ml-kem vs kyber & ml-dsa vs dilithium

[yogappu]

Hello Team,

I noticed a folder of ml-kem in kem along with kyber.
i assumed that ml-kem would be completely based on NIST FIPS203 specification, however i notice differences in that too wrt specification(derand etc..)

So, whats ml-kem exactly based on ? and which is the closest implementation to NIST specification then ?
and is ml-kem-ipd same as ml-kem ? as for tests both are considered same.

[baentsch]

which is the closest implementation to NIST specification then

Those are/should be "ml-kem" and "ml-dsa". Please document any functional (KAT?) differences you can see as Bugs in liboqs.

is ml-kem-ipd same as ml-kem ?

Yes -- for now: The name "ml-kem" will remain the same until the algorithm is finally standardized (but the underlying implementation may change). Please see the documentation.

[yogappu]

@baentsch thanks for the reply. got it now !

Would it be a good idea to set DILITHIUM_MODE = 2/3/5 in https://github.com/open-quantum-safe/liboqs/blob/main/src/sig/ml_dsa/pqcrystals-dilithium-standard_ml-dsa-65-ipd_ref/config.h#L10(actual value supplied from Cmake compile options). Currently its set to 2 in all files.

Also(being greedy :) ), i noticed you as a contributor in kyber github repo. Is it possible to check pq-crystals/kyber#74

Thank you again and sorry for the multiple questions.

Regards.

[baentsch]

Currently its set to 2 in all files.

This indeed seems weird. But this code is/should be a direct copy of the upstream code base. Any feedback, @bhess? But @yogappu, for liboqs this is no problem as this define is overruled by the "compile_opts" setting e.g., here. Maybe easier done than changing the upstream code.

Is it possible to check pq-crystals/kyber#74

I'm not sure who's looking after that repo; I do not (and my contribution was a "historical mistake" :-)

[bhess]

Yes, it's a direct copy from upstream where the default is 2 if DILITHIUM_MODE is not supplied. We set DILITHIUM_MODE with compile opts as @baentsch correctly says.

Re. pq-crystals/kyber#74, I'll comment upstream.

[yogappu]

Closing as queries answered