Token Erasure as a Footprint of Implicit Vocabulary Items in LLMs
This work addresses a fundamental mechanistic question in LLM interpretability for researchers, though it appears incremental as it builds on existing probing techniques.
The study tackled the problem of how LLMs convert arbitrary groups of tokens into higher-level representations by identifying a 'token erasure' effect in early layers, and proposed a method to read out the implicit vocabulary, demonstrating results on Llama-2-7b and Llama-3-8B models.
LLMs process text as sequences of tokens that roughly correspond to words, where less common words are represented by multiple tokens. However, individual tokens are often semantically unrelated to the meanings of the words/concepts they comprise. For example, Llama-2-7b's tokenizer splits the word "northeastern" into the tokens ['_n', 'ort', 'he', 'astern'], none of which correspond to semantically meaningful units like "north" or "east." Similarly, the overall meanings of named entities like "Neil Young" and multi-word expressions like "break a leg" cannot be directly inferred from their constituent tokens. Mechanistically, how do LLMs convert such arbitrary groups of tokens into useful higher-level representations? In this work, we find that last token representations of named entities and multi-token words exhibit a pronounced "erasure" effect, where information about previous and current tokens is rapidly forgotten in early layers. Using this observation, we propose a method to "read out" the implicit vocabulary of an autoregressive LLM by examining differences in token representations across layers, and present results of this method for Llama-2-7b and Llama-3-8B. To our knowledge, this is the first attempt to probe the implicit vocabulary of an LLM.