This paper proposes a recurrent structure to mimic the auto-regressive Transformers such that their quadratic time complexity is reduced to linear.
They replace the softmax attention heads with a MLP-dot one and limit the later tokens to attend to only previous ones.
Comments
- The title could be phrased as “compression” or “distillation” instead of “conversion” as their method is very lossy.
- Although they claimed any pre-trained Transformer can be “converted”, they only experimented on a toy Transformer trained on WikiText and they didn’t experiment on any real world Transformers like BART or T5.
Rating
- 5: Transformative: This paper is likely to change our field. It should be considered for a best paper award.
- 4.5: Exciting: It changed my thinking on this topic. I would fight for it to be accepted.
- 4: Strong: I learned a lot from it. I would like to see it accepted.
- 3.5: Leaning positive: It can be accepted more or less in its current form. However, the work it describes is not particularly exciting and/or inspiring, so it will not be a big loss if people don’t see it in this conference.
- 3: Ambivalent: It has merits (e.g., it reports state-of-the-art results, the idea is nice), but there are key weaknesses (e.g., I didn’t learn much from it, evaluation is not convincing, it describes incremental work). I believe it can significantly benefit from another round of revision, but I won’t object to accepting it if my co-reviewers are willing to champion it.
- 2.5: Leaning negative: I am leaning towards rejection, but I can be persuaded if my co-reviewers think otherwise.
- 2: Mediocre: I would rather not see it in the conference.
- 1.5: Weak: I am pretty confident that it should be rejected.
- 1: Poor: I would fight to have it rejected.
0 投票人