[quantization] fix run_arg tiny bug

[leimao]

This fix allows the calibration function to take in more than one positional argument.

[dr-ci]

💊 CI failures summary and remediations

As of commit 3c209a3 (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.

See how this bot performed.

This comment has been revised 24 times.

[codecov]

Codecov Report

Merging #48537 (3c209a3) into master (272f4db) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master   #48537      +/-   ##
==========================================
- Coverage   80.93%   80.93%   -0.01%     
==========================================
  Files        1855     1855              
  Lines      200194   200194              
==========================================
- Hits       162024   162020       -4     
- Misses      38170    38174       +4     

[leimao]

@jerryzh168

[jerryzh168]

OK, thanks. Please try not to use this api since we might want to deprecate it in the future.
this can be replace by

model = prepare(model), 
calibrate(model, inputs)
model = convet(model)

[facebook-github-bot]

@jerryzh168 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[leimao]

OK, thanks. Please try not to use this api since we might want to deprecate it in the future.
this can be replace by

model = prepare(model), 
calibrate(model, inputs)
model = convet(model)

I found you have a similar high-level API for dynamic quantization as well. Can I assume the torch.quantization.quantize_dynamic API will not likely change?

[jerryzh168]

OK, thanks. Please try not to use this api since we might want to deprecate it in the future.
this can be replace by

model = prepare(model), 
calibrate(model, inputs)
model = convet(model)

I found you have a similar high-level API for dynamic quantization as well. Can I assume the torch.quantization.quantize_dynamic API will not likely change?

we won't deprecate these in short term. but not sure about longer term. in general quantize_dynamic is not needed either. you can use

model = prepare(model, qconfig_dict)
model = convert(model)

as well. a single line api does not buy us much and increases the api surface.

For example, in case of static quantization, the api should also include a key word argument dictionary to be complete (currently we only have positional run_args).

this file: https://github.com/pytorch/pytorch/blob/master/torch/quantization/quantize_fx.py shows the long term apis we are thinking of better. basically just prepare, prepare_qat and convert, and user will decide whether to run calibration/training or not. (we also have some warnings in the observer when calibration/training is not run)

[leimao]

@jerryzh168 Thank you. I will try to stay away from the "high-level" wrapper API as much as I can.

[facebook-github-bot]

@jerryzh168 merged this pull request in 0db7346.

[albanD]

Note that this is a BC-breaking and it breaks the quantization tutorials.