b9a8dff7e5
* Fixed several spelling errors under colossalai * Fix the spelling error in colossalai and docs directory * Cautious Changed the spelling error under the example folder * Update runtime_preparation_pass.py revert autograft to autograd * Update search_chunk.py utile to until * Update check_installation.py change misteach to mismatch in line 91 * Update 1D_tensor_parallel.md revert to perceptron * Update 2D_tensor_parallel.md revert to perceptron in line 73 * Update 2p5D_tensor_parallel.md revert to perceptron in line 71 * Update 3D_tensor_parallel.md revert to perceptron in line 80 * Update README.md revert to resnet in line 42 * Update reorder_graph.py revert to indice in line 7 * Update p2p.py revert to megatron in line 94 * Update initialize.py revert to torchrun in line 198 * Update routers.py change to detailed in line 63 * Update routers.py change to detailed in line 146 * Update README.md revert random number in line 402
2.3 KiB
Gradient Accumulation
Author: Shenggui Li, Yongbin Li
Prerequisite
Example Code
Introduction
Gradient accumulation is a common way to enlarge your batch size for training. When training large-scale models, memory can easily become the bottleneck and the batch size can be very small, (e.g. 2), leading to unsatisfactory convergence. Gradient accumulation works by adding up the gradients calculated in multiple iterations, and only update the parameters in the preset iteration.
Usage
It is simple to use gradient accumulation in Colossal-AI. Just add this following configuration into your config file. The integer represents the number of iterations to accumulate gradients.
gradient_accumulation = <int>
Hands-on Practice
We provide a runnable example to demonstrate gradient accumulation. In this example, we set the gradient accumulation size to be 4. You can run the script using this command:
python -m torch.distributed.launch --nproc_per_node 1 --master_addr localhost --master_port 29500 run_resnet_cifar10_with_engine.py
You will see output similar to the text below. This shows gradient is indeed accumulated as the parameter is not updated in the first 3 steps, but only updated in the last step.
iteration 0, first 10 elements of param: tensor([-0.0208, 0.0189, 0.0234, 0.0047, 0.0116, -0.0283, 0.0071, -0.0359, -0.0267, -0.0006], device='cuda:0', grad_fn=<SliceBackward0>)
iteration 1, first 10 elements of param: tensor([-0.0208, 0.0189, 0.0234, 0.0047, 0.0116, -0.0283, 0.0071, -0.0359, -0.0267, -0.0006], device='cuda:0', grad_fn=<SliceBackward0>)
iteration 2, first 10 elements of param: tensor([-0.0208, 0.0189, 0.0234, 0.0047, 0.0116, -0.0283, 0.0071, -0.0359, -0.0267, -0.0006], device='cuda:0', grad_fn=<SliceBackward0>)
iteration 3, first 10 elements of param: tensor([-0.0141, 0.0464, 0.0507, 0.0321, 0.0356, -0.0150, 0.0172, -0.0118, 0.0222, 0.0473], device='cuda:0', grad_fn=<SliceBackward0>)