TVM
PulseAugur coverage of TVM — every cluster mentioning TVM across labs, papers, and developer communities, ranked by signal.
-
GLOW-FM: One-Step Latent Flow Matching for Efficient Low-Light Image Enhancement
Researchers have developed GLOW-FM, a novel framework for efficient low-light image enhancement using a one-step latent flow matching approach. This method operates within the latent space of a Tiny-VAE and employs Term…
-
New method accelerates Bayesian Neural Networks with single forward pass
Researchers have developed a new method to accelerate Bayesian Neural Networks (BNNs) by using a single Probabilistic Forward Pass (PFP) and code generation. This approach significantly reduces the computational cost as…
-
New framework and agent tackle LLM kernel generation for edge devices
Researchers have developed EdgeKernelBench, a framework for evaluating LLM-generated kernels on diverse edge devices, and LEKA, a layered optimization agent. While LLMs can generate kernels with good performance on serv…
-
New framework FBLayout boosts mobile LLM fine-tuning efficiency
Researchers have developed FBLayout, a new framework designed to optimize memory layout for efficient fine-tuning of large language models on mobile GPUs. This approach addresses the memory constraints and layout transf…
-
InstantSfM offers GPU-native structure-from-motion for deep learning era
Researchers have developed InstantSfM, a novel GPU-native structure-from-motion system designed to integrate seamlessly with deep learning pipelines. This system addresses the limitations of traditional CPU-centric SfM …
-
Run Neural Networks on GPU via Vulkan: Libraries, Compilers, or Custom Engines
This article outlines three methods for running trained neural networks on a GPU using the Vulkan API. It suggests integrating existing libraries like TensorFlow Lite or ONNX Runtime, compiling models via ML compilers s…
-
AI agents automate data prep, while new Python ML compiler speeds LLM compression
Researchers have developed a new open-source machine learning compiler stack written in just 5,000 lines of Python. This stack offers unprecedented transparency by lowering large language models to CUDA with six interme…