Comment by groundzeros2015

So do you think your implementation is better? Or there are finer parts of the spec it’s missing?

Your llm has JPEG xl programs in the training data. You copied from GitHub.