Feat: Added requantize function and tests

[calmitchell617]

What does this PR do?

This PR addresses issue #162 by adding a requantize function with the following signature, and appropriate tests.

def requantize(model, state_dict):

The function does the following, inplace:

• Checks to make sure the model isn't a Transformers model which has been distributed via accelerate (ie device_map=auto)
• Moves the model to the meta device
• Quantizes it
• Moves the empty, quantized model to the cpu
• Loads the state dict into the model.

At this point, the user can move the requantized model to their preferred device.

Who can review?

@dacorvo

Also, FYI @ManoBharathi93

[calmitchell617]

Added a new commit creating a new file for the tests. I'm not sure what to do RE:

Here we should also check for weight qtype and parameters. Same for activations.

I am happy to try again with more instruction, or just observe how you would do it with a commit of your own.

[dacorvo]

Don't forget to use conventional commits or the CI will fail.
You can for instance add all commits during review as:

review: whatever I was asked to do by the almighty reviewer

Also try to check style locally before pushing:

$ make style

[calmitchell617]

Ok, I will keep working on it tomorrow. Thanks for your attention on this.

[ManoBharathi93]

@calmitchell617 since you have not done any commits last 2 days I made PR by referring your commit with @dacorvo suggestions on your commit.

[calmitchell617]

Hey @ManoBharathi93, appreciate your persistence. I'm going to continue here anyway, as an exercise.

[calmitchell617]

@dacorvo, I believe the requested changes have been made.

[ManoBharathi93]

Hey @ManoBharathi93, appreciate your persistence. I'm going to continue here anyway, as an exercise.

Looking forward to learn from you @calmitchell617

[dacorvo]

Thanks for the pull-request ! I would require a few changes related to the device placement.

[calmitchell617]

Ok, I will make those changes soon.

FYI, I'm writing a review / usage guide of Quanto for my blog. If you would like to see a draft of it to make sure I'm not misrepresenting anything, I would be happy to send you a draft and give you a chance to critique.

[dacorvo]

Ok, I will make those changes soon.

FYI, I'm writing a review / usage guide of Quanto for my blog. If you would like to see a draft of it to make sure I'm not misrepresenting anything, I would be happy to send you a draft and give you a chance to critique.

That's very nice of you to ask. I will be happy to take a look.

[dacorvo]

Thank you very much for this pull-request !

[dacorvo]

There seems to be an issue with the formatting of your changes. If you applied style locally and don't have that error, please let me know and I will check if we need to update or pin the linter version.

[calmitchell617]

Ah, darn it. Forgot to run make style. I'm too used to working on repos by myself :-D

I amended the commit.