Stories by matt_d

Directionality in Low Precision

O-POPE: High-Frequency Pipelined Outer Product based GEMM acceleration

VideoFDB: Evaluating Full-Duplex Vision-Speech Capabilities in Agents

Strace-ui, Bonsai_term, and the TUI renaissance

Why Larger Models Learn More: Capacity, Interference, Rare-Task Retention

PyTorch's playbook for AI coding, as of May 2026

When does fragmentation occur in the CUDA caching allocator?

Evaluating Apple Silicon for Data Processing [pdf]

PassNet: Scaling Large Language Models for Graph Compiler Pass Generation

Do GPUs Need New Tabular File Formats?

Polyhedral Compilation in MLIR

Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs

Three Trends from MLSys 2026

What are the real problems of continual learning?

What "Memory Compiler" Actually Means: From Bitcells to GDS Tiling

Understanding Inference Scaling for LLMs: Bottlenecks, Trade-Offs, and Perf

MIT EECS/CSAIL Agentic Coding in Practice Seminar Series

The Load-Balance Problem Behind Hybrid Parallelism

Delayed Tensor Parallelism for Faster Transformer Inference

Continuous Diffusion Models Can Obey Formal Syntax

1 2 3 4 5 6 7 8 9 10