26 citations · 50 across the 19 of their papers we have counts for
5 papers · 1 filter
How Sensitive Are Safety Benchmarks to Judge Configuration Choices?
Xinran Zhang
Safety benchmarks such as HarmBench rely on LLM judges to classify model responses as harmful or safe, yet the judge configuration, namely the combination of judge model and judge…
Atomic and Holistic LLM Judges for Reference-Grounded Support Labels: A Prompt-Controlled Comparison
Xinran Zhang
When an LLM judge only has to assign a three-way support label to a candidate answer given a reference, does asking it to decompose the answer into atomic claims help, and at what…
Beyond Creed: A Non-Identity Safety Condition A Strong Empirical Alternative to Identity Framing in Low-Data LoRA Fine-Tuning
Xinran Zhang
How safety supervision is written may matter more than the explicit identity content it contains. We study low-data LoRA safety fine-tuning with four supervision formats built from…
Lingxi: A Diversity-aware Chinese Modern Poetry Generation System
Xinran Zhang, Maosong Sun, Jiafeng Liu +1
Poetry generation has been a difficult task in natural language processing. Unlike plain neural text generation tasks, poetry has a high requirement for novelty, since an easily-un…
Improving Diversity of Neural Text Generation via Inverse Probability Weighting
Xinran Zhang, Maosong Sun, Jiafeng Liu +1
The neural text generation suffers from the text degeneration issue such as repetition. Traditional stochastic sampling methods only focus on truncating the unreliable "tail" of th…