Introducing GEAK, An AMD AI-driven Triton kernel generation framework and evaluation benchmarks. GEAK leverages frontier LLMs with inference-time scaling to automatically produce efficient, accurate GPU kernels for AMD Instinct GPUs, achieving up to 63% correctness and 2.59× speedups over reference implementations.

An end-to-end autonomous research workflow is meant to assist you as the human researcher toward implementing your research ideas.

A LLM-empowered agent that can complete various tasks in Minecraft automatically. It enhances adaptability and generalization by integrating online RL training with a multi-expert memory module. Experiments show that AMD MoEA framework outperforms state-of-the-art methods on MineDOJO tasks.

An improved FFN (IFFN) module for vision transformers that uses AGeLU function and multiple instances to enhance non-linearity, reducing hidden dimensions and computational cost.

QT-ViT replaces softmax-based attention with second-order Taylor expansion, accelerating it via a fast approximation algorithm. It achieves superior performance without knowledge distillation or high-order attention residuals, outperforming previous models.

FDViT employs a flexible downsampling layer to reduce feature map sizes smoothly. Combined with a masked auto-encoder for training, it decreases redundant calculations and information loss.

A domain shift diffusion-based SR model that capitalizes on the generative powers of pretrained diffusion models while significantly enhancing efficiency by initiating the diffusion process with low-resolution (LR) images.

A reward-guided approach that directly learns negative embeddings through gradient descent. The negative embeddings exhibit strong generalization capabilities and can be seamlessly adaptable to T2I and T2V models.

Using an enhanced post-training approach based on intermediate layer distillation and optimized layer selection, our models dramatically reduce KV-cache requirements — without compromising quality.     

A continuous image tokenizer that leverages soft categorical posteriors to aggregate multiple codewords into each latent token,

increasing the representation capacity of the latent space.

An efficient post-training approach to convert MHA models to multi-head latent attention (MLA) using knowledge distillation.

Gumiho combines serial and parallel heads for SPD. Early tokens use sophisticated Transformer architecture serially, later ones use lightweight MLP heads in parallel. Experiments show it outperforms existing methods.

Multimodal speculative decoding enhances inference by parallelizing token prediction and verification across cross-modal contexts, achieving higher acceptance lengths and up to 3× speedups in structured visual-text interpretation tasks.