Conversation
The `get_i2v_mask` method had a hardcoded `device="cuda"` default, which crashes on non-CUDA accelerators (Ascend NPU, etc.) with "Torch not compiled with CUDA enabled" when called without an explicit device argument. Change the default to None and resolve via `self._execution_device`, matching the pattern used across other pipeline methods. Verified on Ascend 910B NPU: torch.zeros(device="cuda") crashes, fix with device-agnostic resolution creates tensors on the correct device.
The VAE decode block in the MiniMax H3 pipeline only enables fp16 autocast on CUDA (`enabled=device.type == "cuda"`), which disables autocast on other accelerators such as Ascend NPU, causing suboptimal performance and potential dtype mismatches. Change to `enabled=device.type != "cpu"` so autocast is enabled on any accelerator device while remaining disabled on CPU. Verified on Ascend 910B NPU: torch.autocast(device_type="npu", dtype=torch.float16, enabled=True) correctly computes in fp16.
github-actions
Bot
added
modular-pipelines
pipelines
size/S
Contributor
|
Hi @li-lizhe, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Please note that PRs without a linked issue are likely to be automatically closed 10 days after this notice. Once the PR links an issue (or gets the |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description: The VAE decode block in the MiniMax H3 modular pipeline only enables fp16 autocast on CUDA devices (
enabled=device.type == "cuda"), which disables autocast on other accelerators such as Ascend NPU, Intel XPU, or Apple MPS.Without autocast on NPU, the VAE decode runs in fp32 (or the tensor's dtype without fp16 acceleration), causing suboptimal performance and potential dtype mismatches when the pipeline expects half-precision computation.
Change:
enabled=device.type == "cuda"→enabled=device.type != "cpu"Verification on Ascend 910B NPU (torch 2.14.0a0 + torch_npu):
torch.autocast(device_type="npu", dtype=torch.float16, enabled=True)produces fp16 output (out.dtype=torch.float16)torch.autocast(device_type="npu", dtype=torch.float16, enabled=False)produces fp32 output (out.dtype=torch.float32)