Thread by @akshay_pachaar

I have been training neural networks for 10 years now.

Here are 16 ways I actively use to optimize model training:

Before we dive in, the following visual covers what we are discussing today.

Let's understand them in detail below!

These are some basic techniques:

1) Use efficient optimizers—AdamW, Adam, etc.

2) Utilize hardware accelerators (GPUs/TPUs).

3) Max out the batch size.

4) Use multi-GPU training through Model/Data/Pipeline/Tensor parallelism. Check the visual👇

5) Bayesian optimization for hyperparameter optimization:

This technique takes informed steps based on the results of the previous hyperparameter configs.

This way, the model converges to an optimal set of hyperparameters much faster.

Check these results 👇

6) Initialize parameters with He or Xavier initialization.

7) For large models, use DeepSpeed, FSDP, YaFSDP, etc.

8) Mixed precision training: Use lower precision float16 along with float32. This leads to faster computation. Check the visual 👇

9) Use DistributedDataParallel, not DataParallel.

10) Use torch.rand(2, 2, device ...) to create a tensor on GPU. A .cuda() call creates a tensor on CPU and then transfers it to GPU, which is slow.

11) Use activation checkpointing in memory constraints👇

12) Use gradient accumulation.

13) Normalize data after transferring to GPU (for integer data, like pixels):

- Normalizing before will transfer 32-bit floats to the GPU.
- Normalizing after will transfer 8-bit floats to the GPU.
- The latter is better.

Check this👇

14) Use momentum

In gradient descent, every parameter update solely depends on the current gradient. This leads to unwanted oscillations during optimization.

Momentum reduces this by adding a weighted average of previous gradient updates to the update rule.

Check this 👇

15-16) Set max_workers and pin_memory in DataLoader.

PyTorch dataloader has two terrible default settings. Update them according to your config.

Speedup is shown in the image below 👇

Those were 16 techniques that I actively use to optimize neural network training.

If I missed something, please drop that in the replies.

Here's the visual again for your reference 👇

That's a wrap!

If you found it insightful, reshare with your network.

Find me → @akshay_pachaar ✔️
For more insights and tutorials on LLMs, AI Agents, and Machine Learning!

https://twitter.com/703601972/status/1961405987203809434