Xuanwu Yin

Senior Manager@Advanced Micro Devices·中国·OpenReview

研究方向

model inference acceleration · speculative decoding · model optimization · quantization · pruning · sparsity · distillation · imaging algorithm · camera algorithm · display algorithm

6.5

Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match

ICLR 2026Poster

5.0

Athena: Enhancing Multimodal Reasoning with Data-efficient Process Reward Models

ICLR 2026Rejected

4.0

Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization

NeurIPS 2025Poster

7.2

Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding

ICML 2025Poster

4.4

Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE

合作者 (20)

Xuanwu Yin

Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match

Athena: Enhancing Multimodal Reasoning with Data-efficient Process Reward Models

SpecVLM: Fast Speculative Decoding in Vision-Language Models

BDQ: Bidirectional Diagonal Quantization for LLMs

Dual LoRA: Enhancing LoRA with Magnitude and Direction Updates

Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization

Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding

Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE