Feature: Multinomial sampling #2850
Draft
Conversation
This file contains 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
|
Great timing as I quite need this function! I just need the CubeCL version, would you say that is already usable? Any plans to continue this PR? Thank you for the contribution |
|
@ArthurBrussee Hey, so I am currently kinda stuck on the CubeCL version, as I am very new to CubeCL in general. I will then want to run some benchmarks to make sure nothing silly is happening, especially with the Candle backend. I tested it a little locally and it seems to work well, but I wanna make sure it is at least as good as the CPU impl for large Tensors + that it uses GPU resources as much as possible. |
Pablo1785
force-pushed
the
feat/multinomial_sampling
branch
from
dd3883c to
b1a0b95
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Pull Request Template
Checklist
run-checks allscript has been executed.Related Issues/PRs
Would resolve:
#1121
Since the discussion seemed to die down in the issue and I found myself in need of the Multinomial distribution sampler, I decided to take a shot at implementing this.
One issue to solve is how to pass the probabilities to the Cube Launch (how to turn a Tensor into a TensorArg?)
Changes
Add multinomial distribution implementation for all backends. Potential breaking change: Distribution can no longer derive Copy, since Multinomial distribution has to contain a collection of probabilities, which could potentially be very large.
Testing
Describe how these changes have been tested.