Large language models (LLMs) can give out erroneous answers to factually rooted questions either as a result of undesired training outcomes or simply because the world has moved on after a certain knowledge cutoff date. Under such scenarios, *knowledge editing* often comes to the rescue by delivering efficient patches for such erroneous answers without significantly altering the rest, where many editing methods have seen reasonable success when the editing targets are simple and direct (e.g., *``what club does Lionel Messi currently play for?''*). However, knowledge fragments like this are often deeply intertwined in the real world, making effectively propagating the editing effect to non-directly related questions a practical challenge (to entertain an extreme example: [*"What car did the wife of the owner of the club that Messi currently plays for used to get to school in the 80s?"*](youtube.com/watch?v=DbwiHC1Fu-E\&t=132s)). Prior arts have coined this task as *multi-hop knowledge editing* with the most popular dataset being MQuAKE, serving as the sole evaluation benchmark for many later proposed editing methods due to the expensive nature of constructing knowledge editing datasets at scale. In this work, we reveal that **up to 33\% or 76\% of \mquake{}'s questions and ground truth labels are, in fact, corrupted in various fashions due to some unintentional clerical or procedural oversights**. Our work provides a detailed audit of MQuAKE's error pattern and a comprehensive fix without sacrificing its dataset capacity. Additionally, we benchmarked almost all proposed MQuAKE-evaluated editing methods on our post-fix dataset, **MQuAKE-Remastered**. We observe that many methods try to overfit the original MQuAKE by exploiting some dataset idiosyncrasies of MQuAKE. We provide a guideline on how to approach such datasets faithfully and show that a simple, minimally invasive approach — **GWalk** — can offer beyond SOTA editing performance without such exploitation. The MQuAKE-Remastered datasets and utilities are available at [huggingface.co/datasets/henryzhongsc/MQuAKE-Remastered](https://huggingface.co/datasets/henryzhongsc/MQuAKE-Remastered) and [github.com/henryzhongsc/MQuAKE-Remastered](https://github.com/henryzhongsc/MQuAKE-Remastered), respectively.

Characterizing and predicting the training performance of modern machine learning (ML) workloads on compute systems with compute and communication spread between CPUs, GPUs, and network devices is not only the key to optimization and planning but also a complex goal to achieve. The primary challenges include the complexity of synchronization and load balancing between CPUs and GPUs, the variance in input data distribution, and the use of different communication devices and topologies (e.g., NVLink, PCIe, network cards) that connect multiple compute devices, coupled with the desire for flexible training configurations. Built on top of our prior work for single-GPU platforms, we address these challenges and enable multi-GPU performance modeling1by incorporating (1) data-distribution-aware performance models for embedding table lookup, and (2) data movement prediction of communication collectives, into our upgraded performance modeling pipeline equipped with inter-and intra-rank synchronization for ML workloads trained on multi-GPU platforms. Beyond accurately predicting the per-iteration training time of deep learning recommendation models (DLRM) models with random configurations with a geomean error of 5.21% on two multi-GPU platforms, our prediction pipeline generalizes well to other types of ML workloads, such as Transformer-based natural language processing (NLP) models with a geomean error of 3.00%. Moreover, even without actually running ML workloads like DLRMs on the hardware, it is capable of generating insights such as quickly selecting the fastest embedding table sharding configuration (with a success rate of 85%).

Building large AI fleets to support the rapidly growing DL workloads is an active research topic for modern cloud providers. Generating accurate benchmarks plays an essential role in designing the fast-paced software and hardware solutions in this space. Two fundamental challenges to make this scalable are (i) workload representativeness and (ii) the ability to quickly incorporate changes to the fleet into the benchmarks.

We devise a performance model for GPU training of Deep Learning Recommendation Models (DLRM), whose GPU utilization is low compared to other well-optimized CV and NLP models. We show that both the device active time (the sum of kernel runtimes) but also the device idle time are important components of the overall device time. We therefore tackle them separately by (1) flexibly adopting heuristic-based and ML-based kernel performance models for operators that dominate the device active time, and (2) categorizing operator overheads into five types to determine quantitatively their contribution to the device active time. Combining these two parts, we propose a critical-path-based algorithm to predict the per-batch training time of DLRM by traversing its execution graph. We achieve less than 10% geometric mean average error (GMAE) in all kernel performance modeling, and 4.61% and 7.96% geomean errors for GPU active time and overall E2E per-batch training time prediction with overheads from individual workloads, respectively. A slight increase of 2.19% incurred in E2E prediction error with shared overheads across workloads suggests the feasibility of using shared overheads in large-scale prediction. We show that our general performance model not only achieves low prediction error on DLRM, which has highly customized configurations and is dominated by multiple factors but also yields comparable accuracy on other compute-bound ML models targeted by most previous methods. Using this performance model and graph-level data and task dependency analysis, we show our system can provide more general model-system co-design than previous methods.

We devise a performance model for GPU training of Deep Learning Recommendation Models (DLRM), which has low GPU utilization (i.e., the percentage of per-batch training time when kernels are running on the device) compared to other well-optimized vision (CV) and natural language processing (NLP) models. We show that both the device active time (the sum of kernel runtimes) and idle time are important components of the overall device time, and can be tackled separately by (1) flexibly adopting heuristic- and ML-based kernel performance models for kernels that dominate the device active time, and (2) categorizing operator overheads into five types to quantitatively determine their contribution to the overall device time. Combining these two parts, we propose a critical-path-based algorithm to predict the per-batch training time of DLRM by traversing its execution graph. We achieve less than 10% geometric mean absolute error (GMAE) in all kernel performance modeling, and 5.23% and 7.96% geomean errors, respectively, for GPU active time and overall end-to-end per-batch training time prediction on the highly-customized and multi-factor dominated DLRM architectures. We also demonstrate our performance model’s ability to generalize to other compute-bound DL models targeted by most previous methods and better assist general model-system co-design than previous work.

Embedding learning is an important technique in deep recommendation models to map categorical features to dense vectors. However, the embedding tables often demand an extremely large number of parameters, which become the storage and efficiency bottlenecks. Distributed training solutions have been adopted to partition the embedding tables into multiple devices. However, the embedding tables can easily lead to imbalances if not carefully partitioned. This is a significant design challenge of distributed systems named embedding table sharding, i.e., how we should partition the embedding tables to balance the costs across devices, which is a non-trivial task because 1) it is hard to efficiently and precisely measure the cost, and 2) the partition problem is known to be NP-hard. In this work, we introduce our novel practice in Meta, namely AutoShard, which uses a neural cost model to directly predict the multi-table costs and leverages deep reinforcement learning to solve the partition problem. Experimental results on an open-sourced large-scale synthetic dataset and Meta's production dataset demonstrate the superiority of AutoShard over the heuristics. Moreover, the learned policy of AutoShard can transfer to sharding tasks with various numbers of tables and different ratios of the unseen tables without any fine-tuning. Furthermore, AutoShard can efficiently shard hundreds of tables in seconds. The effectiveness, transferability, and efficiency of AutoShard make it desirable for production use. Our algorithms have been deployed in Meta production environment. A prototype is available at https://github.com/daochenzha/autoshard

We study embedding table placement for distributed recommender systems, which aims to partition and place the tables on multiple hardware devices (e.g., GPUs) to balance the computation and communication costs. Although prior work has explored learning-based approaches for the device placement of computational graphs, embedding table placement remains to be a challenging problem because of 1) the operation fusion of embedding tables, and 2) the generalizability requirement on unseen placement tasks with different numbers of tables and/or devices. To this end, we present DreamShard, a reinforcement learning (RL) approach for embedding table placement. DreamShard achieves the reasoning of operation fusion and generalizability with 1) a cost network to directly predict the costs of the fused operation, and 2) a policy network that is efficiently trained on an estimated Markov decision process (MDP) without real GPU execution, where the states and the rewards are estimated with the cost network. Equipped with sum and max representation reductions, the two networks can directly generalize to any unseen tasks with different numbers of tables and/or devices without fine-tuning. Extensive experiments show that DreamShard substantially outperforms the existing human expert and RNN-based strategies with up to 19% speedup over the strongest baseline on large-scale synthetic tables and our production tables. The code is available.

Deep learning recommendation systems at scale have provided remarkable gains through increasing model capacity (i.e. wider and deeper neural networks), but it comes at significant training cost and infrastructure cost. Model pruning is an effective technique to reduce computation overhead for deep neural networks by removing redundant parameters. However, modern recommendation systems are still thirsty for model capacity due to the demand for handling big data. Thus, pruning a recommendation model at scale results in a smaller model capacity and consequently lower accuracy. To reduce computation cost without sacrificing model capacity, we propose a dynamic training scheme, namely alternate model growth and pruning, to alternatively construct and prune weights in the course of training. Our method leverages structured sparsification to reduce computational cost without hurting the model capacity at the end of offline training so that a full-size model is available in the recurring training stage to learn new data in real time. To the best of our knowledge, this is the first work to provide in-depth experiments and discussion of applying structural dynamics to recommendation systems at scale to reduce training cost. The proposed method is validated with an open-source deep learning recommendation model (DLRM) and state-of-the-art industrial-scale production models.

We present a practical approach to generate stochastic anisotropic samples with Poisson-disk characteristic over a two-dimensional domain. In contrast to isotropic samples, we understand anisotropic samples as non-overlapping ellipses whose size and density match a given anisotropic metric. Anisotropic noise samples are useful for many visualization and graphics applications. The spot samples can be used as input for texture generation, e.g., line integral convolution (LIC), but can also be used directly for visualization. The definition of the spot samples using a metric tensor makes them especially suitable for the visualization of tensor fields that can be translated into a metric. Our work combines ideas from sampling theory and mesh generation. To generate these samples with the desired properties we construct a first set of non-overlapping ellipses whose distribution closely matches the underlying metric. This set of samples is used as input for a generalized anisotropic Lloyd relaxation to distribute noise samples more evenly. Instead of computing the Voronoi tessellation explicitly, we introduce a discrete approach which combines the Voronoi cell and centroid computation in one step. Our method supports automatic packing of the elliptical samples, resulting in textures similar to those generated by anisotropic reaction-diffusion methods. We use Fourier analysis tools for quality measurement of uniformly distributed samples. The resulting samples have nice sampling properties, for example, they satisfy a blue noise property where low frequencies in the power spectrum are reduced to a minimum.

A shell map is a bijective mapping between shell space and texture space that can be used to generate small-scale features on surfaces using a variety of modeling techniques. The method is based upon the generation of an offset surface and the construction of a tetrahedral mesh that fills the space between the base surface and its offset. By identifying a corresponding tetrahedral mesh in texture space, the shell map can be implemented through a straightforward barycentric-coordinate map between corresponding tetrahedra. The generality of shell maps allows texture space to contain geometric objects, procedural volume textures, scalar fields, or other shell-mapped objects.

The physical interpretation of mathematical features of tensor fields is highly application-specific. Existing visualization methods for tensor fields only cover a fraction of the broad application areas. We present a visualization method tailored specifically to the class of tensor field exhibiting properties similar to stress and strain tensors, which are commonly encountered in geomechanics. Our technique is a global method that represents the physical meaning of these tensor fields with their central features: regions of compression or expansion. The method is based on two steps: first, we define a positive definite metric, with the same topological structure as the tensor field; second, we visualize the resulting metric. The eigenvector fields are represented using a texture-based approach resembling line integral convolution (LIC) methods. The eigenvalues of the metric are encoded in free parameters of the texture definition. Our method supports an intuitive distinction between positive and negative eigenvalues. We have applied our method to synthetic and some standard data sets, and "real" data from earth science and mechanical engineering application.

Source Viewer 3D is a software visualization framework that uses a 3D metaphor to represent software system and analysis data. The 3D representation is based on the SeeSoft pixel metaphor. It extends the original metaphor by rendering the visualization in a 3D space. New, object-based manipulation methods and simultaneous alternative mappings are available to the user.