  calculation.
- Depending on the budget, `Rockmate` defines automatically which activations should be recomputed.

More information on [our repository](https://github.com/topal-team/rockmate).
