Skip to content

[docs] add a tip for compile + offload #11731

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 3 commits into from
Closed

[docs] add a tip for compile + offload #11731

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
edef2da
add a tip for compile + offload
sayakpaul Jun 17, 2025
222ed15
Merge branch 'main' into tip-compile-offload
sayakpaul Jun 17, 2025
b7f414f
more combos.
sayakpaul Jun 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions docs/source/en/optimization/memory.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -302,6 +302,13 @@ compute-bound, [group-offloading](#group-offloading) tends to be better. Group o

</Tip>

<Tip>

When using offloading, users can additionally compile the diffusion transformer/unet to get a
good speed-memory trade-off. First set `torch._dynamo.config.cache_size_limit=1000`, and then before calling the pipeline, add `pipeline.transformer.compile()`.

</Tip>

## Layerwise casting

Layerwise casting stores weights in a smaller data format (for example, `torch.float8_e4m3fn` and `torch.float8_e5m2`) to use less memory and upcasts those weights to a higher precision like `torch.float16` or `torch.bfloat16` for computation. Certain layers (normalization and modulation related weights) are skipped because storing them in fp8 can degrade generation quality.
Expand Down Expand Up @@ -365,6 +372,12 @@ apply_layerwise_casting(
)
```

<Tip>

Layerwise casting can be combined with group offloading.

</Tip>

## torch.channels_last

[torch.channels_last](https://pytorch.org/tutorials/intermediate/memory_format_tutorial.html) flips how tensors are stored from `(batch size, channels, height, width)` to `(batch size, heigh, width, channels)`. This aligns the tensors with how the hardware sequentially accesses the tensors stored in memory and avoids skipping around in memory to access the pixel values.
Expand Down