Results 3 comments of Shehper

So, depending on the GPU you have, you should edit the line defining [flops_promised in estimate_mfu.](https://github.com/karpathy/nanoGPT/blob/eba36e84649f3c6d840a93092cb779a260544d08/model.py#L301) nanoGPT was trained on A100 GPUs, so the code compares the speed of training...

I dug in a bit more and am sharing my findings here. After updating MacOS to version 13.6, the error shared above disappeared. However, `test_interactive_model.py` still fails on mps backend....

A related comment: In `ppo_procgen.py`, episodic information is biased in favor of the first of all episodes by the `break` statement. [Link](https://github.com/vwxyzjn/cleanrl/blob/1ed80620842b4cdeb1edc07e12825dff18091da9/cleanrl/ppo_procgen.py#L244) ``` for item in info: if "episode" in...