Training a language model suite classically requires training each model separately and serving them independently. We improve both training and inference efficiency by stacking sub-models of increasing size into a single nested architecture trained end-to-end. This Matryoshka training framework reduces the total parameter count of the suite, enables low-cost distillation from the largest to all smaller sub-models at every training step, and is well-suited for speculative decoding as the draft model is contained within the verifier. We validate our approach by training a Matryoshka suite comprising 500M, 1.5B, and 3B sub-models. Our suite is on par with independently trained baselines on benchmark performance and validation and out-of-domain perplexities, while using 36% less training compute and improving the throughput of speculative decoding by 14-26%. We also ablate key architectural choices, offering guidance for building strong Matryoshka LM suites.
Kevo is a co-evolving framework that jointly learns to construct a structured knowledge base and reason over it for knowledge-intensive question answering, which leads to larger, better-connected knowledge structures with higher answer reachability, while also improving compositional factual reasoning and controllability compared to standard retrieval baselines.
Ryan Thomas Noonan, Lin-Xi Zhao, Meng-Han Xu et al.· 0 citations
This work proposes continuous-query LMLM (CO-LMLM), where the KB pairs continuous keys with textual knowledge values, a significant departure from prior reliance on relational KB and queries, which outperforms prior LMLMs and vanilla LLMs in both perplexity and factual precision.
Yair Feldman, Lin-Xi Zhao, Nathan Godey et al.· 0 citations