Files
axolotl/docs/gradient_checkpointing.qmd
Wing Lian c9df6efdc2 support offloading layers to CPU (#3512) [skip ci]
* support offloading layers to CPU

* chore: lint

* revert change

* update docs
2026-03-21 22:47:02 -04:00

60 lines
2.4 KiB
Plaintext

---
title: Gradient Checkpointing, Activation Offloading, and Layer Offloading
---
Gradient checkpointing and activation offloading are techniques used to optimize the performance of deep learning
models by reducing the memory footprint and improving computational efficiency.
### Enabling Gradient Checkpointing
```yaml
gradient_checkpointing: true
```
### Enabling Activation Offloading
```yaml
gradient_checkpointing: true # required for activation offloading
activation_offloading: true
```
Activation offloading variants:
The default `activation_offloading: true` offloads activations to CPU and uses CUDA streams
to overlap the communications and computations when offloading.
The `activation_offloading: legacy` naively offloads activations to CPU and without additional optimizations.
For resource constrained environments with limited CPU memory, `activation_offloading: disk` offloads
activations to disk instead of CPU RAM so that much larger context lengths can be trained with minimal memory.
### Enabling Layer Offloading
```yaml
layer_offloading: true
```
Layer offloading reduces GPU memory usage by moving frozen (non-trainable) decoder layer parameters to CPU
and streaming them back to GPU one layer at a time during the forward and backward passes. This is
particularly useful for LoRA/QLoRA training where most of the model's parameters are frozen — only the
trainable adapter weights stay on GPU permanently.
During training, forward and backward hooks on each decoder layer handle the transfer automatically:
- **Forward pass:** Before a layer executes, its frozen params are loaded to GPU. The next layer is
prefetched asynchronously on a separate CUDA stream for overlap.
- **Backward pass:** Same pattern in reverse — the current layer's frozen params are loaded and the
previous layer is prefetched.
After each layer finishes, its frozen params are offloaded back to CPU pinned memory.
This approach trades some CPU-GPU transfer overhead for significant GPU memory savings — the freed memory
is roughly equal to the size of all frozen parameters across all decoder layers, minus one layer's worth
that is kept on GPU at any given time.
**Requirements:**
- CUDA GPU (CPU-only training is not supported for this feature)
- Works with any HuggingFace model architecture that uses decoder layers (Llama, Mistral, Qwen, etc.)
- Best combined with LoRA/QLoRA where most parameters are frozen