[BUG] Refine AutoTuner configuration recommendations

[kuhushukla]

Describe the bug
This bug tracks the work we need to do to improve the autotuner o/p we ask customers to try out for their first GPU run. This issue also relates to #1334 and #1067 .

[amahussein]

Is there a way we can cleanup old issues that overlap with this umbrella?

[parthosa]

Many of these recommendations require a separate recommendation when AutoTuner is run on CPU event logs (i.e. via Qual Tool) vs when AutoTuner is run on GPU event logs (i.e. via Profiling Tool).

I plan to use a class based approach to create QualAutoTuner and ProfilingAutoTuner that extend the base AutoTuner. Now we can have overrides as required. E.g. QualAutoTuner can override BATCH_SIZE_BYTES to be 1 GB as compared to 2 GB that is recommended by ProfilingAutoTuner.

cc: @amahussein

[amahussein]

Many of these recommendations require a separate recommendation when AutoTuner is run on CPU event logs (i.e. via Qual Tool) vs when AutoTuner is run on GPU event logs (i.e. via Profiling Tool).

I plan to use a class based approach to create QualAutoTuner and ProfilingAutoTuner that extend the base AutoTuner. Now we can have overrides as required. E.g. QualAutoTuner can override BATCH_SIZE_BYTES to be 1 GB as compared to 2 GB that is recommended by ProfilingAutoTuner.

cc: @amahussein

Yes, the plan was to do do that when I added the QualAutoTuner, but I did not get the bandwidth to complete it.
Keep me posted with the plan because I remember there was a pass to accomplish that.