Summary of Ffn-skipllm: a Hidden Gem For Autoregressive Decoding with Adaptive Feed Forward Skipping, by Ajay Jaiswal et al.
FFN-SkipLLM: A Hidden Gem for Autoregressive Decoding with Adaptive Feed Forward Skipping
by Ajay Jaiswal, Bodun Hu, Lu Yin, Yeonju Ro, Shiwei Liu, Tianlong Chen, Aditya Akella
First submitted to arxiv on: 5 Apr 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
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 Autoregressive Large Language Models (LLaMa, GPTs) have achieved remarkable success in language understanding and generation. However, these models typically require substantial computational resources for token-by-token generation, which presents a challenge. To mitigate this issue, early-exit and layer-dropping strategies have been proposed. Despite some promising results on metrics like Rough-L/BLUE, our evaluation reveals issues such as generation collapse, hallucination of wrong facts, and performance drops even at trivial exit ratios. We attribute these errors to ineffective handling of the KV cache through state copying during early-exit. This paper proposes FFN-SkipLLM, a novel fine-grained skip strategy for autoregressive LLMs that can skip 25-30% of feed-forward blocks with minimal impact on performance on knowledge-intensive generation tasks. Our extensive experiments across benchmarks like MT-Bench, Factoid-QA, and text summarization demonstrate how this method facilitates faster autoregressive decoding. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are super smart computers that can understand and generate human-like language. But they need a lot of computing power to do so. To make them more efficient, researchers have tried different strategies to reduce the amount of computation needed. However, these efforts haven’t quite worked out as hoped. The models still struggle with generating coherent text and often produce incorrect information. This paper proposes a new approach called FFN-SkipLLM that helps large language models generate text faster without sacrificing too much accuracy. We tested this method on various tasks like machine translation and text summarization, and it showed promising results. |
Keywords
* Artificial intelligence * Autoregressive * Hallucination * Language understanding * Llama * Summarization * Token * Translation
