Papers by Yifeng Gao
Stop Looking for “Important Tokens” in Multimodal Language Models: Duplication Matters More (2025.emnlp-main)
Copied to clipboard
Zichen Wen, Yifeng Gao, Shaobo Wang, Junyuan Zhang, Qintong Zhang, Weijia Li, Conghui He, Linfeng Zhang
| Challenge: | Vision tokens in multimodal large language models often dominate computational overhead due to excessive length compared to linguistic modality. |
| Approach: | They propose a token pruning method which defines an importance criterion for vision tokens and prunes the unimportant vision token during inference. |
| Outcome: | The proposed method can prune 88.9% of vision tokens while maintaining comparable performance. |
Token Pruning in Multimodal Large Language Models: Are We Solving the Right Problem? (2025.findings-acl)
Copied to clipboard
| Challenge: | Multimodal large language models have shown remarkable performance for cross-modal understanding and generation, yet suffer from severe inference costs. |
| Approach: | They propose to prune redundant tokens in MLLMs to reduce computation and storage costs. |
| Outcome: | The proposed method reduces the computational and storage costs of MLLMs by identifying redundant tokens and pruning them. |
Stability Implies Redundancy: Delta Attention Selective Halting for Efficient Long-Context Prefilling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to reduce sequence length rely on heuristics that break compatibility with hardware-efficient kernels like FlashAttention. |
| Approach: | They propose a method that selectively halts stabilized tokens by monitoring layer-wise update dynamics of the self-attention mechanism. |
| Outcome: | The proposed method can reduce prefill complexity while preserving model accuracy and hardware efficiency. |
SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) achieve strong reasoning with Chain-of-Thought prompting, but long and redundant traces substantially increase inference cost. |
| Approach: | They propose a training-free and plug-and-play decoding method that reduces CoT overhead without modifying model weights. |
| Outcome: | Experiments on GSM8K, MMLU, GPQA, and BBH show that SyncThink achieves 62.00% average Top@1 accuracy using 656 generated tokens and 28.68s latency, compared to 61.22%, 2141 tokens, and 92.01s for full CoT decoding. |