Add FP16 support to ptq_evaluate.py and update README argument list

[hkayann]

Reason for this PR

Changes Made in this PR

Testing Summary

Risk Highlight

• This PR includes code from another work (please detail).
• This PR contains API-breaking changes.
• This PR depends on work in another PR (please provide links/details).
• This PR introduces new dependencies (please detail).
• There are coverage gaps not covered by tests.
• Documentation updates required in subsequent PR.

Checklist

• Code comments added to any hard-to-understand areas, if applicable.
• Changes generate no new warnings.
• Updated any relevant tests, if applicable.
• No conflicts with destination dev branch.
• I reviewed my own code changes.
• Initial CI/CD passing.
• 1+ reviews given, and any review issues addressed and approved.
• Post-review full CI/CD passing.

[Giuseppe5]

Just 2 quick comments, otherwise looks good!

Thanks again for this.

[Giuseppe5]

I think half and float16 are equivalent, and for consistency reasons I think I prefer float16.

If I am missing something, let me know please.

[hkayann]

Yes, they are functionally equivalent. My initial thought was that using half could help prevent typos, as the only difference is the letter b.

[hkayann]

Now, I am trying to save custom-bit-width models, for example FP16 with mantissa 9 bits, exponent 6 bits etc, but seems like not possible given the available PyTorch dtypes.

[Giuseppe5]

Unfortunately I'm not sure I can fully help with the second issue, since we mostly focus on minifloat quantization with 8 bits of fewer.

In the meantime, would you mind changing half to float16? I understand the potential for typos but I still prefer trying to be consistent across the codebase, and we never (or rarely) use half instead of float16.

Thanks!

[hkayann]

I can simulate for now, so I have a working workaround which is something. I have done all the requested changes as requested. Many thanks again.

[Giuseppe5]

Thanks for this again :)

I will let the tests run and then merge it