6 papers
Decoupling the "What" and "Where" With Polar Coordinate Positional Embeddings
Anand Gopalakrishnan, Robert Csordás, Jürgen Schmidhuber +1
The attention mechanism in a Transformer architecture matches key to query based on both content -- the what -- and position in a sequence -- the where. We present an analysis indi…
Correcting Gradient-Based Circuit Localization via Interaction-Aware Backpropagation
Joakim Edin, Casper L. Christensen, Róbert Csordás +5
Circuit localization methods aim to identify the subset of model components responsible for specific behaviors in large language models, enabling detailed mechanistic analysis. Mos…
Thoughtbubbles: an Unsupervised Method for Parallel Thinking in Latent Space
Houjun Liu, Shikhar Murty, Christopher D. Manning +1
Current approaches for scaling inference-time compute in transformers train them to emit explicit chain-of-thought tokens before producing an answer. While these methods are powerf…
Do Language Models Use Their Depth Efficiently?
Róbert Csordás, Christopher D. Manning, Christopher Potts
Modern LLMs are increasingly deep, and depth correlates with performance, albeit with diminishing returns. However, do these models use their depth efficiently? Do they compose mor…
MrT5: Dynamic Token Merging for Efficient Byte-level Language Models
Julie Kallini, Shikhar Murty, Christopher D. Manning +2
Models that rely on subword tokenization have significant drawbacks, such as sensitivity to character-level noise like spelling errors and inconsistent compression rates across dif…
Recurrent Neural Networks Learn to Store and Generate Sequences using Non-Linear Representations
Róbert Csordás, Christopher Potts, Christopher D. Manning +1
The Linear Representation Hypothesis (LRH) states that neural networks learn to encode concepts as directions in activation space, and a strong version of the LRH states that model…