Articles by matt_d
3

Tensor Shapes in Pyrefly – Avik Chaudhuri – PyCon US 2026 Typing Summit [video] (youtube.com)

1

BenchEvolver: Frontier Task Synthesis via Solution-Centric Evolution (benchevolver.github.io)

2

JITDomain: Instruction-level JIT code isolation (sciencedirect.com)

2

Serving Transformers: Lessons from the Trenches – Stanford CS25 Transformers [video] (youtube.com)

1

Constrained Adaptive Rejection Sampling (arxiv.org)

2

Training an Agentic Router for Optimal Cost-Performance on SWE Tasks (appliedcompute.com)

1

Diagramming Program Values by Spatial Refinement (brownplt.org)

1

Agent Arena: Causal Evaluation of Agents in the Real World (arena.ai)

1

Can LLMs Reason Structurally? Benchmarking via the Lens of Data Structures (arxiv.org)

1

Recent improvements to the type checker – Swift Compiler (swift.org)

1

Semantic Reification: A New Paradigm for Random Program Generation (sigplan.org)

1

GPU Forecasters: Language Models as Selective Surrogates for Kernel Optimization (arxiv.org)

1

Type-Error Ablation and AI Coding Agents (arxiv.org)

1

Session-Aware Agentic Routing: Continuity-Aware Model Selection for Long-Horizon (vllm.ai)

1

Directionality in Low Precision (constantinides.net)

1

O-POPE: High-Frequency Pipelined Outer Product based GEMM acceleration (arxiv.org)

2

VideoFDB: Evaluating Full-Duplex Vision-Speech Capabilities in Agents (nvidia.com)

51

Strace-ui, Bonsai_term, and the TUI renaissance (janestreet.com)

2

Why Larger Models Learn More: Capacity, Interference, Rare-Task Retention (arxiv.org)

2

PyTorch's playbook for AI coding, as of May 2026 (pytorch.org)

2

When does fragmentation occur in the CUDA caching allocator? (pytorch.org)

2

Evaluating Apple Silicon for Data Processing [pdf] (tum.de)

2

PassNet: Scaling Large Language Models for Graph Compiler Pass Generation (arxiv.org)

3

Do GPUs Need New Tabular File Formats? (arxiv.org)

1

Polyhedral Compilation in MLIR (sajidzubair.substack.com)

1

Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs (arxiv.org)

2

Three Trends from MLSys 2026 (modular.com)

3

What are the real problems of continual learning? (infinitefaculty.substack.com)

1

What "Memory Compiler" Actually Means: From Bitcells to GDS Tiling (thecloudlet.github.io)

5

Understanding Inference Scaling for LLMs: Bottlenecks, Trade-Offs, and Perf (arxiv.org)

1

MIT EECS/CSAIL Agentic Coding in Practice Seminar Series (csail.mit.edu)

1

The Load-Balance Problem Behind Hybrid Parallelism (hecate0821.github.io)

1

Delayed Tensor Parallelism for Faster Transformer Inference (kog.ai)

1

Continuous Diffusion Models Can Obey Formal Syntax (arxiv.org)

1

HartBreaker: Deterministic Fuzzing of Multi-Hart RISC-V CPUs (ethz.ch)

4

Tuning LLVM's SLP Vectorizer Cost Model (kaving.me)

1

A Friendly Tour of Substructural, Uniqueness, Ownership, Capabilities and more! (federicobruzzone.github.io)

1

FlashLib: Bringing Flash Magic to Classical Machine Learning Operators (flashml-org.github.io)

1

FML-Bench: A Controlled Study of AI Research Agent Strategies (arxiv.org)

2

Finding deadlocks in CuTe kernels with SPIN (metaworld.me)

2

A Case for Tracing Based DSL Kernel Languages (metaworld.me)

1

You don't need all the LLM benchmarks (smola.org)

1

Elusive order of async GPU kernels: scheduling, abstractions, DSL implications (ianbarber.blog)

1

MileStone: A Multi-Objective Compiler Phase Ordering Framework (arxiv.org)

3

SSV: Sparse Speculative Verification for Efficient LLM Inference (arxiv.org)

2

Characterizing Real-World Bugs in Tile Programs for Automated Bug Detection (arxiv.org)

3

Characterization of machine learning compilers for LLM inference on NVIDIA GPUs (springer.com)

1

Chip design from the bottom up – Reiner Pope [video] (youtube.com)

2

LT2: Linear-Time Looped Transformers (charlesdddd.github.io)

2

Event Tensor: A Unified Abstraction for Compiling Dynamic Megakernel (arxiv.org)

1

PopPy: Opportunistically Exploiting Parallelism in Python Compound AI Apps (arxiv.org)

33

CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs (arxiv.org)

1

[RFC] Open Access to Standards Documents – LLVM Project (llvm.org)

3

Curly braces: An evolution of UNIX and C (thalia.dev)

1

NanoTag: Systems Support for Efficient Byte-Granular Overflow Detection on Arm (github.com/ice-rlab)

1

InferenceBench: A Benchmark for Open-Ended Inference Optimization by AI Agents (inferencebench.ai)

1

Tracking Capabilities for Safer Agents (arxiv.org)

2

Scalable Packed Layouts for Vector-Length-Agnostic ML Code Generation (arxiv.org)

1

Verifying EDA and compiler optimizations once and for all (samuelcoward.co.uk)

1

StepStone: LLM-Based GPU Kernel Driver Fuzzing via User-Space Libraries [pdf] (ucr.edu)

1

Graded Modal Types for Memory and Communication Safety (kent.ac.uk)

1

Systems Are Changing: The Architect's Role in the Era of Agentic Co-Design (sigarch.org)

1

Code-Specify-Test-Debug-Prove: Flexibly Integrating Separation Logic [pdf] (cam.ac.uk)

3

Detecting Relaxed Memory Concurrency Bugs in C and C++ Compilers (lukegeeson.com)

2

The downgrading semantics of memory safety (Extended version) (arxiv.org)

1

Direction-Preserving Number Representations (arxiv.org)

1

On the Unreasonable Effectiveness of PBT for Validating Formal Specifications (proofsandintuitions.net)

2

Understanding, Analyzing, and Optimizing Agentic AI: A CPU-Centric Perspective (arxiv.org)

2

Getting Confidence in (Agentic) Code (ucsd-cse-115-215.github.io)

1

Compute Optimal Tokenization: Scaling Laws for Data Compression in LLMs (co-tok.github.io)

1

Nitsum: Serving Tiered LLM Requests with Adaptive Tensor Parallelism (wuklab.io)

1

SuperInfer: SLO-Aware Rotary Scheduling and Memory Management for LLM Inference (supercomputing-system-ai-lab.github.io)

7

[flagged] KV cache is becoming the memory hierarchy of inference (touchdown-labs.com)

1

Ada-MK: Adaptive MegaKernel Optimization via DAG-Based Search for LLM Inference (arxiv.org)

42

How to Write to SSDs [pdf] (vldb.org)

1

Scalable GPU Acceleration of Scalar Functions in Analytical Databases [pdf] (vldb.org)

1

The agent principal-agent problem (crawshaw.io)

1

FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale (frontier-cs.org)

1

Demystifying the Silence of Correctness Bugs in PyTorch Compiler (arxiv.org)

1

Fork, Explore, Commit: OS Primitives for Agentic Exploration (arxiv.org)

1

Systematically Auditing AI Agent Benchmarks with BenchJack (arxiv.org)

8

mimalloc: A new, high-performance, scalable memory allocator for the modern era (microsoft.com)

1

Let AI Agents Write Your Serving Stack with VibeServe (washington.edu)

2

TLX: Hardware-Native, Evolvable MIMW GPU Compiler for Large-Scale Production (arxiv.org)

2

TorchLean: Verified Neural Networks in Lean (robertj1.com)

98

Deterministic Fully-Static Whole-Binary Translation Without Heuristics (arxiv.org)

1

Dynamic Persistent Tile Scheduling w/ Cluster Launch Control (CLC) on Blackwell (colfax-intl.com)

1

VibeServe: Can AI Agents Build Bespoke LLM Serving Systems? (github.com/uw-syfi)

2

CCL-Bench 1.0: A Trace-Based Benchmark for LLM Infrastructure (arxiv.org)

1

Microbenchmark-Driven Analytical Performance Modeling Across Modern GPUs (arxiv.org)

2

PyTorch DevLog (pytorch.org)

2

VDCores: Resource Decoupled Programming and Execution for Asynchronous GPU (arxiv.org)

1

Aurora: A Leverage-Aware Optimizer for Rectangular Matrices (tilderesearch.com)

1

The Two Abstractions of System Design: Hide or Reduce (muratbuffalo.blogspot.com)

1

Practical Formal Verification for MLIR Programs (arxiv.org)

1

Kerncap: Automated Kernel Extraction and Isolation for AMD GPUs (arxiv.org)

2

Capsules: Compile-time lock discipline in OxCaml (kcsrk.info)

1

Data Race Freedom in OxCaml (kcsrk.info)

2

cuda-oxide: a custom rustc backend for compiling GPU kernels in pure Rust (github.com/nvlabs)

3

A case study with Aeneas and jxl-rs (protzenko.fr)