Xuanwu Yin#
Xuanwu Yin leads the model optimization team, driving work on model quantization, sparsity, speculative decoding, and efficient training/inference across multiple platforms. His team delivers high-performance, production-ready solutions for large language models, vision-language models, and image/video-generation pipelines, while providing direct support to customers.
Posts by Xuanwu Yin
Athena-PRM: Enhancing Multimodal Reasoning with Data-Efficient Process Reward Models
Learn how to utilize a data-efficient Process Reward Model to enhance the reasoning ability of the Large Language/Multimodal Models.
Breaking the Accuracy-Speed Barrier: How MXFP4/6 Quantization Revolutionizes Image and Video Generation
Explore how MXFP4/6, supported by AMD Instinct™ MI350 series GPUs, achieves BF16-comparable image and video generation quality.
SparK: Query-Aware Unstructured Sparsity with Recoverable KV Cache Channel Pruning
In this blog we will discuss SparK, a training-free, plug-and-play method for KV cache compression in large language models (LLMs).
Týr-the-Pruner: Search-based Global Structural Pruning for LLMs
This blog introduces Týr-the-Pruner, a search-based, end-to-end framework for global structural pruning of large language models (LLMs).
Gumiho: A New Paradigm for Speculative Decoding — Earlier Tokens in a Draft Sequence Matter More
Gumiho boosts LLM inference with early-token accuracy, blending serial + parallel decoding for speed, accuracy, and ROCm-optimized deployment.
Technical Dive into AMD's MLPerf Inference v5.1 Submission
In this blog, we share the technical details of how we accomplish the results in our MLPerf Inference v5.1 submission.
Slim Down Your Llama: Pruning & Fine-Tuning for Maximum Performance
This blog describes the technical details of how we prune and fine tune the Llama 3.1 405B model in our MLPerf Inference v5.1 submission.
Reproducing the AMD Instinct™ GPUs MLPerf Inference v5.1 Submission
In this blog, we will provide step by step instruction on how to reproduce AMD's MLPerf Inference v5.1 Submission
Introducing AMD EVLM: Efficient Vision-Language Models with Parameter-Space Visual Conditioning
A novel approach that replaces visual tokens with perception-conditioned weights, reducing compute while maintaining strong vision-language performance.