Skip to content

Top Best Ask Show New Jobs

somnial | Better HN

somnial

3 karmaJoined October 30, 202510 submissions

Recent submissions

1

Pushing memory bound CUDA kernels past the speed of light with data compression (opens in new tab)

(fergusfinn.com)

2somnial28d ago0

2

Speculative KV coding: ~4× losslessly compressed KV cache using a small model (opens in new tab)

(fergusfinn.com)

2somnial1mo ago0

3

70x faster cold(ish) starts for SGLang (opens in new tab)

(fergusfinn.com)

1somnial1mo ago0

4

LLM powered data structures: A lock-free binary search tree (opens in new tab)

(fergusfinn.com)

1somnial5mo ago0

5

Parallel Primitives for Multi-Agent Workflows (opens in new tab)

(fergusfinn.com)

1somnial5mo ago0

6

Scheduling in LLM Inference (opens in new tab)

(fergusfinn.com)

1somnial7mo ago0

7

How fast can an LLM go? (opens in new tab)

(fergusfinn.com)

2somnial7mo ago0