Curious about the torch.compile baseline in benchmark_litgpt

Was curious to know why the [`reduce-overhead` option was not used as a baseline](https://github.com/Lightning-AI/lightning-thunder/blob/3bde9e695385ba37fb365c5ed1cd1cc71995aba2/thunder/benchmarks/benchmark_litgpt.py#L261) and if there's a comparison somewhere using `reduce-overhead` ?

Thanks !

cc @apaz-cli

Start at thunder/benchmarks/benchmark_litgpt.py around line 261 and inspect how the torch.compile baseline is configured. Check whether a reduce-overhead comparison already exists; done would be a documented explanation of the baseline choice and, if absent, a reproducible comparison.

Written by the indexing model from the issue text.