Describe the bug
In DDIMPipeline, the eta parameter is documented to be within the range [0, 1], but there is currently no validation enforcing this constraint.
From the docstring:
"eta corresponds to η in paper and should be between [0, 1]"
However, users can pass arbitrary values (e.g., negative or >1) without any warning or error, which may lead to undefined or unintended behavior during inference.
This creates a mismatch between:
- documented API contract
- actual implementation behavior
I’d be happy to submit a PR to add validation if this is considered appropriate.
Reproduction
from diffusers import DDIMPipeline
pipe = DDIMPipeline.from_pretrained("google/ddpm-cifar10-32")
# Invalid eta values (should ideally raise error or warning)
image = pipe(eta=-1.0)
image = pipe(eta=2.0)Both calls execute without validation or error.
Logs
Either:
- Raise ValueError if eta is outside [0, 1]
- Or explicitly document that values outside the range are allowed (if intentional)
System Info
- diffusers: local dev version (from source)
- Python: 3.13
- OS: Windows
- PyTorch: 2.11.0
Who can help?
No response
Describe the bug
In DDIMPipeline, the eta parameter is documented to be within the range [0, 1], but there is currently no validation enforcing this constraint.
From the docstring:
However, users can pass arbitrary values (e.g., negative or >1) without any warning or error, which may lead to undefined or unintended behavior during inference.
This creates a mismatch between:
I’d be happy to submit a PR to add validation if this is considered appropriate.
Reproduction
Both calls execute without validation or error.
Logs
Either: - Raise ValueError if eta is outside [0, 1] - Or explicitly document that values outside the range are allowed (if intentional)System Info
Who can help?
No response