Fix Image GPT

[teddykoker]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

Fixes Image GPT. Images were not being properly quantized before. See code:

    def forward(self, x, classify=False):
        x = _shape_input(x)

        # TODO(teddykoker): this is a hack to quantize images into `vocab_size` bins.
        # This only works with 1 channel images; something like KNN needs to be used
        # for RGB. Assumes data is in [0.0, 1.0].
        x = torch.round(x * (self.hparams.vocab_size - 1)).long()

        return self.gpt(x, classify)

Also fixes docs.

Review

@williamFalcon

Discussion

Implementing KNN for image quantization is fairly straight forward. I would be willing to implement at some point. But I am unsure where this code would go as is it is an independent preprocessing step that just needs to be run once (in practice I would just create my own dataset for the new quantized images).

[codecov-commenter]

Codecov Report

Merging #108 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #108   +/-   ##
=======================================
  Coverage   91.91%   91.91%           
=======================================
  Files          77       77           
  Lines        3944     3944           
=======================================
  Hits         3625     3625           
  Misses        319      319           

Flag	Coverage Δ
#unittests	91.91% <100.00%> (ø)

Impacted Files	Coverage Δ
pl_bolts/models/vision/image_gpt/gpt2.py	100.00% <100.00%> (ø)
pl_bolts/models/vision/image_gpt/igpt_module.py	98.78% <100.00%> (ø)

---

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 024b574...17048c6. Read the comment docs.