[PT2E][Quant] Refactor quantizer and qnnpack qantizer code to support dqlinear config #99399
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
… dqlinear config This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/99399
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ✅ No FailuresAs of commit f38d2ba: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Apr 18, 2023
kimishpatel
added a commit
that referenced
this pull request
Apr 18, 2023
… dqlinear config This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) ghstack-source-id: 186387917 Pull Request resolved: #99399
jerryzh168
reviewed
Apr 18, 2023
| _QUANT_CONFIG_TO_ANNOTATOR = {} | ||
|
|
||
|
|
||
| def register_annotator(quantization_configs: List[QuantizationConfig]): |
There was a problem hiding this comment.
why is this indexed by quantization_config? are we still planning to support module fqn/type based config or other existing features in QConfigMapping or do we just want to support global qconfig?
There was a problem hiding this comment.
So the way I imagined this to be used is like this:
qnnpack, i will rename this to xnnpack, quantizer has two configs:
symmetric_static_config
symmetric_dynamic_config.
THis decorator registers annotation function for each.
Now if I want to quantize a model with dynamic q linear for linear and static quant for the rest, I would do it like this with composable quantizer.
dq_linear_config = QNNPACKQuantizer.get_dynamic_linear_config()
dq_linear_quantizer = QNNPACKQuantizer().set_global(dq_linear_config)
static_qconfig = QNNPACKQuantizer.get_symmetric_config()
static_quantizer = QNNPACKQuantizer().set_global(static_config)
composed_quantizer = ComposableQuantizer([dq_linear_quantizer, static_quantizer])
We can alternatively do this via:
qnnpack_quantizer = QNNPACKQuantizer().set_global(static_config)
# following api doesnt yet exist
qnnpack_quantizer.set_module_type_config(torch.nn.Linear, QNNPACKQuantizer.get_dynamic_linear_config()
I think both are fine. RIght now I am doing this mainly because it felt somewhat easier to use, but I am open to just doing via set_module_config or similar API.
There was a problem hiding this comment.
Let me think a bit more. If it makes sense to just set_module_type_config api, then I will remove the config-to-annotator mapping part from the PR and follow up separately.
There was a problem hiding this comment.
Ok I thought about this and I dont think we should do
dq_linear_config = QNNPACKQuantizer.get_dynamic_linear_config()
dq_linear_quantizer = QNNPACKQuantizer().set_global(dq_linear_config)
static_qconfig = QNNPACKQuantizer.get_symmetric_config()
static_quantizer = QNNPACKQuantizer().set_global(static_config)
composed_quantizer = ComposableQuantizer([dq_linear_quantizer, static_quantizer])
There was a problem hiding this comment.
But for now I will use annotator api until we do this better
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
|
just to confirm the plan is to follow up with set_module_type_config api right? |
jerryzh168
approved these changes
Apr 22, 2023
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
kimishpatel
added a commit
that referenced
this pull request
Apr 28, 2023
… dqlinear config Pull Request resolved: #99399 This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. ghstack-source-id: 187514867 Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/)
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [ghstack-poisoned]
… to support dqlinear config" This diff introduces a few refactors: - Move observer creation to utils.py. - Use quantization spec to supply args to observers. - Use annotation function registration corresponding QuantizationConfig. This will be later used in dynamic quantized linear. Differential Revision: [D45073790](https://our.internmc.facebook.com/intern/diff/D45073790/) [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 |
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.
Stack from ghstack (oldest at bottom):
This diff introduces a few refactors:
will be later used in dynamic quantized linear.
Differential Revision: D45073790