Last active
December 21, 2020 12:34
-
-
Save nilesh0109/9b8e68cca6d36cb7e226f76bb7a5ff08 to your computer and use it in GitHub Desktop.
Accumulating gradients for large batch training on single GPU.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| TARGET_BATCH_SIZE, BATCH_FIT_IN_MEMORY = 256, 32 | |
| accumulation_steps = int(TARGET_BATCH_SIZE / BATCH_FIT_IN_MEMORY) | |
| network.zero_grad() # Reset gradients tensors | |
| for i, (imgs, labels) in enumerate(dataloader): | |
| preds = network(imgs) # Forward pass | |
| loss = loss_function(preds, labels) # Compute loss function | |
| loss = loss / accumulation_steps # Normalize our loss (if averaged) | |
| loss.backward() # Backward pass | |
| if (i+1) % accumulation_steps == 0: # Wait for several backward steps | |
| optim.step() # Perform an optimizer step | |
| network.zero_grad() # Reset gradients tensors |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment