Organizations: University of Zurich · ETH Zurich · Deutsche Bank Securities
Abstract
A simple way to improve the performance of almost any machine learning model is not to train a single but several models with diverse algorithms which will make slightly distinct kinds of predictions and errors on the same data, and thus improve the average predictions and robustness. However, making predictions using a whole ensemble of models is cumbersome and computationally too expensive to allow deployment to a large number of users, especially if the models are large neural nets. In response to this, we introduce a layer and point wise projection mapping, which maps student and teacher representations into an aligned high-dimensional embedding space during training process. The proposed approach combined with LoRA injection reduces the student model trainable parameters to less than 1% of the teacher model, while significantly improving word error rate (WER) compared to other distillation methods, as demonstrated in ablation studies. Unlike a mixture of experts, our method can be trained rapidly and in parallel.
A compressed student has two shapes that need not agree: the weight it deploys at inference and the weight family its training can reach. We show that a state-of-the-art weight-inheritance distiller, Low-Rank Clone (LRC), deploys a full-width student MLP but ties training to a teacher-induced slice, leaving 62.5-81.4% of each deployed matrix's independent linear degrees of freedom unreachable-paid for at inference, never trainable. Our principle is one line: train what you deploy. From the identical LRC warm start, we make the training object the entire deployed matrix, with no change in deployed shape, deployed parameter count, or inference FLOPs, via two mergeable realizations (Dense-LRC and CORE-LRC) that both collapse to one deployed weight. This recovers stranded capacity: taking the stronger realization per teacher, +2.36/+2.71/+10.45 Avg9 over matched-budget plain-LRC baselines across three teachers (Llama3.2-3B, Llama3.1-8B, Qwen2.5-3B), with the largest gain on the widest teacher (Qwen), where it reaches the original recipe's approx. 20B-token accuracy at 10B tokens (2x token efficiency); there the strictly same-lineage arm still recovers +6.39, the fully controlled figure. Controls strongly support attributing the gain to the enlarged reachable set, rather than to added parameters or the recipe. From approx. 10B distillation tokens plus a short SFT, a half-parameter 1.5B student matches its approx. 9T-token teacher's 9-task macro-average, within evaluation noise and with a residual MMLU deficit, and a 2.7B student beats Meta's own official compression of Llama3.1-8B at ~900x fewer compression tokens (a token count under unmatched recipes, not a compute claim). All results are from single-seed runs on the LRC backbone.
Logit-based knowledge distillation for autoregressive language models usually aligns teacher and student next-token distributions over the entire vocabulary. However, this global objective overlooks relative preferences among likely token alternatives. Existing local approaches often select candidate tokens from either the teacher or the student alone. Teacher-only selection can miss tokens that the student considers likely, while student-only selection can rely on an inaccurate ranking early in training. We propose Adaptive Local Relational Alignment (ALRA), a position-specific framework combining student proposals with teacher guidance. At each valid prediction position, the student proposes likely tokens, while the teacher's most probable token is included as an anchor. ALRA adjusts the number of selected tokens according to how broadly the teacher distributes probability within this candidate set relative to the current batch. Adaptive Local Divergence retains the mass-matching term and separately matches the relative token distributions within the selected and remaining vocabulary regions. Unlike the exact full-vocabulary decomposition, it replaces the teacher-mass coefficients of the two conditional terms with unit coefficients, preventing either term from being downweighted solely because its region has low teacher probability. Student-Weighted Pairwise Relational Alignment emphasizes high-probability token pairs with small student probability gaps and gives less weight to unlikely or clearly separated pairs. Experiments on The Pile with randomly initialized 200M- and 500M-parameter students across nine zero-shot benchmarks yield average accuracies of 36.62% and 37.40%. ALRA exceeds the strongest competing distillation baseline by 0.94 and 0.83 percentage points and improves over pre-training without distillation by 2.31 and 2.91 points, respectively.
Quang Hoang Trung, Quang Huu Hieu, Nguyen Van Hoang Phuc +1
We study distillation for large language models under explicit compute constraints, with the goal of producing student models that are not only cheaper to train, but structurally efficient at inference time. While prior approaches to parameter-efficient distillation, such as LoRA, reduce adaptation cost, they leave the dense backbone unchanged and therefore fail to deliver meaningful inference savings. We propose Budgeted LoRA, a distillation framework that treats model compression as a structured compute allocation problem. Instead of using a fixed student architecture, we introduce a global compute budget that sets the final target fraction of dense computation retained. Under this constraint, the model redistributes capacity across dense and low-rank pathways via (i) module-level dense retention coefficients, (ii) adaptive low-rank allocation, and (iii) post-training compression that selectively removes, approximates, or preserves dense components. This formulation yields a family of students controlled by a single budget dial. Empirically, Budgeted LoRA matches standard LoRA perplexity at a moderate budget with a 1.74x compressed-module speedup; at an aggressive budget it achieves a 4.05x speedup with moderate perplexity degradation, and it preserves higher accuracy on function-style in-context learning probes. These results suggest that, under compute-constrained distillation, retaining behavior is less about matching perplexity or removing more parameters than it is about controlling how dense computation is transferred to low-rank pathways.