Hao Peng · page 2
Papers
- Pre-training Distillation for Large Language Models: A Design Space Exploration (2025 · 14 citations)
Knowledge distillation for large language models is usually applied after pre-training, when a student is fine-tuned on responses written by a stronger teacher. - MAVEN-FACT: A Large-scale Event Factuality Detection Dataset (2024 · 6 citations)
Event Factuality Detection (EFD) asks whether a mentioned event actually happened, might happen, or did not happen. - MixKVQ: Query-Aware Mixed-Precision KV Cache Quantization for Long-Context Reasoning (2026 · 5 citations)
Reasoning-oriented LLMs produce very long Chain-of-Thought outputs, so their KV cache consumes large amounts of GPU memory and bandwidth. - LLMs are Vulnerable to Malicious Prompts Disguised as Scientific Language (2025 · 5 citations)
The paper asks whether safety-aligned LLMs can be pushed into producing stereotyped or toxic output when a harmful request is framed in academic terms. - Activation-Guided Local Editing for Jailbreaking Attacks (2026 · 1 citation)
The paper addresses the weaknesses of current jailbreak techniques used to red-team aligned LLMs. Token-level attacks tend to yield gibberish suffixes that filters catch easily and that rarely transfer to other models. Prompt-level attacks depend on manual crafting or costly iterative search. The authors present… - WildReward: Learning Reward Models from In-the-Wild Human Interactions (2026)
Reward models for LLM training usually depend on large sets of human-annotated preference pairs, which are expensive to collect.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/hao-peng/page/2/