MX4 quantization #2659
Closed
MX4 quantization #2659
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
Summary:
Implement MX4 quantization-dequantization ops
Usage:
**Quantization:**
```
quantized_output = torch.ops.fbgemm.quantize_mx_cuda(
A,
split_sizes,
scale_bits=8,
ebits=2,
mbits=3,
max_norm=6.0f,
mx_group_size=32,
)
```
where
`A` is 1-D input tensor and
`split_sizes` is list of int containing number of elements in each rank e.g., `split_sizes = [1024, 2048]` for 2 ranks. Note that each value needs to be a multiple of `group size`.
The output is ensured to be 16-byte aligned. Anything less than 16 bytes is padded to make 16 bytes.
Given that the `group_size` is 32, for the best performance, the value should be a multiple of 32 x 16 = 512.
**Dequantization:**
```
dequanted_output = torch.ops.fbgemm.dequantize_mx_cuda(
quantized_output,
split_sizes,
mx_group_size=32,
)
```
Reviewed By: sryap
Differential Revision: D57145102
|
This pull request was exported from Phabricator. Differential Revision: D57145102 |
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D57145102 |
|
This pull request has been merged in eb7ccec. |
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Outdated suggestions cannot be applied.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.
Summary:
Implement MX4 quantization-dequantization ops
Usage:
Quantization:
where
Ais 1-D input tensor andsplit_sizesis list of int containing number of elements in each rank e.g.,split_sizes = [1024, 2048]for 2 ranks. Note that each value needs to be a multiple ofgroup size.The output is ensured to be 16-byte aligned. Anything less than 16 bytes is padded to make 16 bytes.
Given that the
group_sizeis 32, for the best performance, the value should be a multiple of 32 x 16 = 512.Dequantization:
Reviewed By: sryap
Differential Revision: D57145102