Bitpacking

[vayuda]

Based on this issue: #284

Adding this first iteration of packing/unpacking algorithms to support lower bit dtypes into protoype/

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/291

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit b01550f with merge base 5485929 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[gau-nernst]

Just curious. container_size can be determined from data.dtype right? e.g. uint8 -> 8, uint16 -> 16. (there is also this - https://pytorch.org/docs/stable/type_info.html#torch.torch.iinfo).
Also, is it assumed that data.dtype has container_size number of bits? What if data use larger or smaller bit-width than container_size? e.g. store int4 in int32, then request to pack to int8. Depending on what are your assumptions to the inputs, perhaps some kind of type checking and/or type casting is good.

[msaroufim]

Nice first step let's keep iterating on cuda mode to figure out how to promote this to a stable feature