PyTorch
PulseAugur coverage of PyTorch — every cluster mentioning PyTorch across labs, papers, and developer communities, ranked by signal.
13 天有情绪数据
-
AI代理自动化数据准备，而新的Python ML编译器加速LLM压缩
研究人员开发了一个仅用5000行Python编写的新开源机器学习编译器栈。该栈通过将大型语言模型降低到具有六个中间表示的CUDA，提供了前所未有的透明度。它旨在易于修改且针对CUDA进行了优化，与PyTorch或TVM等更复杂的系统形成对比。此外，AI代理因其自动化探索性数据分析和数据准备任务的潜力而受到关注，有望为数据科学家节省大量时间。
-
新的C++引擎HASE在多智能体强化学习训练中达到33M步/秒
研究人员开发了一种名为捉迷藏引擎 (HASE) 的新C++引擎，旨在显著提高在去中心化、部分可观察环境中的强化学习智能体训练效率。通过利用面向数据设计和优化的内存处理，HASE在单个智能体上实现了高达每秒3300万步的惊人吞吐量。该引擎大大缩短了多智能体策略的训练时间，使得复杂的协作行为能在几分钟内学会。
-
New algorithm speeds up EigenDecomposition for large matrices in deep learning
Researchers have developed a new batch-efficient algorithm for EigenDecomposition (ED), a critical computation in computer vision and deep learning. This divide-and-conquer approach aims to overcome the computational bo…
-
VkSplat 流水线通过 Vulkan 计算提升 3D 高斯溅射训练性能
研究人员开发了 VkSplat，一种利用 Vulkan 计算进行 3D 高斯溅射 (3DGS) 训练的新型流水线，可提高性能和兼容性。与传统的 CUDA 和 PyTorch 方法相比，这种新方法将速度提高了 3.3 倍，并将 VRAM 使用量减少了 33%。VkSplat 值得注意的是，它是第一个在不同 GPU 供应商上实现最先进结果的全 Vulkan 3DGS 训练流水线。
-
Neural ODEs advance with mixed precision training and causal forecasting methods
Researchers have developed a new mixed-precision training framework for Neural Ordinary Differential Equations (Neural ODEs) to reduce computational costs. This framework uses low-precision computations for evaluating n…
-
DeepSeek V4 First Release Adaptation Behind: Why does Ascend insist on not doing a CUDA compatibility layer?
Huawei's Ascend AI accelerators are forging a unique path by eschewing CUDA compatibility to build an independent ecosystem. This strategy focuses on deep architectural changes in their latest Ascend 950 chips to addres…
-
研究对NLP任务的AutoML和BiLSTM进行基准测试，结果好坏参半
研究人员比较了传统机器学习方法与深度学习模型在各种自然语言处理任务中的表现，包括细粒度情感分类和情感分析。研究使用了20种情感文本分类数据集和印度尼西亚电子商务评论等数据集。研究结果普遍表明，深度学习模型，特别是双向长短期记忆（BiLSTM）网络，通过更好地捕捉文本中的上下文细微差别，通常能获得更优越的性能。然而，传统的机器学习方法，如支持向量机和逻辑回归，在准确性方面仍然具有竞争力，并且在某些数据集上提供更高的计算效率。
-
New HDET method explores hyperparameters for large model training
Researchers have introduced Hyperparameter-Divergent Ensemble Training (HDET), a novel method designed to optimize the training of large neural networks. HDET repurposes data-parallel replicas to simultaneously explore …
-
IBM Research integrates vLLM into its RITS Platform for AI development
IBM Research has integrated vLLM, an open-source library for fast LLM inference, into its RITS Platform. This integration aims to enhance the platform's capabilities by leveraging vLLM's efficient processing for large l…
-
PointTransformerX offers portable, efficient 3D point cloud processing without sparse algorithms
Researchers have developed PointTransformerX (PTX), a new vision transformer backbone for processing 3D point clouds that eliminates the need for custom CUDA operators. This PyTorch-native model achieves competitive acc…
-
New Rose optimizer offers low VRAM, fast convergence, and great results
A new PyTorch optimizer named Rose has been released under the Apache 2.0 license. Developed by Matthew K., Rose is designed to be stateless, offering significantly lower VRAM usage compared to optimizers like AdamW, wi…
-
HubRouter offers sub-quadratic routing for sequence models, improving throughput
Researchers have developed HubRouter, a novel module designed to replace computationally expensive O(n^2) attention layers in sequence models with a more efficient O(nM) hub-mediated routing system. This new primitive u…
-
NVIDIA Cosmos Predict 2.5 fine-tuned for robots; new ShadowPEFT method emerges
NVIDIA has released a guide for fine-tuning its Cosmos Predict 2.5 world model for robot video generation using parameter-efficient techniques like LoRA and DoRA. This method allows for adaptation to specific domains, s…
-
Together AI Cloud 通过 TorchForge 和工具集成增强 RL 管道
Together AI 正在增强其云平台以支持高级强化学习 (RL) 管道，集成 TorchForge 和 Monarch 进行分布式训练。该平台现在提供低延迟 GPU 通信和异构调度，用于混合 CPU/GPU 工作负载，这对于复杂的 RL 任务至关重要。与 Together CodeSandbox 和 Code Interpreter 的新集成允许 RL 代理与工具交互并执行代码，从而将它们的能力扩展到传统的游戏场景之外。
-
Kimi K2 model boasts 1T parameters and SOTA HLE, while Soumith Chintala departs PyTorch
Kimi K2, a new model from Kimi, boasts 1 trillion parameters and achieves state-of-the-art results on the HLE benchmark. It also demonstrates capabilities in BrowseComp and TauBench. Separately, Soumith Chintala has dep…
-
Thinking Machines names Soumith Chintala new CTO, forms vLLM team
Thinking Machines has appointed Soumith Chintala as its new Chief Technology Officer, succeeding Barret Zoph. Chintala, who has a decade of experience in AI research and development, is expected to lead the company's te…
-
OCaml ecosystem Raven offers type-safe ML tools mirroring Python libraries
Raven is a new ecosystem of OCaml libraries designed for numerical computing, machine learning, and data science. It aims to provide type-safe alternatives to popular Python libraries such as NumPy, JAX, and PyTorch. Th…
-
FormalVerifML offers enterprise-grade formal verification for machine learning models
A new open-source framework called FormalVerifML has been released, utilizing Lean 4 for the formal verification of machine learning models. This tool aims to provide mathematically rigorous proofs of properties like ro…
-
Julia's Micrograd.jl series explores automatic differentiation for ML
This article introduces Micrograd.jl, a new automatic differentiation package for the Julia programming language. It aims to fill a gap in comprehensive tutorials for AD in Julia, requiring a solid understanding of both…
-
OpenAI's Code Interpreter is de facto GPT 4.5, experts suggest
The Code Interpreter feature within ChatGPT is being discussed as a significant advancement, potentially equivalent to a GPT 4.5 model. This tool allows ChatGPT to write and execute Python code within a sandboxed enviro…