enable backward for log1p (sparse layouts) #88155
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/88155
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 274846a: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
amjames
added
module: sparse
cc nikitaved pearu cpuhrsch bhosmer [ghstack-poisoned]
ghstack-source-id: 93d98e44b99c40083844274263f8c79c36a6779c Pull Request resolved: #88155
cc nikitaved pearu cpuhrsch bhosmer [ghstack-poisoned]
ghstack-source-id: 13be4fa066994ba23aa7a72e9e207b58762eeb67 Pull Request resolved: #88155
|
@cpuhrsch this is ready for another look when you get a chance |
| // If grad is sparse we can't divide by the n-d (self + 1).conj(), so we | ||
| // must multiply by the recipricol, layout of grad is preserved which is | ||
| // important to gradcheck | ||
| return grad * self_p1_conj.reciprocal_(); |
There was a problem hiding this comment.
Why is that? Because it is not supported?
There was a problem hiding this comment.
Yes, div supports scalar or zero-dim dense tensor divisors only.
There was a problem hiding this comment.
Ah, I see. We can have it enabled for COO at least now that mul is merged :)
cc nikitaved pearu cpuhrsch bhosmer [ghstack-poisoned]
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Pull Request resolved: pytorch#88155 Approved by: https://github.com/cpuhrsch
Pull Request resolved: pytorch#88155 Approved by: https://github.com/cpuhrsch
Stack from ghstack (oldest at bottom):
cc @nikitaved @pearu @cpuhrsch @bhosmer