[quant][graphmode] FP16 quant support - Insert cast operators #40709
Conversation
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
Summary: Cast to kHalf and back to kFloat before the linear operator to mimic FP16 quant support Test Plan: python test/test_quantization.py test_convert_dynamic_fp16 Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
This was referenced Jun 29, 2020
facebook-github-bot
added
the
oncall: jit
…ors" Summary: Cast to kHalf and back to kFloat before the linear operator to mimic FP16 quant support Test Plan: python test/test_quantization.py test_convert_dynamic_fp16 Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit cabf163 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 6 times. |
jerryzh168
reviewed
Jun 29, 2020
| @@ -578,11 +589,9 @@ bool is_module( | |||
| const auto& match_vmap = match.values_map; | |||
| Value* relu = match_vmap.at(vmap.at(vname)); | |||
| auto type = relu->type()->cast<ClassType>(); | |||
jerryzh168
reviewed
Jun 29, 2020
| @@ -44,6 +44,9 @@ TORCH_API bool isScalar(Value* v); | |||
| // Check if value is the input of the graph | |||
| TORCH_API bool hitGraphInput(Value* value); | |||
|
|
|||
| // Return the module name that corresponds to the value. | |||
| TORCH_API c10::optional<std::string> get_module_name(Value* value); | |||
There was a problem hiding this comment.
should we use the same naming convention like other functions? I just noticed this problem, probably the filter functions should follow the name convention as well. but we can do that later.
jerryzh168
reviewed
Jun 29, 2020
| if (quant_type == QuantType::DYNAMIC && isNoopObserver(observer->input(0))) { | ||
| dequant = insertFP16CastOps(g, observer_out); | ||
| } else if ( | ||
| quant_type == QuantType::DYNAMIC && !isWeight(module, observer_out)) { | ||
| Value* dtype = g->insertGetAttr(self, qparam_names.back()); | ||
| std::tie(choose_qparams, quant, dequant) = insertChooseQParamQuantDequant( | ||
| g, observer_out, dtype, at::Symbol::aten(quantize_func)); |
There was a problem hiding this comment.
probably better to add a comment for else branch, it's accepting both dynamic + activation and static quant
jerryzh168
approved these changes
Jun 29, 2020
jerryzh168
reviewed
Jun 29, 2020
…ors" Summary: Cast to kHalf and back to kFloat before the linear operator to mimic FP16 quant support Test Plan: python test/test_quantization.py test_convert_dynamic_fp16 Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
…ors" Summary: Cast to kHalf and back to kFloat before the linear operator to mimic FP16 quant support Test Plan: python test/test_quantization.py test_convert_dynamic_fp16 Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
|
This pull request has been merged in 55b5ab1. |
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:
Summary:
Cast to kHalf and back to kFloat before the linear operator to mimic FP16 quant support
Test Plan:
python test/test_quantization.py test_convert_dynamic_fp16
Reviewers:
Subscribers:
Tasks:
Tags:
Differential Revision: D22335977