Enable dispatch to tinygemm int4 and int8 kernels for quantized tensor#230
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/230
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 2a8dc5d with merge base e7bbbd2 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Contributor
Author
|
waiting for #227 to be landed to actually test it |
cpuhrsch
reviewed
May 9, 2024
| None if len(args) == 2 else args[2], | ||
| ) | ||
| if weight_qtensor.input_quant_func is not None: | ||
| if weight_qtensor.input_quant_func is None: |
Contributor
There was a problem hiding this comment.
Let's please remove input_quant_func and discuss using AffineQuantizedTensor as an organizing principle :)
Contributor
Author
There was a problem hiding this comment.
this is used for testing 8da4w right now, can this be done later until there is a better alternative?
cpuhrsch
approved these changes
May 9, 2024
cpuhrsch
requested changes
May 10, 2024
cpuhrsch
left a comment
Contributor
There was a problem hiding this comment.
Setting request changes just so we don't forget to remove input_quant_func
Contributor
Author
will do in next PR |
cpuhrsch
reviewed
May 10, 2024
|
|
||
|
|
||
| # TODO: add padding support | ||
| class TinygemmAffineQuantizedTensor(AffineQuantizedTensor): |
cpuhrsch
requested changes
May 10, 2024
cpuhrsch
left a comment
Contributor
There was a problem hiding this comment.
CI is red and let's please not add a TinygemmAffineQuantizedTensor
Contributor
Author
|
sorry, CI should be fixed now. For tinygemm, it's because it is quantizing things differently, we need to think of how to unify this |
b5950a3 to
449a5d8
Compare
cpuhrsch
approved these changes
May 13, 2024
61befaa to
943bf13
Compare
…ed tensor Summary: This adds some dispatch to the tinygemm kernels for cuda, although need to resolve implementation mismatch problem for tinygemm first Test Plan: python test/quantization/test_quant_api.py -k test_quantized_tensor_subclass_int4 python test/quantization/test_quant_api.py -k test_quantized_tensor_subclass_int8 Reviewers: Subscribers: Tasks: Tags:
cpuhrsch
reviewed
May 14, 2024
| _replace_with_custom_fn_if_matches_filter( | ||
| model, | ||
| lambda linear_mod: dynamic_quant(linear_mod, (torch.randn(1, linear_mod.in_features))), | ||
| lambda linear_mod: dynamic_quant(linear_mod, (torch.randn(1, linear_mod.in_features),)), |
Contributor
There was a problem hiding this comment.
Interesting. Why is that extra comma needed now?
Contributor
There was a problem hiding this comment.
By all means don't be blocked on this comment haha
cpuhrsch
reviewed
May 14, 2024
|
|
||
| self.assertTrue(torch.equal(scale, scale_ref)) | ||
| torch.testing.assert_close(zero_point_float, zero_point_ref, rtol=0.00001, atol=torch.max(scale)*0.03) | ||
| self.assertTrue(torch.equal(zero_point, zero_point_ref)) |
lancerts
pushed a commit
to lancerts/ao
that referenced
this pull request
May 17, 2024
pytorch#230) Summary: This adds some dispatch to the tinygemm kernels for cuda, although need to resolve implementation mismatch problem for tinygemm first Test Plan: python test/quantization/test_quant_api.py -k test_quantized_tensor_subclass_int4 python test/quantization/test_quant_api.py -k test_quantized_tensor_subclass_int8 Reviewers: Subscribers: Tasks: Tags:
yanbing-j
pushed a commit
to yanbing-j/ao
that referenced
this pull request
Dec 9, 2024
* clean up gguf loading. Move model loading to meta. * remove cpu * Fix CI and validation scripts (pytorch#154) * missing device (pytorch#232) * Use generator args to group all arguments to generator (pytorch#231) * prompt * chat_mode, num_samples * Move more generator args to use dataclass (pytorch#233) * prompt * chat_mode, num_samples * move more args * more gen args * update * args * undo some changes * typos * Minor lint fixes (pytorch#236) * remove redundancy & remove int4 linear test from ET tests (pytorch#237) * remove redundancy * no int4 linear on ET * small changes --------- Co-authored-by: Guang Yang <42389959+guangy10@users.noreply.github.com> Co-authored-by: Michael Gschwind <61328285+mikekgfb@users.noreply.github.com> Co-authored-by: Mergen Nachin <mnachin@meta.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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:
This adds some dispatch to the tinygemm kernels for cuda, although need to resolve implementation mismatch problem for tinygemm first
Test Plan:
TODO
Reviewers:
Subscribers:
Tasks:
Tags: