text-generation-inference

8b4cd2a9 - fix: skip cuda graphs that will oom and improve free memory logging