fixes to matmul and linear benchmarks

[vkuzo]

Stack from ghstack (oldest at bottom):

• -> #320

Summary:

for matmul benchmarks, unbreaks them - we need the scales to be fp32, not integers

for linear benchmarks, aligns default settings to current best supported path (compile on, dynamic scaling)

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: D59877198

[vkuzo]

@vkuzo has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

This pull request has been merged in pytorch-labs/float8_experimental@ec8b46cda737cb72c0769eba42341edf50111e22.