Summary of Is It a Free Lunch For Removing Outliers During Pretraining?, by Baohao Liao et al.
Is It a Free Lunch for Removing Outliers during Pretraining?
by Baohao Liao, Christof Monz
First submitted to arxiv on: 19 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The novel softmax function introduced in the paper “qtransformer” aims to pretrain language models in an outlier-free manner, making them more suitable for quantization. However, this approach was found to degrade performance when used with full precision. To address this, the authors enhanced the method by ensuring its normalization is invariant to sequence length, which bridges the gap between pretraining and fine-tuning. This improved method also allows for successful pretraining of causal language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are getting bigger, and that’s a problem! Quantization helps make them smaller, but sometimes it makes things go wrong. A special kind of math called “softmax” was recently developed to help with this. It works pretty well, but only if you’re doing things in a special way. If you don’t do things just right, it can actually make things worse! To fix this, the scientists came up with an even better version that takes into account how long the sequence of words is. This helps make sure things work smoothly when we go from training to testing. |
Keywords
» Artificial intelligence » Fine tuning » Precision » Pretraining » Quantization » Softmax
