Mainly Matmul

A practical look at what a megawatt of power is worth and the factors that shape its value.

Predicting multiple tokens jointly with a tensor-train language modelling head.

Why DSpark's parallel generation still depends on autoregressive sampling.

Use BatchBench to model real LLM traffic, measure the tokens a deployment serves, and turn throughput into an input and output token price.

Starting from quadratic attention, derive linear attention, DeltaNet, Gated DeltaNet, and finally Kimi Delta Attention—then connect the recurrence to its chunkwise Triton implementation.

A draft note on what would make an open source model genuinely frontier in late 2026.

A look at DFlash, DiffusionGemma, and Nemotron-Labs-Diffusion as three approaches to diffusion language models.

Improving SGLang’s Offloading Engine.

Using tensor network notation to understand multi-head attention, MQA, talking-heads attention, and DeepSeek's MLA.

A draft post comparing OCR systems on math, tables, scans, and dense technical pages.

Speculative decoding where draft tokens are generated during queueing time and verified when the request is serviced.

High throughput inference of LLMs using JIT weight offloading to optimize KV Cache.