Summary of Sparsepo: Controlling Preference Alignment Of Llms Via Sparse Token Masks, by Fenia Christopoulou et al.
SparsePO: Controlling Preference Alignment of LLMs via Sparse Token Masks
by Fenia Christopoulou, Ronald Cardenas, Gerasimos Lampouras, Haitham Bou-Ammar, Jun Wang
First submitted to arxiv on: 7 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); 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 The proposed SparsePO method for Preference Optimization (PO) in language models aims to learn how to weight reward and KL divergence contributions at the token level. By introducing flexible objectives, this approach allows for automatic learning of token weights that best align with human-desired behaviors. The paper presents two variants of weight-masks, one derived from the reference model itself and another learned on the fly. Experimental results demonstrate improved performance in sentiment control, dialogue, text summarization, and text-to-code generation tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Language models can be trained to follow human preferences using Preference Optimization (PO). The current method focuses on all tokens contributing equally to the loss function. However, human preference is not affected by each word equally, but rather depends on specific words or phrases. The proposed SparsePO method introduces flexible objectives that allow for automatic learning of token weights. This approach assigns meaningful weights to tokens according to the target task and improves performance in various tasks. |
Keywords
» Artificial intelligence » Loss function » Optimization » Summarization » Token
