Skip to content

CHANGELOG 0.10.0: clarify sm100f fwd/bwd limits and CUEQ_TORCH_COMPILE modes #276

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

Open
hsadasivan wants to merge 2 commits into
base: main
Choose a base branch
from
Open

CHANGELOG 0.10.0: clarify sm100f fwd/bwd limits and CUEQ_TORCH_COMPILE modes #276

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d7ff3ef
update CHANGELOG 0.10.0: clarify sm100f fwd/bwd limits and CUEQ_TORCH…
hsadasivan Apr 22, 2026
548d5ff
CHANGELOG: clarify CUEQ_TORCH_COMPILE applies to fwd pass only
hsadasivan Apr 22, 2026
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
4 changes: 2 additions & 2 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@

### Added
- Python 3.14 support finalized, including a fix for stale tuple hashes in `SegmentedTensorProduct` after in-place operand mutation, and updated CI matrix ([#272](https://github.com/NVIDIA/cuEquivariance/pull/272))
- [Torch/JAX] `cuet.triangle_attention`/`cuex.triangle_attention`: new faster sm100f (CC 10.0/10.3) forward kernel for hidden_dim ≤ 256, bwd hidden_dim ≤ 128; `bias` is cast to q/k/v dtype (instead of always float32) under sm100f; non-contiguous input tensors are handled internally — no manual contiguity assertion is required as long as shape requirements are met; updated docstrings. Only available on cu13 builds ([#260](https://github.com/NVIDIA/cuEquivariance/pull/260))
- [Torch/JAX] `cuet.triangle_attention`/`cuex.triangle_attention`: new faster sm100f (CC 10.0/10.3) forward kernel supporting hidden_dim ≤ 256 (backward acceleration limit remains hidden_dim ≤ 128, unchanged); `bias` is cast to q/k/v dtype (instead of always float32) under sm100f; non-contiguous input tensors are handled internally — no manual contiguity assertion is required as long as shape requirements are met; updated docstrings. Only available on cu13 builds ([#260](https://github.com/NVIDIA/cuEquivariance/pull/260))
- [JAX] MACE `flax.nnx` example restructured to use `nnx.split` + `@jax.jit` on `(graphdef, state)` instead of `@nnx.jit` on the module, removing the Python-side nnx graph traversal overhead from each training/inference step ([#261](https://github.com/NVIDIA/cuEquivariance/pull/261))
- [JAX] NVTX markers added to the MACE examples to make step boundaries visible in `nsys` profiles ([#266](https://github.com/NVIDIA/cuEquivariance/pull/266))

Expand All @@ -16,7 +16,7 @@
- [Torch/JAX] `cuet.attention_pair_bias`/`cuex.attention_pair_bias`: fixed incorrect results when the hidden dimension is not a multiple of 32; the previous torch fallback for these cases is removed as the kernel now handles them correctly

### Notes
- [Torch] The `CUEQ_TORCH_COMPILE` environment variable (experimental) enables `torch.compile` for `cuet.triangle_attention`; useful for non-contiguous tensor inputs on Ampere/Hopper architectures
- [Torch] `CUEQ_TORCH_COMPILE` (experimental): set to a non-zero integer to enable `torch.compile` for the forward pass of `cuet.triangle_attention`; useful for non-contiguous tensor inputs on Ampere/Hopper architectures. Supported modes: `1` → `"default"`, `2` → `"reduce-overhead"`, `3` → `"max-autotune"`, `4` → `"max-autotune-no-cudagraphs"`

### Documentation
- Fixed tutorial format issues ([#274](https://github.com/NVIDIA/cuEquivariance/pull/274))
Expand Down
Loading