Best paper award at SKILL24
Published:
Best paper award for: Uncovering the Token Splitting Effect in Soft Prompts for Multi-Model LLM Training
Posts by Collection
awards
portfolio
Portfolio item number 1
Short description of portfolio item number 1
Portfolio item number 2
Short description of portfolio item number 2
projects
BERTchen: Training the best and most efficient German BERT model
Published:
We investigate ways to increase the efficiency of BERT pretraining in order to create the most efficient German BERT model possible.
Uncovering the Token Splitting Effect in Soft Prompts for Multi-Model LLM Training
Published:
We investigate how multi-model tuning affects the interpretability and transferability of soft prompts.
Learning to Predict Usage Options of Product Reviews with LLM-Generated Labels
Published:
We evaluate the annotation quality of different labeling sources on a custom task.
Benchmarking the environmental footprint of BERT models
Published:
We benchmarked the power consumption and CO2e emission of pretraining BERT models.
Transforming Sorted Arrays to Binary Search Trees using Reinforcement Learning
Published:
We investigate how methods of reinforcement learning can be used for algorithm discovery using LLMs.
publications
Uncovering the Token Splitting Effect in Soft Prompts for Multi-Model LLM Training
Published in SKILL24, 2024
We investigate how multi-model tuning affects the interpretability and transferability of soft prompts. Best paper award at SKILL 2024.
Learning to Predict Usage Options of Product Reviews with LLM-Generated Labels
Published in Arxiv, 2024
We evaluate the annotation quality of different labeling sources on a custom task.
CooperBench: Benchmarking Cooperation in Coding Agents
Published in COLM 2026, 2026
A benchmark of over 600 collaborative coding tasks, showing that state-of-the-art coding agents succeed less often when they work together than when they work alone.
Prompt-Robust Language Models: Which Training Strategies Work?
Published in Findings of EMNLP 2026, 2026
We reproduce and compare prompt-robustness training strategies under controlled conditions and diagnose why they fall short.