Krisztián Flautner

Joshua Brodsky, Dhravid Kumar, Savini Kashmira et al.

Machine learning models are increasingly embedded in everyday software, and most of their runtime is spent in a small set of compute kernels such as matrix multiplication, convolution, and normalization. Optimizing these kernels is one of the most direct ways to reduce latency and cost, but it has traditionally required expert engineers to hand-write low-level GPU code. Agentic systems built on large language models (LLMs) can now generate and optimize kernels with far less human effort, yet existing tools are largely evaluated on randomly generated tensors and isolated kernels, emit standalone CUDA code that developers must manually reintegrate, mostly target only LLM PyTorch models, and offer limited support for inspecting and debugging results. We present Kernel Forge, an open-source, end-to-end agentic harness that accepts any unmodified PyTorch model in place. Kernel Forge supports vision, diffusion, and LLM workloads, uses Monte Carlo Tree Search (MCTS) to explore multiple optimization paths rather than a single linear refinement chain, and ships with a graphical user interface for monitoring progress, inspecting candidate kernels, and debugging failures. We evaluate Kernel Forge on four PyTorch models spanning vision, diffusion, and LLM workloads on an NVIDIA DGX Spark with GB10 GPU. With only 50 optimization iterations per kernel, it optimizes 14 kernels to outperform PyTorch eager mode, reaching $1.52\times$ on adaptive\_avgpool2d in ResNet-50, $1.70\times$ on group\_norm in Stable Diffusion 3.5 Medium, $2.83\times$ on softmax in Gemma 4 E2B, and $1.54\times$ on softmax in Qwen 3.5 35B-A3B.

Haojie Ye, Arjun Laxman, Yichao Yuan et al.

Money laundering detection faces challenges due to excessive false positives and inadequate adaptation to sophisticated multi-stage schemes that exploit modern financial networks. Graph analytics and AI are promising tools, but they struggle with the fuzziness of laundering patterns, which exhibit structural and temporal variations. Conventional data mining techniques require the detailed enumeration of pattern variants, which not only complicates the analyst's task to specify them, but also leads to large run-time overheads and difficulty training accurate AI models. The paper presents BlazingAML, a scalable AML system design that introduces: 1. A novel multi-stage framework for expressing fuzzy money laundering patterns 2. A domain-specific compiler that transforms high-level pattern descriptions into high-performance code for CPU and GPU back-ends The multi-stage abstraction decomposes complex laundering schemes into logical stages connected by graph operations, enabling diverse patterns to be expressed using unified primitives while capturing structural and temporal fuzziness. The compiler applies sophisticated optimizations, eliminating manual parallel programming requirements for financial analysts. Evaluation on IBM AML datasets shows BlazingAML achieves the same F1 score as state-of-the-art approaches while delivering 210x and 333x higher speedup on CPU and GPU respectively, with superior scalability.

Priyal Deep, Shane Emmons, Amy Fox et al.

LLM-powered applications routinely embed secrets in system prompts, yet models can be tricked into revealing them. We built an adaptive attacker that evolves its strategies over hundreds of rounds and tested it against nine defense configurations across more than 20,000 attacks. Every defense that relied on the model to protect itself eventually broke. The only defense that held was output filtering, which checks the model's responses via hardcoded rules in separate application code before they reach the user, achieving zero leaks across 15,000 attacks. These results demonstrate that security boundaries must be enforced in application code, not by the model being attacked. Until such defenses are verified by tools like Swept AI, AI systems handling sensitive operations should be restricted to internal, trusted personnel.

Chandra Irugalbandara, Ashish Mahendra, Roland Daynauth et al.

Many companies use large language models (LLMs) offered as a service, like OpenAI's GPT-4, to create AI-enabled product experiences. Along with the benefits of ease-of-use and shortened time-to-solution, this reliance on proprietary services has downsides in model control, performance reliability, uptime predictability, and cost. At the same time, a flurry of open-source small language models (SLMs) has been made available for commercial use. However, their readiness to replace existing capabilities remains unclear, and a systematic approach to holistically evaluate these SLMs is not readily available. This paper presents a systematic evaluation methodology and a characterization of modern open-source SLMs and their trade-offs when replacing proprietary LLMs for a real-world product feature. We have designed SLaM, an open-source automated analysis tool that enables the quantitative and qualitative testing of product features utilizing arbitrary SLMs. Using SLaM, we examine the quality and performance characteristics of modern SLMs relative to an existing customer-facing implementation using the OpenAI GPT-4 API. Across 9 SLMs and their 29 variants, we observe that SLMs provide competitive results, significant performance consistency improvements, and a cost reduction of 5x~29x when compared to GPT-4.

Roland Daynauth, Christopher Clarke, Krisztian Flautner et al.

Deciding which large language model (LLM) to use is a complex challenge. Pairwise ranking has emerged as a new method for evaluating human preferences for LLMs. This approach entails humans evaluating pairs of model outputs based on a predefined criterion. By collecting these comparisons, a ranking can be constructed using methods such as Elo. However, applying these algorithms as constructed in the context of LLM evaluation introduces several challenges. In this paper, we explore the effectiveness of ranking systems for head-to-head comparisons of LLMs. We formally define a set of fundamental principles for effective ranking and conduct a series of extensive evaluations on the robustness of several ranking algorithms in the context of LLMs. Our analysis uncovers key insights into the factors that affect ranking accuracy and efficiency, offering guidelines for selecting the most appropriate methods based on specific evaluation contexts and resource constraints.

Savini Kashmira, Jayanaka Dantanarayana, Thamirawaran Sathiyalogeswaran et al.

This paper presents GraphMend, a high-level compiler that eliminates FX graph breaks in PyTorch 2 programs. Although PyTorch 2 introduced TorchDynamo and TorchInductor to enable just-in-time graph compilation, unresolved dynamic control flow and unsupported Python constructs often fragment models into multiple FX graphs. These fragments force frequent fallbacks to eager mode, incur costly CPU-to-GPU synchronizations, and reduce optimization opportunities. GraphMend addresses this limitation by analyzing and transforming source code before execution. Built on the Jac compilation framework, GraphMend introduces two code transformations that remove graph breaks due to dynamic control flow and Python I/O functions. This design allows PyTorch's compilation pipeline to capture larger, uninterrupted FX graphs without requiring manual refactoring by developers. Evaluation across eight Hugging Face models shows that GraphMend removes all fixable graph breaks due to dynamic control flow and Python I/O functions, driving the break count to 0 in 6 models and reducing it from 5 to 2 in another model. On NVIDIA RTX 3090 and A40 GPUs, GraphMend achieves up to 75% latency reductions and up to 8% higher end-to-end throughput. These results demonstrate that high-level code transformation is an effective complement to PyTorch's dynamic JIT compilation pipeline, substantially improving both usability and performance.

Savini Kashmira, Jayanaka L. Dantanarayana, Krisztián Flautner et al.

Conventional Retrieval Augmented Generation (RAG) approaches are common in text-based applications. However, they struggle with structured, interconnected datasets like knowledge graphs, where understanding underlying relationships is crucial for accurate retrieval. A common direction in graph-based retrieval employs iterative, rule-based traversal guided by Large Language Models (LLMs). Such existing iterative methods typically combine reasoning with single hop traversal at each step, making them vulnerable to LLM reasoning errors and hallucinations that ultimately hinder the retrieval of relevant information. To address these limitations, we propose GraphRunner, a novel graph-based retrieval framework that operates in three distinct stages: planning, verification, and execution. This introduces high-level traversal actions that enable multi-hop exploration in a single step. It also generates a holistic traversal plan, which is verified against the graph structure and pre-defined traversal actions, reducing reasoning errors and detecting hallucinations before execution. GraphRunner significantly reduces LLM reasoning errors and detects hallucinations through validation. Our evaluation using the GRBench dataset shows that GraphRunner consistently outperforms existing approaches, achieving 10-50% performance improvements over the strongest baseline while reducing inference cost by 3.0-12.9x and response generation time by 2.5-7.1x, making it significantly more robust and efficient for graph-based retrieval tasks.

Roland Daynauth, Christopher Clarke, Krisztian Flautner et al.

The LLM-as-a-Judge paradigm offers a scalable, reference-free approach for evaluating language models. Although several calibration techniques have been proposed to better align these evaluators with human judgment, prior studies focus primarily on narrow, well-structured benchmarks. As a result, it remains unclear whether such calibrations generalize to real-world, open-ended tasks. In this work, we show that SOTA calibrated evaluators often fail in these settings, exhibiting weak or even negative correlation with human judgments. To address this, we propose SLMEval, a novel and efficient calibration method based on entropy maximization over a small amount of human preference data. By estimating a latent distribution over model quality and reweighting evaluator scores accordingly, SLMEval achieves strong correlation with human evaluations across two real-world production use cases and the public benchmark. For example, on one such task, SLMEval achieves a Spearman correlation of 0.57 with human judgments, while G-Eval yields a negative correlation. In addition, SLMEval reduces evaluation costs by 5-30x compared to GPT-4-based calibrated evaluators such as G-eval.

Savini Kashmira, Jayanaka L. Dantanarayana, Joshua Brodsky et al.

Retrieval-Augmented Generation (RAG) is one of the leading and most widely used techniques for enhancing LLM retrieval capabilities, but it still faces significant limitations in commercial use cases. RAG primarily relies on the query-chunk text-to-text similarity in the embedding space for retrieval and can fail to capture deeper semantic relationships across chunks, is highly sensitive to chunking strategies, and is prone to hallucinations. To address these challenges, we propose TOBUGraph, a graph-based retrieval framework that first constructs the knowledge graph from unstructured data dynamically and automatically. Using LLMs, TOBUGraph extracts structured knowledge and diverse relationships among data, going beyond RAG's text-to-text similarity. Retrieval is achieved through graph traversal, leveraging the extracted relationships and structures to enhance retrieval accuracy, eliminating the need for chunking configurations while reducing hallucination. We demonstrate TOBUGraph's effectiveness in TOBU, a real-world application in production for personal memory organization and retrieval. Our evaluation using real user data demonstrates that TOBUGraph outperforms multiple RAG implementations in both precision and recall, significantly improving user experience through improved retrieval accuracy.

Christopher Clarke, Yuzhao Heng, Yiping Kang et al.

Conventional approaches to text classification typically assume the existence of a fixed set of predefined labels to which a given text can be classified. However, in real-world applications, there exists an infinite label space for describing a given text. In addition, depending on the aspect (sentiment, topic, etc.) and domain of the text (finance, legal, etc.), the interpretation of the label can vary greatly. This makes the task of text classification, particularly in the zero-shot scenario, extremely challenging. In this paper, we investigate the task of zero-shot text classification with the aim of improving the ability of pre-trained language models (PLMs) to generalize to both seen and unseen data across varying aspects and domains. To solve this we introduce two new simple yet effective pre-training strategies, Implicit and Explicit pre-training. These methods inject aspect-level understanding into the model at train time with the goal of conditioning the model to build task-level understanding. To evaluate this, we construct and release UTCD, a new benchmark dataset for evaluating text classification in zero-shot settings. Experimental results on UTCD show that our approach achieves improved zero-shot generalization on a suite of challenging datasets across an array of zero-shot formalizations.

Jason Mars, Yiping Kang, Roland Daynauth et al.

Today's production scale-out applications include many sub-application components, such as storage backends, logging infrastructure and AI models. These components have drastically different characteristics, are required to work in collaboration, and interface with each other as microservices. This leads to increasingly high complexity in developing, optimizing, configuring, and deploying scale-out applications, raising the barrier to entry for most individuals and small teams. We developed a novel co-designed runtime system, Jaseci, and programming language, Jac, which aims to reduce this complexity. The key design principle throughout Jaseci's design is to raise the level of abstraction by moving as much of the scale-out data management, microservice componentization, and live update complexity into the runtime stack to be automated and optimized automatically. We use real-world AI applications to demonstrate Jaseci's benefit for application performance and developer productivity.