Cmprsr: Abstractive Token-Level Question-Agnostic Prompt Compressor

Motivated by the high costs of using black-box Large Language Models (LLMs), we introduce a novel prompt compression paradigm, under which we use smaller LLMs to compress inputs for the larger ones. We present the first comprehensive LLM-as-a-compressor benchmark spanning 25 open- and closed-source models, which reveals significant disparity in models' compression ability in terms of (i) preserving semantically important information (ii) following the user-provided compression rate (CR). We further improve the performance of gpt-4.1-mini, the best overall vanilla compressor, with Textgrad-based compression meta-prompt optimization. We also identify the most promising open-source vanilla LLM - Qwen3-4B - and post-train it with a combination of supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), pursuing the dual objective of CR adherence and maximizing the downstream task performance. We call the resulting model Cmprsr and demonstrate its superiority over both extractive and vanilla abstractive compression across the entire range of compression rates on lengthy inputs from MeetingBank and LongBench as well as short prompts from GSM8k. The latter highlights Cmprsr's generalizability across varying input lengths and domains. Moreover, Cmprsr closely follows the requested compression rate, offering fine control over the cost-quality trade-off.

Packed-Ensemble Surrogate Models for Fluid Flow Estimation Arround Airfoil Geometries

Physical based simulations can be very time and computationally demanding tasks. One way of accelerating these processes is by making use of data-driven surrogate models that learn from existing simulations. Ensembling methods are particularly relevant in this domain as their smoothness properties coincide with the smoothness of physical phenomena. The drawback is that they can remain costly. This research project focused on studying Packed-Ensembles that generalize Deep Ensembles but remain faster to train. Several models have been trained and compared in terms of multiple important metrics. PE(8,4,1) has been identified as the clear winner in this particular task, beating down its Deep Ensemble conterpart while accelerating the training time by 25%.