Pytorch

Fibre optic sensing, such as distributed acoustic sensing (DAS), has become a widespread technology for geophysical studies. To process the large-scale datasets produced by DAS, several machine learning methods have been proposed. However, without standardization of data and models, these methods lack comparability and interoperability. This introduces a gap between model developers and practitioners analyzing DAS data and inhibits adoption of deep learning for DAS. To address these limitations, here we present SeisBench DAS, an extension to the SeisBench library for machine learning in seismology. SeisBench DAS defines standard formats for DAS benchmark datasets, including standardised metadata and labels, and DAS models. It builds on the xdas framework for data ingestion and virtual array handling, and on PyTorch for reading and applying the machine learning models. Importantly, SeisBench provides an engine to efficiently apply deep learning models to diverse formats of DAS data, bridging the gap between model developers and practitioners. SeisBench DAS is designed as an open and extensible framework, allowing to easily incorporate future developments in deep learning for DAS.

Architecture category. Hybrid method: a physics-based analytical pencil-beam (PB) dose engine followed by a 3-D convolutional residual-correction network (RepVGG-U-Net). We addressed the DoseRAD2026 proton dose-prediction task with PyDoseRT Proton, a GPU-accelerated engine implemented in PyTorch and augmented by a learned residual toward Monte Carlo (MC) accuracy. A double-Gaussian PB kernel was calibrated to GATE/Geant4 integrated depth doses in water in two stages: a classical per-energy curve fit, then a gradient-based fit of the full 3-D dose through the PyTorch physics engine as it retains a differentiable execution path for gradient-based optimization of dose-dependent objectives. The engine computes each beamlet on a beam's-eye-view (BEV) lattice with variance-preserving Gaussian splitting, an analytic nuclear halo, and a Fermi-Eyges heterogeneity term, then rotates the result into the patient frame. Additionally, a compact residual U-Net predicts an additive correction in BEV space. It is conditioned on voxelwise material-label embeddings, a discrete energy embedding and spot size. The same model was used for all anatomical sites (thoracic and abdominal). It was trained with a patient-space L1 objective emphasizing the scored high-dose region and multi-scale BEV deep supervision. The submitted CT configuration obtained preliminary-test beamlet MAE 0.0066, image-z IDD distance 0.0025, plan MAE 0.0049, 98.30% gamma pass rate (1%/1 mm), and DVH error 0.460.

In modern AI frameworks, GPU kernels are key to overall system performance. Combining usability, portability, and near-handwritten CUDA performance, Triton is widely adopted for implementing GPU kernels. Recent advances show the potential of large language models (LLMs) to automatically generate Triton kernels, reducing the manual effort required from expert kernel developers. Several benchmarks evaluate LLM-generated Triton kernels. However, they suffer from three key limitations: (1) they restrict tasks to PyTorch-to-Triton translation, failing to reflect the diversity and complexity of real-world Triton tasks; (2) they evaluate only individual-kernel performance rather than end-to-end performance, the core criterion for real-world deployment in AI frameworks; and (3) they rely on manually written evaluation scripts for individual kernels, which may contain flaws that models can exploit to bypass correctness checks and obtain inflated scores. To address these limitations, we introduce RealisticTritonBench, the first benchmark to derive Triton kernel generation tasks from real-world pull requests in popular AI frameworks, enabling realistic, production-like evaluation. RealisticTritonBench systematically extracts PRs that modify Triton kernels from popular open-source AI frameworks and transforms them into generation tasks with concrete engineering contexts. Each task takes a natural language requirement as input and requires a corresponding Triton kernel implementation, with a complete and reproducible evaluation environment. Unlike prior benchmarks focused on isolated kernel performance, RealisticTritonBench integrates generated kernels into their original frameworks and evaluates them using end-to-end tests, enabling a more faithful assessment. We evaluate leading LLMs on RealisticTritonBench and find that they still struggle with real-world Triton kernel generation tasks.

Augmentation can corrupt a training example when an image and its annotations receive different random changes. A crop must use the same coordinates for the image, mask, boxes, keypoints, stereo views, video frames, or volume. Code paths that choose these values separately can silently misalign the data. AlbumentationsX keeps the transform list, probabilities, annotation settings, and random seed in one Compose object. Each call chooses random values once and applies them to every supported part of the training example. The library keeps each object's mask, box, and label together and lets projects add their own transforms. It can also save the pipeline definition, show what happened in one call, and run that call again. The examples place Compose after files have been decoded into arrays and before PyTorch groups examples into a batch. AlbumentationsX executes the declared transforms. Practitioners still decide whether a flip, crop, color change, or other operation preserves the correct label for their task.

We systematically investigate finite-difference (FD) derivative computation in Physics-Informed Neural Networks (PINNs) as an alternative to automatic differentiation (AD). On three benchmark PDEs we show that, with a properly calibrated step size, FD matches AD in accuracy on every problem while running faster across the full tested batch-size range and using substantially less GPU memory, and that a stochastic variant we propose outperforms AD on a stationary problem. We further show that for neural architectures with inter-sample dependencies (e.g. BatchNorm, self-attention) the standard PyTorch autograd idiom is silently incorrect; the correct per-sample alternative is computationally infeasible at PINN-relevant batch sizes, while FD provides a forward-only approximation that is empirically an order of magnitude closer to the true per-sample derivative.

Activation checkpointing minimizes the runtime of neural networks under a given memory budget, by selecting which intermediate tensors to store and which to recompute. PyTorch solves this as a 0/1 knapsack problem, where operations from a joint forward-backward computation graph are items with a memory cost (weight) and a runtime saving (value). The default solver, dp_knapsack, allocates a full dynamic programming (DP) table of shape $(n+1) \times (W+1)$, where $n$ is the number of operations and $W$ is the quantized memory budget. This method is resource-hungry and crashes at $n = 100$ items on a machine with 64 GB RAM. In this paper, we introduce dp_knapsack_sliding_hirschberg, which combines the sliding window trick and Hirschberg's algorithm to reduce peak memory from $O(nW)$ to $O(W)$ while preserving the exact optimal solution. Our experiments show successful knapsack execution at $n = 2000$, where dp_knapsack fails at $n = 100$, a 20$\times$ increase in computable problem size. In addition, our benchmarks show a consistent 25-28% runtime speedup over dp_knapsack. The implementation is merged into PyTorch and released in version 2.10.

Ptychography neural networks suffer from scaling inconsistencies when generalizing out of distribution, limiting their real world viability. We address this scaling mismatch using a factorization strategy which decouples the learned object texture from measurement scaling, enabling a single trained network to produce measurement-consistent reconstructions across varying illumination conditions. This requires predicting the learned object in real and imaginary units instead of the canonical amplitude and phase representation. We additionally introduce a synthetic object sampling strategy that minimizes phase distribution mismatch between synthetic training data and experimental targets. These improvements yield up to a 5x reduction in Fourier error over the previous PtychoPINN-torch baseline across 5 experimental datasets spanning multiple beamlines and facilities.

Training and deployed inference often cross export, conversion, and platform-specific runtime boundaries. Meganeura asks whether one compact native compiler can span both phases on consumer GPUs. Its typed static graph, automatic differentiation, optimizer, checkpoint, memory planner, and runtime lower specialized programs through Vulkan and Metal. We compare five matched workloads with PyTorch on NVIDIA and AMD discrete GPUs, an AMD APU, Apple silicon, and an Intel iGPU. The protocol separates strict f32 from validated fast paths and gates forward and backward independently. Forty-eight of 50 device-workload-mode cells pass both gates; the other two share one unresolved backward-reference disagreement on a newly supported APU. In strict f32, Meganeura wins 12 of 20 GPU-referenced minimal-latency cells and has a median valid training gap of 1.8x. On the discrete AMD GPU, four of five inference workloads are within 1.10x of compiled ROCm PyTorch and three training workloads are faster. Under accelerated contracts, the worst training gap is 4.6x. Compilation takes 0.1-2.4 seconds versus 6-96 seconds for torch.compile on supported GPU paths; the stripped binary is 13 MiB. Dispatch profiles localize the largest gaps to convolution derivatives and attention backward. A physical Android XR case study transfers a Meganeura-trained decoder into an Adreno/OpenXR application sharing the graphics queue. The results show that general consumer graphics APIs can support a compact shared train-to-deploy stack at useful, sometimes vendor-competitive performance. The measured gaps point to kernel coverage, scheduling, and arithmetic policy rather than an identified API limitation.

Modern reranking recipes---billion-scale cross-encoders, mixture-of-experts (MoE) backbones, and distillation against strong teachers---have outpaced the training infrastructure available to most academic groups. Existing Tevatron reranker training relies on the Hugging Face Trainer with DeepSpeed or PyTorch FSDP1, but these backends lack efficient support for large-scale MoE training. We present Tevatron 3.0, which integrates a Megatron-Core training backend into Tevatron while preserving its data pipeline, evaluation workflow, and Hugging Face-compatible checkpoints. We benchmark existing distributed training configurations against the new backend, showing that Megatron matches FSDP reranker quality and training efficiency under comparable data-parallel settings, is up to 22% faster in the recommended single-node configuration, and supports both LoRA and full-parameter fine-tuning. Crucially, expert parallelism enables training a 30B-parameter Qwen3-30B-A3B MoE reranker, which is infeasible with PyTorch FSDP1. Using this framework, we conduct a controlled comparison of MoE versus dense models, LoRA versus full-parameter tuning, and distillation versus contrastive training on BEIR-15 with three first-stage retrievers, and report serving throughput for Hugging Face and vLLM. We find that the MoE reranker matches dense 8B quality while activating less than half as many parameters and achieving substantially higher inference throughput. We will release the framework and trained checkpoints.

Over the last decade, neural networks have been applied to an increasingly diverse range of applications, including data with rich geometric, topological, or symmetry-related structure. As a result, researchers have increasingly drawn inspiration from topology, algebra, and geometry. Despite this rich algorithmic development, the supporting software ecosystem remains fragmented. Many important methods exist only as research prototypes in unmaintained repositories. We address this by introducing Topology, Algebra, and Geometry Torch (TAGTorch), an open-source, PyTorch-based library that unifies tools inspired by topology, algebra, and geometry, including data-preprocessing methods, architectures, training techniques, and model analysis tools. We describe the design philosophy of TAGTorch and then discuss its current architecture and capabilities, highlighting areas where it can fill gaps in the current software ecosystem. We conclude with a discussion of our future development priorities for the library.

Agentic reinforcement learning research is constant algorithm modification, new estimators, new pipeline stages, new rollout schemes, and in mainstream frameworks each change threads through layers of trainer, distributed backend, and rollout glue: the cost lands on the researcher at every iteration. Molt is a PyTorch-native training framework built to keep that cost small: a codebase compact and clean enough for a researcher to hold in their head, and for an AI coding assistant to read and reason about in its entirety, so the algorithm flow can be traced and changed end to end. The agent is an ordinary program, and one asynchronous loop trains multimodal and mixture-of-experts policies while never training on a token it did not generate, consistent in tokens, policy versions, and model semantics. Leanness does not cost performance: under a matched, fully asynchronous protocol, Molt is statistically comparable to a state-of-the-art Megatron-based stack. Molt is open source and provides recipes and containers at https://github.com/NVIDIA-NeMo/labs-molt.

In RLHF pipelines, reward scoring blocks policy updates. Slow scoring bottlenecks the entire loop, since no update runs until every rollout gets a score. And yet most setups just default to PyTorch eager mode or torch.compile, no one checks if that's actually fastest. Scoring itself is small. Rollout generation eats far more of a typical RLHF step. But scoring and generation fight over the same CPU and GPU resources, so a faster scoring engine doesn't shrink step time on its own. It mainly frees up capacity generation can use instead. We built a native C++ inference engine on ONNX Runtime. First step: confirm correctness. Output matched the PyTorch reference to 5.7 x 10^-6 on CPU and 4.2 x 10^-3 on GPU, close enough to trust. Then we tested it against PyTorch eager mode, torch.compile, and FastAPI, on both CPU and GPU. CPU was decisive. Our engine beat every baseline, confidence intervals didn't even overlap. GPU gave a different view: we beat PyTorch and FastAPI, but torch.compile came out ahead. Further testing traced the speedup to ONNX Runtime itself, not C++ as a language. And batching strategy mattered more than either the language or the runtime choice, more than we expected. The results are from repeated, independent runs, since single runs just aren't reliable enough to trust.

As with most ``in the wild'' collections of the natural world, the North America Camera Trap Images (NACTI) dataset exhibits long-tailed class imbalance, with the largest class covering over 50% of its 3.7M images. Building on the PyTorch Wildlife model, we systematically evaluate Long-Tail Recognition (LTR) methodologies to benchmark species recognition performance, including specialised loss functions and LTR-sensitive regularisation. Our optimised configuration achieves state-of-the-art 99.40% Top-1 accuracy on the NACTI test split, significantly outperforming standard baselines and previously reported top performances. To assess robustness under domain shifts (e.g., night-time captures, occlusion, motion-blur), we extend our evaluation across three independent reduced-bias test sets (including ENA-Detection, Caltech Camera Traps and Missouri Camera Traps). Across these out-of-distribution (OOD) evaluations, our LTR-enhanced model consistently demonstrates substantially stronger generalisation capabilities compared to standard cross-entropy approaches. However, qualitative and quantitative analyses underline that current LTR optimisations cannot fully overcome representational bottlenecks, resulting in catastrophic predictive breakdown for rare `Tail' classes under severe domain shift. For maximum reproducibility, all dataset splits, key code, and network weights are published with this paper at https://github.com/ZehuaLiuY/Species-Classification.

Physics-Informed Neural Networks (PINNs) solve PDEs by incorporating physical constraints into neural-network training, but large-scale problems are limited by automatic-differentiation memory overhead and inefficient execution of grid-based PDE operators. We present FlashPDE, a drop-in fused operator library for grid-based scientific machine learning. FlashPDE replaces fragmented PyTorch finite-difference execution with differentiable Triton kernels. Each operator integrates fused stencil evaluation, an analytic discrete-adjoint backward pass, and boundary-gradient correction within a unified PyTorch autograd Function interface. The library provides 14 differentiable PDE operators covering 17 configurations across 1D--3D elliptic, parabolic, and Navier--Stokes systems, while remaining independent of neural architectures and training strategies. Experiments on an NVIDIA A100 GPU show that FlashPDE reduces peak memory usage by up to 37.0x compared with coordinate-based automatic differentiation and reduces CUDA kernel launches by up to 3.5x compared with eager PyTorch finite-difference implementations. Across six representative PDE benchmarks, FlashPDE achieves up to 2.30x end-to-end time-to-solution speedup and up to 19.2x kernel-level acceleration while maintaining numerical agreement with PyTorch finite-difference references. FlashPDE provides a hardware-efficient execution layer that bridges differentiable PDE solvers and GPU-optimized numerical computation within the PyTorch ecosystem.

OpenLanguageModel (OLM) is an open-source PyTorch library for building and pretraining small language models while keeping their machinery visible. In OLM, model code reads like the architecture: components are ordinary modules, while Block, Residual, Repeat, and Parallel describe how they are wired. The resulting model can move unchanged from a teaching notebook to a complete pretraining run or a research ablation. OLM connects this readable model layer to tokenizers, local and streaming datasets, optimization, mixed precision, callbacks, checkpoints, and hardware-aware CPU, single-GPU, and single-node multi-GPU execution. We demonstrate the full path by tracing GPT-2 from diagram to code, launching a FineWeb-Edu training script, replacing one attention component, and letting AutoTrainer configure the available machine. The package includes 27 presets across nine familiar model families and documentation that progresses from LM fundamentals to architecture research. Validation shows close agreement with independent reference implementations, 90.6% four-GPU weak-scaling efficiency for a 348M-parameter workload, compact architecture edits, and positive early usability results. OLM is MIT-licensed and available through PyPI, GitHub, and its documentation site.

Existing GPU kernel generation benchmarks draw problems from synthetic or curated sources that diverge from deployed workloads. We present Atrex-Bench, a benchmark whose 30 operators and 440 shapes are sampled directly from full-cluster production inference traces of compute-limited, memory-rich GPUs. Each problem carries an importance weight derived from its share of observed GPU time, weighted by application card-hours and computed separately for the serving phases in which it runs, together with a per-problem roofline ceiling, so the aggregate score emphasizes the kernels that consume the most serving time. Evaluating six frontier coding agents on Atrex-Bench shows that even the best vanilla model reaches only ${\sim}10\%$ of the hardware roofline on production operators; and correctness alone overstates capability, since much of the apparent pass rate comes from PyTorch fallbacks rather than kernels the model wrote. To close this gap, we co-release Atrex-Kernel-Agent (AKA), a profile-driven kernel-optimization agent that combines iterative measure-revise search, optimization dropout for escaping stalled search contexts, and a layered GPU-optimization knowledge base (298 reference-kernel files and 244 optimization-knowledge documents, plus external upstream reference projects for API/ISA lookup). In a controlled case study, the agent converts zero-FlyDSL fallbacks into real kernels that match or exceed hand-tuned production baselines.

Neural decompilation is increasingly studied as a code-generation problem, yet its evaluation methodology remains underdeveloped for modern languages. We present a systematic empirical study of fine-tuning effectiveness and metric validity for Dart Ahead-of-Time (AOT) neural decompilation. We evaluate six fine-tuned model variants across three base architectures (4B-8B parameters) using three metrics: CodeBLEU, compile@k, and pass@k on a new 154-task HumanEval-Dart benchmark. Our study yields three principal findings grounded in paired task-level statistical tests. First, no fine-tuning configuration produces a statistically significant pass@k improvement. The sole positive case yields +0.71 pp (McNemar p=0.21), while fine-tuning the strongest base (Qwen3-8B) causes a highly significant regression of -5.65 pp (p<0.001). This capacity-dependent trend is consistent across architectures but needs broader scale sweeps. Second, cross-lingual interference from Swift training is highly significant at 4B (-2.66 pp, p<0.001) but statistically indistinguishable from zero at 8B, consistent with the scaling hypothesis. Third, we demonstrate metric divergence: CodeBLEU and compile@k can improve significantly while pass@k moves in the opposite direction. This has implications for any LLM code generation task where fine-tuning targets superficial similarity. Error analysis reveals assembly sequence length is the strongest predictor of task difficulty (p=0.001), with a capability cliff at 200 instructions. We contribute the HumanEval-Dart benchmark, a Dart-adapted CodeBLEU, and empirical evidence that pass@k must be the primary evaluation metric for neural decompilation.

In this paper, we present Kiwano, an open-source toolkit designed to advance research and evaluation for speaker verification. Kiwano provides a lightweight yet extensible framework built on PyTorch, offering standardized recipes, pretrained models, and integration of several widely used speaker verification architectures. The toolkit emphasizes reproducibility, by delivering transparent training pipelines, unified evaluation protocols and ready-to-use baselines across multiple corpora. Beyond conventional training and inference, Kiwano includes tools for benchmarking, experiment tracking and rapid prototyping of new architectures. To foster community adoption, the toolkit is distributed under the Apache 2.0 license, accompanied by comprehensive documentation and reproducible experiments. By lowering entry barriers and standardizing evaluation practices, Kiwano contributes a valuable resource for both academic research and applied development in speaker verification. The toolkit is publicly available at: https://github.com/kiwano-toolkit/kiwano/

Neurosymbolic semantics is fragmented: classical, fuzzy, probabilistic and neural systems each define truth by their own inductive rules. NeSyCat, extending ULLER, subsumes them under a single inductive definition of truth, parametric in a strong monad and an aggregation structure on truth-values. NeSyCat has so far lacked an account of predicates and functions learned by neural networks. We provide NeSyCat Torch as the missing link and interpret computational symbols via neural networks, implementing the framework in probabilistic programming and tensor-based backends. We use the distribution monad for reference semantics and metric evaluation, and complement it by a monad for numerically stable, differentiable training: the lazy log-tensor monad over the log-semiring. For efficient training in batches, we furthermore employ a batch monad. The axioms are the source code: written once in monad-based do-notation, monadic bind performs marginalisation, lazily pruning unneeded branches. On MNIST addition, our HaskTorch, JAX, and PyTorch implementations outperform LTN and DeepProbLog in speed and accuracy, while achieving nearly the accuracy of DeepStochLog. However, unlike DeepStochLog, we stay in a uniform framework that applies to many first-order NeSy approaches. Namely, the construction is parametric in the monad; instantiating it with, e.g., the Giry monad extends the approach to continuous probability (working out a neural representation here is left for future work).

Kolmogorov-Arnold Networks (KANs) have recently emerged as a promising alternative to traditional multilayer perceptrons by replacing linear weights with learnable univariate functions. Despite their theoretical advantages in interpretability and expressiveness, practical research of KANs remains difficult due to high computational costs and inconsistent feature support across existing frameworks. This paper introduces KANLib, a modular, extensible, and computationally efficient framework for developing and evaluating KAN architectures. KANLib unifies core concepts from existing implementations, including PyKAN, EfficientKAN, and FastKAN, within a consistent software architecture that emphasizes flexibility, feature parity, and high performance. The framework supports two basis function types, adaptive grid rescaling, grid extension, and fine-grained architectural customization while maintaining compatibility with standard PyTorch workflows. Experimental evaluation on the California Housing benchmark demonstrates that KANLib reproduces the predictive behavior of established reference KAN implementations while achieving competitive computational efficiency. Furthermore, the framework enables the exploration of architectural variations beyond standard KAN formulations with only minor impacts on predictive performance. Overall, KANLib provides a robust foundation for future research on scalable and extensible KAN architectures.

Translating deep learning models from PyTorch's flexible, object-oriented design to JAX's functional, stateless setup is usually a manual and error-prone task. Automated migration is challenging because Large Language Models (LLMs) struggle with strict and dynamic API alignment and are prone to mistakes for exacting operations. We propose a fully autonomous system that combines In-Context Learning (ICL) with oracle-driven self-debugging. First, we curated an ICL context that serves as a strict reference for idiomatic JAX styling and test case generation. Second, instead of depending on the LLM to deduce mathematical outputs, we run the source PyTorch modules to get their actual dynamic tensor states. This creates an unchangeable execution oracle. We then use an autonomous agentic loop to synthesize tests based on the oracle data. The test cases are executed repeatedly, and the traceback is sent back to the LLM for self-correction. Ablations show that combining ICL references with oracle grounding and self-debugging greatly outperforms pure instructional and basic agentic baselines. This improvement does not add an excessive computational overhead. Our lightweight pipeline achieves 91% numerical equivalence (compared to baseline: 9%, instruction + self-debugging: 27%) on neural modules, providing a highly reliable, scalable blueprint for cross-framework migration. This has been validated across several state-of-the-art models including SAM (segment anything), T5, Code Whisper amongst others showing high numerical equivalency. Code: https://github.com/AI-Hypercomputer/accelerator-agents/tree/main/MaxCode

This paper traces, with explicit numerical values, how PyTorch's automatic differentiation (AD) engine computes gradients for Physics-Informed Neural Network (PINN) training -- a setting that requires two levels of differentiation: computing the physics derivative $\hat{y}'(t)=d\hat{y}/dt$ through the network, and computing parameter gradients $\nabla_θL$ of a loss that itself depends on $\hat{y}'(t)$. Using a 1-3-3-1 multilayer perceptron and the initial value problem $y'(t)+y(t)=0$, $y(0)=1$, we trace the complete pipeline at every node: the computational graph built during the forward pass, the reverse-mode backward traversal that computes all 22 parameter gradients in a single pass, and the graph-on-graph mechanism by which \texttt{create_graph=True} enables correct differentiation through the physics-informed residual. Every adjoint value is verified against the hand derivations of Tahimi (2026), connecting the $P/Q$ sensitivity framework to the vector--Jacobian products used by PyTorch's autograd engine.

ANEForge is a Python package that programs the Apple Neural Engine (ANE), the fixed-function neural accelerator on every recent Apple device, directly and without CoreML. In production the engine is reachable only through CoreML, which treats it as a scheduling option: no configuration requires the ANE, and a model can silently run on the CPU or GPU instead. ANEForge compiles a lazy tensor graph, built from 58 fused operators and 19 native bridge operators, into a single ANE program. The program is dispatched through the same ANE daemon and kernel-driver stack as Apple's internal framework. Beyond inference, the package reaches the engine's native fused attention, streams int8, int4, and sparse weights, keeps decoder and optimizer state resident across steps, and runs the forward pass, backward pass, and optimizer update of training on the engine. A small fused program completes a call in about 90us, near the engine's 70us per-program dispatch floor, and a pretrained ResNet-18 forward runs end-to-end in 0.33ms. ResNet-18, a sentence encoder, and a Vision Transformer run end-to-end against framework references, and a Stable Diffusion U-Net validates its forward pass. ANEForge targets Apple Silicon under macOS 14 and later. Each release is verified against a recorded macOS and ANE-compiler version.

Gaussian splatting methods have become increasingly popular for neural reconstruction of the real world. However, they are often limited in scale and resolution due to compute and memory constraints. We present a multi-GPU Gaussian splatting approach that scales reconstruction to higher resolutions and larger scenes while abstracting away the code complexity typically associated with distributing a model. To accomplish this, we propose a PyTorch backend that distributes the Gaussian parameters and splatting operators across GPUs via CUDA unified memory and NVLink. Because distribution occurs at the operator level, the model code requires no explicit cross-device communication. More broadly, the backend exposes multiple GPUs as an aggregate PyTorch device and supports other PyTorch operators. We demonstrate city-scale reconstructions with street-level detail consisting of over 1 billion Gaussian splats, more than 25 times as many as the current state of the art.

As deep learning models scale, managing, inspecting, and modifying large checkpoints has become increasingly challenging. Researchers often need to alter model weights for layer restructuring, precision casting, low-rank factorization, and architectural debugging, yet these workflows often rely on fragile ad-hoc Python scripts. Here, we introduce BrainSurgery, a tool for robust and reproducible "tensor surgery" on neural network checkpoints, and provide a system demonstration covering four examples and three case studies from model upcycling to LoRA extraction. By abstracting storage formats and memory management, BrainSurgery executes complex transformations through declarative YAML plans. It supports structural modifications, mathematical transformations, and tensor reshaping through expressive regex and structural targeting, while built-in assertions validate tensor shapes, data types, and values to prevent silent errors. We envision that BrainSurgery will provide a strong foundation for future research through its reproducible and validated operations.

Tensor program optimization is essential for modern machine learning systems, but its search space is enormous. Existing auto-schedulers reduce measurement cost with learned cost models, yet they usually evaluate each candidate as a static code snapshot, ignoring the schedule trajectory that produced it. This makes them insensitive to action dependencies and vulnerable to superficial code variations. We propose a \emph{world-model-inspired} evaluator that models schedule evaluation as action-conditioned latent dynamics over program states. Starting from the initial program, it rolls out scheduling actions in a continuous latent space with a lightweight transition model, avoiding expensive AST mutation and repeated code encoding. The final dynamic representation is combined with action and hardware features to rank candidates. Implemented in TVM AutoScheduler, our method improves representative-subgraph latency over Ansor by 1.37$\times$ on GPU and 1.54$\times$ on CPU under the same 64-trial budget. It also matches Ansor-10K within 2.2% geometric mean using 10$\times$ fewer measurements, and accelerates full-model inference over PyTorch/PyTorch-opt(cuDNN) by 4.61$\times$/3.67$\times$ geometric mean.

TorchKM is an open-source library for kernel machines, including support vector machines, kernel logistic regression, and kernel quantile regression, with GPU acceleration. The library features a scikit-learn-style API and is designed to exploit GPU-friendly linear algebra, accelerating the full training and model-selection pipeline through intelligent reuse of matrix operations. Benchmarks show competitive predictive performance with substantial speedups over standard baselines. The efficiency and programmable design also make TorchKM a kernel-learning component for AI-driven workflows. Code and documentation are available at https://github.com/YikaiZhang95/torchkm, and the package can be easily installed via PyPI.

Transformer inference increasingly depends on specialized compiler and runtime support, but real model graphs still require semantic decisions about which regions are worth specializing and which CUDA implementation families are plausible. We present AgentCompile, an LLM-guided CUDA inference compiler that uses LLM outputs only as advisory search metadata. Given compiler-derived region summaries and bounded candidate spaces, the LLM proposes semantic labels, candidate priorities, parameter hints, and risk annotations; the compiler materializes CUDA candidates through templates, checks interface and hardware constraints, validates candidates empirically, selects implementations by measured latency, and falls back when specialization is unsupported or unprofitable. In end-to-end autoregressive generation, AgentCompile averages 5.66x, 4.05x, and 4.26x speedup over PyTorch eager on Qwen3-1.7B, Qwen3-4B, and Llama-3.2-1B-Instruct, respectively, across five representative workloads. We will open-source the project.

nnAudio is an open-source audio feature extraction toolbox for deep learning, but its use in current environments is hindered by TorchScript incompatibilities, inverse-transform edge cases, and dependency drift. We present a targeted modernization for modern PyTorch and scientific Python. We resolve TorchScript compilation failures in STFT and iSTFT by removing dynamic state mutation and module construction from scripted code paths and tightening argument handling in inverse-related helpers. We clarify inverse-STFT behavior by restricting reliable inversion to the uniform-bin setting (freq_scale=`no') and raising explicit runtime errors for unsupported frequency scales, preventing silently degraded reconstructions. We restore CFP compatibility with modern SciPy and ensure VQT reduces to CQT when gamma = 0. Regression tests cover the new STFT/iSTFT behaviors, and the updated codebase passes the full repository test suite in a modern Python environment. These improvements provide a more robust foundation for differentiable audio analysis in research and deployment.

As robot control shifts toward large-scale reinforcement learning with in-loop dynamics computation, the community's reliance on CPU-bound libraries such as Pinocchio creates a throughput bottleneck in GPU-based training pipelines. We present BARD (Batched Articulated Rigid-body Dynamics), a self-contained PyTorch implementation of Featherstone's rigid-body dynamics algorithms, optimized for batched GPU evaluation and automatic differentiation. Three design choices make this efficient: a tiered lazy-evaluation cache that avoids redundant tree traversals, matmul-free joint transforms via pre-computed Rodrigues constants, and level-parallel propagation that reduces sequential operations to tree-depth batched steps. On five robot models (7-23 DOFs), BARD matches Pinocchio numerically while reaching up to 64x higher throughput for Forward Kinematics and 63x for Jacobians at batch size 4096 on an NVIDIA H200. We validate differentiability through gradient-based system identification on a 7-DOF manipulator, recovering link masses to 1.24% mean error under 5% torque noise, and integrate BARD into an Isaac Lab AMP training pipeline for an 11-DOF spined quadruped with 4096 parallel environments, where it is 8.5x faster than Pinocchio and 2.0x faster than ADAM for in-loop dynamics. BARD is open-sourced at: https://github.com/YueWang996/bard-pytorch-dynamics.

We introduce Fuzzy PyTorch, a framework for rapid evaluation of numerical variability in deep learning (DL) models. As DL is increasingly applied to diverse tasks, understanding variability from floating-point arithmetic is essential to ensure robust and reliable performance. Tools assessing such variability must be scalable, efficient, and integrate seamlessly with existing frameworks while minimizing code modifications. Fuzzy PyTorch enables this by integrating stochastic arithmetic into PyTorch through Probabilistic Rounding with Instruction Set Management, a novel library interfacing with Verificarlo, a numerical analysis compiler. The library offers stochastic rounding mode and a novel mode; up-down rounding. Comparative evaluations show Fuzzy PyTorch maintains model performance and achieves runtime reductions of 5x to 60x versus Verrou, a state-of-the-art tool. We further demonstrate scalability by running models from 1 to 341 million parameters, confirming applicability across small and large DL architectures. Overall, Fuzzy PyTorch provides an efficient, scalable, and practical solution for assessing numerical variability in deep learning, enabling researchers and practitioners to quantify and manage floating-point uncertainty without compromising performance or computational efficiency.

In a landscape of high-performance distributed ML systems, JAX has emerged as a framework of choice. However, JAX's modular design philosophy leaves it without a standardized checkpointing solution. In this paper, we introduce Orbax, a modular, JAX-native checkpointing library that abstracts the complexities of distributed accelerator systems while also providing flexibility for user-friendly checkpoint manipulations throughout the ML model lifecycle. We demonstrate performance exceeding comparable PyTorch competitors by up to 3.5$\times$ for saving and 2$\times$ for loading. The library is available at https://github.com/google/orbax.

Scientific machine learning often requires combining known physics with unknown parameters or correction terms learned from data. Existing approaches either ignore known structure, encode it as a soft penalty, or require hand-written PyTorch code for each equation. We present The Neural Compiler, a system that translates programs written in a first-order Scheme-like expression language into frozen, differentiable PyTorch modules. These modules match the source program to floating-point precision and provide gradients through autograd. In hybrid models, the compiled module encodes known physics exactly while learned components model the unknown remainder. We evaluate the compiler across six experiment domains: Feynman physics equations, Lotka-Volterra dynamics, a damped pendulum, a one-dimensional heat equation, three-dimensional vector mechanics, and compositional generalization. Compiled modules match hand-coded PyTorch implementations numerically for single equations, showing no accuracy loss from compilation. With only 1 to 4 trainable parameters, compiled models recover physical constants to less than 1 percent error in most cases, while standard PINN baselines with more than 8500 parameters show 7 to 93 percent error. Compiled modules also compose with zero error, while neural approximations can accumulate large errors in deep composition chains. The main value of the compiler is not improved accuracy over hand-coded equations, but systematic composability: it generates correct, differentiable modules from symbolic specifications without rewriting each equation by hand. The system supports 51 primitive operations, including vector and matrix algebra, enabling PDE discretizations and hybrid scientific models. This string-in, module-out interface also provides a natural target for large language models that translate scientific descriptions into executable differentiable modules.

We show that singular value decomposition of the lm_head} weight matrix of a transformer-based large language model -- requiring only five lines of PyTorch and no model inference -- reveals interpretable semantic subspaces directly from the model weights. Each left singular vector identifies the vocabulary tokens most readily selected when the hidden state aligns with the corresponding singular direction; inspecting these clusters exposes the model's training data composition and curation philosophy. Analysing GPT-OSS-120B, Gemma-2-2B, and Qwen2.5-1.5B, we find that singular value spectra and vocabulary cluster structures differ systematically across models: GPT exhibits a graduated hierarchy of functionally differentiated subspaces; Gemma is dominated by pre-nineteenth-century English orthography, forming a stepwise clustering structure that may contribute to high output controllability; and Qwen exhibits broad multilingual coverage alongside subspaces whose vocabulary the authors have determined to be ethically inappropriate for direct publication. Base-instruct comparison reveals that ethically concerning subspaces originate in pretraining and are not removed by post-training alignment. We introduce the Vocabulary Cluster Score (VCS) to quantify subspace coherence, and the Weighted Projection Score (WPS) as a static glitch token detector; applying WPS to GPT-OSS-120B recovers shokubutsu-hyakka-tsu (ID 137606), a well-known glitch token widely reported in the CJK language community, without any model inference. We propose a taxonomy of root causes for problematic vocabulary content and call for lm_head} SVD analysis to be adopted as a standard pre-release safety auditing step. Our findings further suggest directions toward SVD-guided tokenizer optimisation and more controllable LLM design.

Modern LLMs typically require multistage training pipelines to achieve strong downstream performance, with post-training serving as the main interface for adapting open-weight models. We introduce torchtune, a PyTorch-native library designed to streamline the post-training lifecycle of LLMs, enabling efficient fine-tuning, experimentation, and deployment-oriented workflows. Unlike many existing fine-tuning frameworks, which often optimize for ease of use, specialized recipes, or hardware efficiency at the cost of transparency and extensibility, torchtune emphasizes modularity, hackability, and direct access to the underlying PyTorch components. In this paper, we present the design principles behind torchtune, describe how they are reflected in its model builders, training recipes, and distributed training stack, and evaluate the library across representative post-training settings. We compare against popular fine-tuning frameworks, including Axolotl and Unsloth, and show that torchtune provides strong performance and memory efficiency across many settings while remaining flexible enough for rapid research iteration. These results position torchtune as a practical foundation for reproducible LLMs post-training research.

Sutra is a typed, purely functional programming language whose compiled forward pass is a PyTorch neural network. The compiler beta-reduces the whole program -- primitives, control flow, string I/O -- to one fused tensor-op graph over a frozen embedding substrate. Rotation binding, unbind, bundle, polynomial Kleene three-valued logic, and tail-recursive loops all lower to tensor operations; the Kleene connectives are Lagrange-interpolated polynomials exact on the {-1, 0, +1} truth grid. Validation is one fact tested two ways. (1) The same program runs on four frozen embeddings spanning two modalities -- three text encoders (nomic-embed-text, all-minilm, mxbai-embed-large) and one protein language model (ESM-2) -- and decodes bundles at 100% accuracy through width k=8 on every substrate, where the textbook Hadamard product has already collapsed (2.5% on mxbai-embed-large, 7.5% on all-minilm). (2) PyTorch autograd flows through the actually compiled graph: a fuzzy-rule classifier written in .su trains from random init (18.7 +/- 9.5%; chance = 20%, five classes) to 100.0 +/- 0.0% (three seeds) by backpropagating through the emitted graph, the symbolic source unmodified. A weighted variant additionally trains a scalar cosine gain and writes it back into the .su source as a numeric literal; recompiling reproduces the trained behaviour to ~2e-7 per logit, so the trained model is itself legible, recompilable code. The same artifact is therefore both a logic program and a trainable neural network.

This work presents an end-to-end strategy for solving inverse problems constrained by Partial Differential Equations within a fully differentiable Machine Learning framework. The proposed formulation provides a unified and user-friendly methodology applicable to a wide range of problems, from data assimilation to closure modeling. Our approach combines a baseline differentiable PDE solver, which predicts the state w from the nonlinear system $R(w) = 0$, with a generic additive, parametrized, and differentiable correction $f_φ(w)$, with trainable parameters $φ$. We show how to optimize phi within a fully differentiable Python workflow by reformulating the PDE as an implicit layer, enabling its integration into arbitrary objective functions, while leveraging PyTorch's automatic differentiation graph. The method is demonstrated on the Reynolds-Averaged Navier-Stokes equations for compressible flows, where the closure term, or a portion of it, is modeled using trainable parameters or a Neural Network. The first application considers the 2D NASA Wall-Mounted Hump test case, where a production-term parameter is optimized against time-averaged LES data. A second application is carried out on the VKI LS-59 turbine blade, where the Spalart-Allmaras eddy viscosity field is reconstructed through the optimization of a trainable spatial field. A dataset is generated starting from the VKI LS-59 turbine blade geometry using the differentiable BROADCAST solver with the Spalart-Allmaras turbulence model. The results highlight the flexibility of the framework, showing its applicability beyond turbulence modeling to a broader class of physics-informed PDE-constrained problems with data-driven components.

Gaussian process (GP) inference in Python is dominated by libraries such as GPyTorch and GPflow, which are built on deep-learning frameworks and inherit their dispatch overhead and dependency footprint. We present LightGP, a dependency-free C++17 library for GP regression with Python bindings, supporting Apple Metal and NVIDIA CUDA backends alongside tuned CPU paths via Apple Accelerate and OpenBLAS. LightGP provides four inference paths -- exact Cholesky, matrix-free conjugate gradients, sparse variational free energy, and structured kernel interpolation with FFT -- covering problems from $N{=}100$ to $N{=}500{,}000$. On an Apple M4, LightGP CPU is 2.6--8.7$\times$ faster than GPyTorch CPU for exact GP and ${\sim}1.5\times$ faster for sparse GP at every scale tested. On an NVIDIA RTX~3060, LightGP CUDA is 2.3--6.7$\times$ faster than GPyTorch CUDA for exact GP up to $N{=}2{,}048$, with GPyTorch closing the gap at $N{=}4{,}096$. A fused matrix-free kernel-vector product on Metal achieves 32$\times$ over the explicit path at $N{=}20{,}000$ with $O(N)$ memory, and an FFT-accelerated SKI matvec via Accelerate vDSP runs in sub-millisecond time at $N{=}200{,}000$. LightGP compiles as a single static library with zero external dependencies and is installable via \texttt{pip install lightgp

Modern generative modeling has grown into a broad collection of related but often separately implemented paradigms, including denoising diffusion models, score-based stochastic differential equations, flow matching, variational autoencoders, normalizing flows, adversarial models, and energy-based models. For newcomers, this fragmentation makes it difficult to compare training objectives, inference procedures, sampling algorithms, and conditioning mechanisms within a single coherent codebase. We introduce V ENOM, an educational PyTorch toolkit that implements representative generative modeling families under a unified, MNIST-first interface. V ENOM emphasizes breadth, readability, reproducible entry points, and consistent training and sampling APIs rather than large-scale performance engineering. The package currently includes diffusion and score-based models, flow matching and one-step generators, variational autoencoders, normalizing flows, generative adversarial networks, and energy-based models. It provides separate training and sampling scripts, classifier and classifier-free guidance examples, bilingual tutorial notebooks, and a model-family organization that supports teaching, prototyping, and lightweight benchmarking.

This paper presents MiniGPT, a compact from-scratch implementation of GPT-style autoregressive language modeling in PyTorch. The aim is to rebuild the core GPT pipeline from first principles after studying the design of nanoGPT by Andrej Karpathy, while keeping the model and training code independently written in a single notebook. MiniGPT implements token and positional embeddings, causal multi-head self-attention, pre-LayerNorm Transformer blocks, residual connections, feed-forward MLP layers, next-token cross-entropy training (teacher forcing), validation tracking, checkpoint selection, and autoregressive text generation. This paper evaluates the implementation on Tiny Shakespeare dataset using character-level tokenization. A baseline 0.83M-parameter model reaches a validation loss of 1.7236 after 3000 training iterations. A stronger 10.77M-parameter configuration, using a larger context length and improved training settings, reaches a best validation loss of 1.4780 and generates text with recognizable Shakespeare-style dialogue structure. MiniGPT does not introduce a new language-model architecture. Instead, it documents a clear and reproducible implementation path from raw text to trained character-level generation, including design choices, training behavior, generation quality, and practical limitations.

GPU kernel optimization is increasingly critical for efficient deep learning systems, but writing high-performance kernels still requires substantial low-level expertise. Recent AI coding agents can iteratively read code, invoke compilers and profilers, and refine implementations, yet existing kernel benchmarks evaluate single LLM calls rather than full agent workflows, and none include both kernel-to-kernel optimization and unseen-configuration generalization testing. We present AgentKernelArena, an open-source benchmark for measuring AI coding agents on GPU kernel optimization. The benchmark contains 196 tasks spanning HIP-to-HIP optimization, Triton-to-Triton optimization, and PyTorch-to-HIP translation, and evaluates complete agent workflows in isolated workspaces using gated compilation, correctness, and performance checks, centralized scoring and an unseen-configuration generalization protocol that tests whether optimizations transfer to input configurations the agent never observed. Across production agents including Cursor Agent, Claude Code, and Codex Agent, we find near-perfect compilation and high correctness rates on most task categories, with the strongest configurations achieving mean speedups of up to 6.89x on PyTorch-to-HIP, 6.69x on HIP-to-HIP, and 2.13x on Triton-to-Triton tasks. Our unseen-configuration evaluation shows that HIP-to-HIP and Triton-to-Triton optimizations largely transfer to unseen input shapes, while PyTorch-to-HIP exhibits substantial correctness drops, indicating that agents generating kernels from scratch frequently hardcode shape-specific assumptions. AgentKernelArena is designed as a modular, extensible framework for rigorous evaluation of agentic GPU kernel optimization across agents, tasks, and hardware targets.

DeepLog is an operational neurosymbolic framework that unifies logic and deep learning within standard PyTorch workflows. While existing neurosymbolic systems focus on a particular paradigm and semantics, DeepLog serves as a universal backend that can emulate many systems in the neurosymbolic alphabet soup. By treating diverse neurosymbolic languages as high-level specifications, the DeepLog software automatically compiles them into optimized arithmetic circuits. This design lowers the barrier for machine learning practitioners by treating logic as composable modules, while providing neurosymbolic developers with a shared, high-performance basis for prototyping new integration strategies. The code is available here: https://github.com/ML-KULeuven/deeplog

jNO (jax Neural Operators) is a JAX-native library for neural operators and foundation models with unified support for both data-driven and physics-informed training. Its core design is a tracing system in which domains, model calls, residuals, supervised losses, and diagnostics are written in one symbolic language and compiled into one optimization pipeline. This allows users to move between operator regression, mesh-aware residual evaluation, and PDE-constrained training without restructuring the surrounding code. jNO also supports multi-model compositions, fine-grained control at parameter level (model, optimizer, and learning rate), hyperparameter tuning, and JAX-native workflows for translated PDE foundation-model families. The source repository is available at https://github.com/FhG-IISB/jNO.

JPEG decode is routine ML infrastructure, but Python decoder choices are often justified by single-process, single-thread microbenchmarks. We audit this evaluation assumption with thirteen Python-accessible JPEG decode paths on five matched 16 vCPU Google Cloud CPUs: Intel Emerald Rapids, AMD Zen 4, AMD Zen 5, ARM Neoverse V2, and ARM Neoverse N1. ImageNet validation is the workload, not a new dataset contribution: each run decodes the full 50,000-image split from memory and reports single-thread throughput for all decoders, PyTorch \texttt{DataLoader} throughput for eligible decoders at worker counts $\{0,2,4,8\}$, and decoder skip behavior. The evaluation protocol changes the supported conclusion. On Neoverse V2, \texttt{imageio} is ninth in single-thread throughput yet lands in the top DataLoader tier with \texttt{torchvision}; on Zen 4, \texttt{torchvision} rises from seventh single-thread to the top measured DataLoader tier; on Neoverse N1, \texttt{imagecodecs} is the single-thread leader but fifth at peak DataLoader throughput. We also find that worker-count conclusions differ between Zen 4 and Zen 5, TensorFlow has a large single-thread ARM penalty, and strict native JPEG decoders/wrappers reject the same rare ImageNet JPEG. For PyTorch DataLoader workloads, \texttt{torchvision} and \texttt{simplejpeg} form the strongest measured zero-skip tier: \texttt{torchvision} has the highest mean normalized throughput, while \texttt{simplejpeg} has the highest minimum. OpenCV remains a robust general-purpose fallback above 90% of the platform-local winner on every tested CPU. We release raw JSON, generated tables/figures, and an executable local/cloud benchmark framework.

Over the last two decades, language modeling has experienced a shift from the use of predominantly recurrent architectures that process tokens sequentially during training and inference to non-recurrent models that process sequence elements in parallel during training, which results in greater training efficiency and stability at the expense of lower inference throughput. Here we introduce the Structured Recurrent Mixer, an architecture that allows for algebraic conversion between a sequence parallel representation at train time and a recurrent representation at inference, notably without the need for specialized kernels or device-specific memory management. We show experimentally that this dual representation allows for greater training efficiency, higher input information capacity, and larger inference throughput and concurrency when compared to other linear complexity models. We postulate that recurrent models are poorly suited to extended sequence length scaling for information-rich inputs typical of language, but are well suited to scaling in the sample (batch) dimension due to their constant memory per sample. We provide Mojo/MAX inference implementations of SRMs exhibiting 12x the throughput and 170x the concurrency of similarly powerful Transformers inferenced on vLLM, increases characteristic of Pytorch implementations resulting in a 30% increase in compute-constant GSM8k Pass@k. We conclude by demonstrating that SRMs are effective reinforcement learning training candidates.

Local execution of AI on edge devices is important for low latency and offline operation. However, deploying models on diverse hardware remains fragmented, often requiring model conversion or complete reimplementation outside the PyTorch ecosystem where the model was originally authored. We introduce ExecuTorch, a unified PyTorch-native deployment framework for edge AI. ExecuTorch enables seamless deployment of machine learning models across heterogeneous compute environments. It scales from embedded microcontrollers to complex system-on-chips (SoCs) with dedicated accelerators, powering devices ranging from wearables and smartphones to large compute clusters. ExecuTorch preserves PyTorch semantics while allowing customization, support for optimizations like quantization, and pluggable execution "backends". These features together enable fast experimentation, allowing researchers to validate deployment behavior entirely within PyTorch, bridging the gap between research and production.

SwiGLU is the dominant MLP activation in modern large language models, yet its intermediate tensor materialization costs 9-37% of MLP execution time. We present two complementary CUTLASS-based SM90 kernels that fuse SwiGLU into GeMM at the tile level. Kernel-1 overlaps Swish computation on the Gate accumulator with Up-tile loading using the Pingpong warp-specialized schedule; Kernel-2 interleaves SwiGLU with tile stores via a custom Epilogue Visitor Tree. Evaluated on Qwen-2.5 models (0.5B-72B) on NVIDIA H100, our kernels achieve up to 2.47x speedup over PyTorch, shifting workloads from memory-bound to compute-bound and reaching 79.5% peak BF16 utilization. We demonstrate that torch.compile cannot replicate this fusion (3-7x slower than our kernels), validating the need for hand-crafted tile-level design. Our fused kernels are also numerically superior, achieving zero mismatches compared to 4.5-11% for cuBLAS.

Deep neural networks dominate modern machine learning, while alternative function approximators remain comparatively underexplored at scale. In this work, we revisit kernel methods as drop-in components for standard deep learning pipelines. We introduce \emph{Sparse Kernels} (SKs), a differentiable, localized, and lazy variant of kernel ridge regression (KRR) that defers training to inference time and reduces to the solution of small local systems. We integrate SKs into PyTorch as modular layers that preserve end-to-end trainability, and we show that they expose three distinct sets of parameters -- feature representations, target values, and evaluation points -- each of which can be fixed or learned. This decomposition broadens the design space available to practitioners, enabling, in particular, training-free transfer, nonlinear probing, and hybrid kernel-neural models. Across convolutional networks, vision transformers, and reinforcement learning, SK-based modules serve two complementary roles: in some settings, they match the performance of trained neural readouts with substantially less training; in others, they augment existing models and improve their performance when used as additional components. Our results suggest that kernel methods, once made scalable and differentiable, can be readily integrated with deep learning rather than treated as a separate paradigm.

We present a PyTorch package that compiles neural networks and their weights from Turing machine descriptions, producing models that exactly simulate the specified machine without any training. Given a transition function and a set of terminal states, the package constructs a model whose forward pass corresponds to one step of the Turing machine. Two architectures are implemented, each realizing a different theoretical result: (1) a transformer with self-attention, cross-attention, and feedforward layers based on Wei, Chen, and Ma (2021), and (2) a recurrent network based on Siegelmann and Sontag (1995) that encodes the stack in a Cantor set. We develop the constructions from first principles, showing how ReLU networks implement Boolean circuits (AND, OR, NOT, XOR gates and their composition into DNF formulas and binary adders) and how hard attention implements positional lookup on the tape. The package serves as a concrete, runnable reference for the symbolic-neural bridge, and as a foundation for future work on the stability of constructed solutions under gradient-based optimization. Code is available at https://github.com/jonrbates/turing.

We present VkSplat, a high-performance, cross-vendor 3D Gaussian Splatting (3DGS) training pipeline implemented fully in Vulkan compute, addressing performance and compatibility limitation of existing training pipelines. With various optimizations, we achieve $3.3\times$ speed and $33\%$ VRAM reduction over CUDA+PyTorch baseline, maintaining quality, and demonstrating compatibility across GPU vendors. To the best of our knowledge, this is the first fully-Vulkan-based 3DGS training pipeline that achieves state-of-the-art performance. Code: \href{https://github.com/harry7557558/vksplat}{https://github.com/harry7557558/vksplat}

We introduce HubRouter, a pluggable module that replaces O(n^2) attention layers with O(nM) hub-mediated routing, where M << n is a small number of learned hub tokens. We demonstrate it in two from-scratch architectures: a Jamba-style hybrid and a 12-layer Transformer; retrofit into pretrained models is a tested negative case. HubRouter implements an encode-decode-score-council pipeline: M learned hubs cross-attend to all tokens, tokens project against hubs for routing fingerprints, a score head selects top-k tokens, and a sparse council attends only to the selected subset. We validate HubRouter in three settings. (1) Hub-Jamba yields a nominal 4.2% PPL improvement (200.2 vs 209.0, single seed; possibly within seed noise) and up to ~90x training throughput at sequence length 1024 in matched PyTorch-native baselines; an optimised baseline would narrow this to ~10-15x. (2) Graduated replacement of 25% of Transformer attention layers gives the best perplexity in our matched-budget sweep (268.0 vs 282.4 pure Transformer). (3) Hub-GPT provides strictly causal routing, achieving PPL 211.5 +/- 0.4 over 3 seeds (post council-causal fix); approximately 3 PPL worse than Jamba's 208.5 +/- 0.7, a measurable quality cost for avoiding O(n^2) computation. Post-fix, chunk size C has little effect; the pre-fix chunk-size benefit was an artifact of a bidirectional-council leak we found in adversarial review. A multi-seed hub-count sweep (~105 runs across M=1-32) reveals M=8-14 as the reliably-converging sub-band (4-5/5 seeds); M=6 is rescued to 5/5 by orthogonal regularization, while M>=20 shows increasing seed sensitivity. Companion paper arXiv:2603.20997 (Basu, 2026) defines the routing diagnostic task. Code and scripts will be released.

Deep learning (DL) has become a cornerstone of modern machine learning (ML) praxis. We introduce the R package mlr3torch, which is an extensible DL framework for the mlr3 ecosystem. It is built upon the torch package, and simplifies the definition, training, and evaluation of neural networks for both tabular data and generic tensors (e.g., images) for classification and regression. The package implements predefined architectures, and torch models can easily be converted to mlr3 learners. It also allows users to define neural networks as graphs. This representation is based on the graph language defined in mlr3pipelines and allows users to define the entire modeling workflow, including preprocessing, data augmentation, and network architecture, in a single graph. Through its integration into the mlr3 ecosystem, the package allows for convenient resampling, benchmarking, preprocessing, and more. We explain the package's design and features and show how to customize and extend it to new problems. Furthermore, we demonstrate the package's capabilities using three use cases, namely hyperparameter tuning, fine-tuning, and defining architectures for multimodal data. Finally, we present some runtime benchmarks.

Token pruning methods for Vision Transformers (ViTs) promise quadratic reductions in attention FLOPs by dropping uninformative patches. Yet standard variable-length attention APIs -- including FlashAttention-2's varlen and PyTorch's NestedTensor SDPA -- fail to translate these savings into proportional wall-clock gains at the short post-pruning sequence lengths typical of ViTs ($\leq$197 tokens). We identify a dispatch-overhead bottleneck: at these lengths, host-side kernel dispatch consumes ${\sim}$50,$μ$s regardless of workload, exceeding the actual GPU compute time at moderate-to-high pruning rates. We present a lightweight bidirectional Triton attention kernel whose dispatch floor is ${\sim}$24,$μ$s -- roughly 2.17$\times$ lower than FlashAttention-2 varlen -- allowing pruning savings to become visible in wall-clock time. Integrated into a complete pack-attend-unpack pipeline and evaluated on an NVIDIA RTX 4000 Ada Generation GPU, our system achieves 1.88$\times$ end-to-end throughput over padded PyTorch SDPA at standard 224$\times$224 inputs, scaling to 2.51$\times$ at 384$\times$384. Against FlashAttention-2 varlen -- the strongest baseline -- our kernel delivers 9-12% higher throughput at serving batch sizes (BS=1-4), and 2.17$\times$ lower kernel latency at 80% token pruning. Numerical correctness is verified with max absolute logit difference $<$0.004 and bit-exact top-1 predictions.

News recommender systems are devised to alleviate the information overload, attracting more and more researchers' attention in recent years. The lack of a dedicated learner-oriented news recommendation toolkit hinders the advancement of research in news recommendation. We propose a PyTorch-based news recommendation toolkit called NewsTorch, developed to support learners in acquiring both conceptual understanding and practical experience. This toolkit provides a modular, decoupled, and extensible framework with a learner-friendly GUI platform that supports dataset downloading and preprocessing. It also enables training, validation, and testing of state-of-the-art neural news recommendation models with standardized evaluation metrics, ensuring fair comparison and reproducible experiments. Our open-source toolkit is released on Github: https://github.com/whonor/NewsTorch.

Knowledge distillation (KD) is widely used to compress and post-train large language models (LLMs), yet many existing frameworks execute teacher inference with the same training-oriented backend as student optimization, leading to suboptimal efficiency. In this paper, we propose KDFlow, a novel framework for LLM distillation that features a decoupled architecture and employs SGLang for teacher inference. KDFlow combines SGLang for teacher inference with PyTorch FSDP2 for student optimization, allowing each model to run on a backend tailored to its workload. To enable efficient full-vocabulary distillation in this decoupled architecture, KDFlow transfers the teacher's final hidden states via Ray's object store and recomputes teacher logits on each student worker using a frozen copy of the teacher's output head. Furthermore, our framework supports both off-policy and on-policy distillation and incorporates cross-tokenizer algorithms through highly extensible and user-friendly APIs. Experiments show that KDFlow achieves a 1.44$\times$ to 6.36$\times$ speedup over MS-SWIFT in off-policy distillation and a 1.43$\times$ to 1.75$\times$ speedup over verl in on-policy distillation. KDFlow further scales to 64 GPUs, achieving 3.68$\times$ and 2.52$\times$ strong-scaling speedups in two representative model configurations. The code and documentation are publicly available.

Multi-hop all-reduce is the de facto backbone of large model training. As the training scale increases, the network often becomes a bottleneck, motivating the reduction of the volume of transmitted data. Accordingly, recent systems have demonstrated significant acceleration of the training process using gradient quantization. However, these systems are not optimized for multi-hop aggregation, where entries are partially summed multiple times along their aggregation topology. We present DynamiQ, a quantization framework that bridges the gap between quantization best practices and multi-hop aggregation. DynamiQ introduces novel techniques to better represent partial sums, codesigned with a decompress accumulate recompress fused kernel to facilitate fast execution. We extend PyTorch DDP to support DynamiQ over NCCL P2P, and across different LLMs, tasks, and scales, we demonstrate consistent improvement of up to 34.2% over the best among state-of-the-art methods such as Omni-Reduce, THC, and emerging standards such as MXFP4, MXFP6, and MXFP8. Further, DynamiQ is the only evaluated method that consistently reaches near-baseline accuracy (e.g., 99.9% of the BF16 baseline) and does so while significantly accelerating the training.

Large language models (LLMs) have shown strong performance on code translation between widely used programming languages. However, translation becomes much less reliable for domain-specific code, where correctness depends on framework-specific APIs and execution semantics. One example is translating deep-learning code from PyTorch to JAX, where LLM outputs often contain subtle bugs or non-idiomatic usage that prevents execution or changes behavior. Prior work suggests that curated bug-fix data from LLM-generated code can help improve code generation quality, but such resources are still limited for PyTorch-to-JAX translation. In this work, we introduce T2J, a benchmark of LLM translation bugs paired with developer-written fixes for PyTorch-to-JAX code. We start from 20 kernels in the TorchLeet dataset, translate them to JAX using the weak LLM gpt-4o-mini, and hire software developers to debug and repair the generated JAX implementations. We then use T2J to improve PyTorch-to-JAX translation for the weak LLM gpt-4o-mini via in-context learning. Our evaluation shows that using T2J yields up to 20% improvement of our proposed metric T2J-CodeTrans-Score.

This paper presents GraphMend, a compiler technique that automatically fixes FX graph breaks in PyTorch 2 programs. Although PyTorch 2 introduced TorchDynamo and TorchInductor to enable just-in-time graph compilation, certain code patterns still cause graph breaks that force execution to fall back to Python eager mode, introducing costly CPU-GPU synchronization and reducing optimization opportunities. Our investigation of 195 Hugging Face models reveals that 13.8% of models exhibit graph breaks. GraphMend automatically eliminates fixable breaks through source-level program analysis and transformations. It analyzes AST-level program structure to identify graph-break patterns and applies transformations only when their semantic preservation can be statically established. These transformations enable PyTorch to capture larger, uninterrupted FX graphs without manual refactoring by developers. We evaluate GraphMend on all 27 models found to exhibit graph breaks in our investigation. GraphMend eliminates 107 of 147 graph breaks (73%), fully fixing all breaks in 21 models. In our experiments on NVIDIA GPUs, GraphMend achieves up to 26x cold-start speedup, 5x on average, and up to 1.39x steady-state forward pass speedup. These results demonstrate that semantics-aware source-level analysis and transformation are effective complements to PyTorch's dynamic JIT compilation pipeline, substantially improving both usability and performance.

Training deep learning models on single-cell datasets with hundreds of millions of cells requires loading data from disk, as these datasets exceed available memory. While random sampling provides the data diversity needed for effective training, it is prohibitively slow due to the random access pattern overhead, whereas sequential streaming achieves high throughput but introduces biases that degrade model performance. We present scDataset, a PyTorch data loader that enables efficient training from on-disk data with seamless integration across diverse storage formats. Our approach combines block sampling and batched fetching to achieve quasi-random sampling that balances I/O efficiency with minibatch diversity. On Tahoe-100M, a dataset of 100 million cells, scDataset achieves more than two orders of magnitude speedup compared to true random sampling while working directly with AnnData files. We provide theoretical bounds on minibatch diversity and empirically show that scDataset matches the performance of true random sampling across multiple classification tasks and model architectures.

We present VLMEvalKit: an open-source toolkit for evaluating large multi-modality models based on PyTorch. The toolkit aims to provide a user-friendly and comprehensive framework for researchers and developers to evaluate existing multi-modality models and publish \textbf{reproducible} evaluation results. In VLMEvalKit, we implement over 450+ large multi-modality model configurations, including both proprietary APIs and open-source models, and support 330+ benchmarks across diverse multi-modal benchmarks. By implementing a single interface, new models can be easily added to the toolkit, while the toolkit automatically handles the remaining workloads, including data preparation, distributed inference, prediction post-processing, and metric calculation. VLMEvalKit has also evolved to a broader evaluation suite spanning video/audio, document understanding, GUI grounding, spatial reasoning, safety, scientific reasoning, and multi-turn dialogue. Based on the evaluation results obtained with the toolkit, we host the OpenVLM Leaderboard, a comprehensive leaderboard to track the progress of multi-modality learning research. The toolkit is released on https://github.com/open-compass/VLMEvalKit and is actively maintained.