[quant] Compute scale and zero point automatically in testing::_quantize #46232

z-a-f · 2020-10-13T02:43:13Z

Stack from ghstack:

[quant] Statically quantized LSTM #46229 [quant] Statically quantized LSTM
[quant] Allow replacement of the custom modules in the "prepare" step #46426 [quant] Allow replacement of the custom modules in the "prepare" step
[quant] Custom module fix #46228 [quant] Custom module fix
[quant] Compute scale and zero point automatically in testing::_quantize #46232 [quant] Compute scale and zero point automatically in testing::_quantize
[quant] Quantized flip dispatch #46235 [quant] Quantized flip dispatch

Differential Revision: D24271061

[ghstack-poisoned]

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

dr-ci · 2020-10-13T06:21:13Z

💊 CI failures summary and remediations

As of commit f027bc5 (more details on the Dr. CI page):

2/2 failures possibly* introduced in this PR
- 2/2 non-CircleCI failure(s)

Extra GitHub checks: 1 failed

Failed: Codecov - codecov/patch

codecov.io: 1 failed

Failed: codecov/patch

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.

See how this bot performed.

This comment has been revised 12 times.

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

codecov · 2020-10-13T21:03:08Z

Codecov Report

Merging #46232 into gh/z-a-f/74/base will decrease coverage by 0.00%.
The diff coverage is n/a.

@@                 Coverage Diff                  @@
##           gh/z-a-f/74/base   #46232      +/-   ##
====================================================
- Coverage             68.34%   68.33%   -0.01%     
====================================================
  Files                   410      410              
  Lines                 53804    53813       +9     
====================================================
+ Hits                  36770    36771       +1     
- Misses                17034    17042       +8

Impacted Files	Coverage Δ
torch/testing/_internal/common_quantized.py	`85.41% <ø> (-6.54%)`	⬇️
torch/testing/_internal/expecttest.py	`77.55% <0.00%> (-1.03%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cd6ee10...f027bc5. Read the comment docs.

supriyar · 2020-10-13T21:07:19Z

torch/testing/_internal/common_quantized.py

    qx = np.round(x / scale + zero_point).astype(np.int64)
    qx = np.clip(qx, qmin, qmax)
    qx = qx.astype(dtype)
-    return qx
+    return qx, scale, zero_point


Are we actually using the returned scale and zero_point anywhere in this PR?

Only in one case I needed it -- this is optional, but because the current function returns the int_repr of the x, it losses information about the scale and zero_point, especially if it is inferred internally in this function

supriyar · 2020-10-13T21:07:43Z

torch/testing/_internal/common_quantized.py

@@ -21,16 +21,27 @@ def _conv_output_shape(input_size, kernel_size, padding, stride, dilation,
                     * (dilation - 1)) / stride) + 2 * output_padding + 1

 # Quantization references
-def _quantize(x, scale, zero_point, qmin=None, qmax=None, dtype=np.uint8):
+def _quantize(x, scale=None, zero_point=None, qmin=None, qmax=None,
+              dtype=np.uint8):


Might be a good idea to update the docstring for this function to reflect the new behavior.

Do you mean the comment here? This is just an internal function for testing -- do we have a docstring?

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

github-actions · 2022-04-12T23:36:08Z

Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as Stale.
Feel free to remove the Stale label if you feel this was a mistake.
If you are unable to remove the Stale label please contact a maintainer in order to do so.
If you want the bot to never mark this PR stale again, add the no-stale label.
Stale pull requests will automatically be closed after 30 days of inactivity.

[quant] Compute scale and zero point automatically in testing::_quantize

ae74b42

[ghstack-poisoned]

This was referenced Oct 13, 2020

[quant] Custom module fix #46228

Closed

[quant] Statically quantized LSTM #46229

Closed

z-a-f requested review from raghuramank100 and supriyar October 13, 2020 02:44

z-a-f added 2 commits October 12, 2020 20:01

Update on "[quant] Compute scale and zero point automatically in test…

0026828

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

Update on "[quant] Compute scale and zero point automatically in test…

ae3b62d

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

z-a-f mentioned this pull request Oct 13, 2020

[quant] Quantized flip dispatch #46235

Closed

Update on "[quant] Compute scale and zero point automatically in test…

bbc0701

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

supriyar reviewed Oct 13, 2020

View reviewed changes

z-a-f requested a review from supriyar October 14, 2020 19:08

supriyar approved these changes Oct 14, 2020

View reviewed changes

Update on "[quant] Compute scale and zero point automatically in test…

f027bc5

…ing::_quantize" Differential Revision: [D24271061](https://our.internmc.facebook.com/intern/diff/D24271061) [ghstack-poisoned]

z-a-f mentioned this pull request Oct 15, 2020

[quant] Allow replacement of the custom modules in the "prepare" step #46426

Closed

facebook-github-bot added the cla signed label Oct 30, 2020

github-actions bot added the Stale label Apr 12, 2022

github-actions bot closed this May 12, 2022

facebook-github-bot deleted the gh/z-a-f/74/head branch June 12, 2022 14:24

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[quant] Compute scale and zero point automatically in testing::_quantize #46232

[quant] Compute scale and zero point automatically in testing::_quantize #46232

z-a-f commented Oct 13, 2020 •

edited

dr-ci bot commented Oct 13, 2020 •

edited

codecov bot commented Oct 13, 2020 •

edited

supriyar Oct 13, 2020

z-a-f Oct 14, 2020

supriyar Oct 13, 2020

z-a-f Oct 14, 2020

github-actions bot commented Apr 12, 2022

[quant] Compute scale and zero point automatically in testing::_quantize #46232

[quant] Compute scale and zero point automatically in testing::_quantize #46232

Conversation

z-a-f commented Oct 13, 2020 • edited

dr-ci bot commented Oct 13, 2020 • edited

💊 CI failures summary and remediations

Extra GitHub checks: 1 failed

codecov.io: 1 failed

codecov bot commented Oct 13, 2020 • edited

Codecov Report

supriyar Oct 13, 2020

Choose a reason for hiding this comment

z-a-f Oct 14, 2020

Choose a reason for hiding this comment

supriyar Oct 13, 2020

Choose a reason for hiding this comment

z-a-f Oct 14, 2020

Choose a reason for hiding this comment

github-actions bot commented Apr 12, 2022

z-a-f commented Oct 13, 2020 •

edited

dr-ci bot commented Oct 13, 2020 •

edited

codecov bot commented Oct 13, 2020 •

edited