[BUG] Transformers model padding should be to "longest"

[pingzhili]

Describe the bug

transformers model padding is not efficient,

lighteval/src/lighteval/models/transformers/transformers_model.py

Line 863 in 75f4429

padding="max_length", # we pad to the longest sequence

As the comments say, "we pad to the longest sequence", it should be padding="longest"

To Reproduce

lighteval accelerate \ "pretrained=gpt2" \ "leaderboard|truthfulqa:mc|0|0"

Expected behavior

The initial sequence length does not have to be that long (max length)

Version info

0.8.0

[NathanHB]

Hey ! Being fixed in #669