You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Feature Request] Consider Array API acceleration for deepmd/utils/tabulate_math.py #5352
The new shared tabulation math module centralizes a large amount of numerical logic. If these operators can be expressed with the Array API consistently, we may be able to:
improve backend portability,
reduce backend-specific special casing,
enable acceleration on compatible backends/devices,
and align this module with the broader Array API direction already discussed in the project.
Possible scope
Audit functions in deepmd/utils/tabulate_math.py for Array API compatibility.
Identify places that still rely on NumPy-specific behavior.
Refactor eligible operators to Array API equivalents.
Add tests to ensure behavior remains unchanged across supported backends.
Notes
This issue is split out from PR #5339 because it is a follow-up optimization/refactor rather than a blocker for the current PR.
Activity
refactor: use Array API in tabulate math
refactor: use Array API-compatible tabulate math (#5366)
refactor: use Array API-compatible tabulate math (deepmodeling#5366)
refactor: use Array API-compatible tabulate math (deepmodeling#5366)