Knowledge distillation (KD) is widely used for compressing a teacher model to a smaller student model, reducing its inference cost and memory footprint while preserving model capabilities. However, current KD methods for auto-regressive sequence models (e.g., large language models) suffer from missing a standardized objective function. Moreover, the recent use of student-generated outputs to address training-inference mismatches has significantly escalated computational costs. To tackle these issues, we introduce DistiLLM, a more effective and efficient KD framework for auto-regressive language models. DistiLLM comprises two components: (1) a novel skew Kullback-Leibler divergence loss, where we unveil and leverage its theoretical properties, and (2) an adaptive off-policy approach designed to enhance the efficiency in utilizing student-generated outputs. Extensive experiments, including instruction-following tasks, demonstrate the effectiveness of DistiLLM in building high-performing student models while achieving up to 4.3$\times$ speedup compared to recent KD methods.

We introduce a heuristic scheduling algorithm for real-time adaptive traffic signal control to reduce traffic congestion. This algorithm adopts a lane-based model that estimates the arrival time of all vehicles approaching an intersection through different lanes, and then computes a schedule (i.e., a signal timing plan) that minimizes the cumulative delay incurred by all approaching vehicles. State space, pruning checks and an admissible heuristic for A* search are described and shown to be capable of generating an intersection schedule in real-time (i.e., every second). Due to the effectiveness of the heuristics, the proposed approach outperforms a less expressive Dynamic Programming approach and previous A*-based approaches in run-time performance, both in simulated test environments and actual field tests.