Tensorizing Engram: Sharing Latents Across N-Gram Embeddings is Beneficial in LLMs
About
Modern language models represent text using discrete token-level embeddings, which forces recurring multi-token patterns to be learned implicitly across Transformer layers. Both Over-tokenized Transformers and Engram attempt to address this limitation by explicitly incorporating multi-token (n-gram) memories. However, they rely on separate hash tables for each n-gram order, which introduces hash collisions and prevents nested n-grams from sharing the underlying latent structures. To address these issues, we propose Tensorized Engram (TN-gram), a compact memory module that represents tensorized n-gram embeddings through shared factors in the Canonical Polyadic (CP) form. TN-gram learns shared token-position factors together with order-absorption vectors to encode the embeddings of different n-gram order. Comprehensive experiments demonstrate that TN-gram matches or even outperforms Engram-style n-gram modules while requiring much fewer parameters.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Commonsense Reasoning | WinoGrande | Accuracy53.04 | 1581 | |
| Conversational Question Answering | CoQA | Accuracy8.92 | 35 | |
| Coreference Resolution | Winogender | Accuracy56.41 | 15 | |
| Multiple-choice Question Answering | Multiple-choice tasks (HS-ZS, ARC, COPA, CSQA, PIQA, OBQA, HS, LSAT-AR, BoolQ, LangID) | Average Accuracy37.82 | 6 | |
| Diverse Reasoning | BIG-Bench Hard | BBH QA Score12.32 | 6 |