Stories by martianlantern

Gumbel Max trick for LLM sampling

Strassen's Matmul with AVX 512

System Design Interview: An insider's guide (Alex Xu) [pdf]

3 Months of Using Neovim

Fail Faster on Your Idea

Fail Faster on Your Ideas

Distilling persona vectors into LLM weights

Fast Median Filter over arbitrary datatypes

Updating My Bash Prompt

How to Establish SSH over Tor

Median Filter over Arbitrary Datatypes

Show HN: Python Asyncio Puzzles

LLM Routing Strategies

Paged Attention Performance Analysis

Paged Attention Performance Analysis

ThinkMesh: A Python lib for parallel thinking in LLMs

Train a small GPT to solve mazes

Reasoning Cot Injection

Injecting doubts in the CoT of reasoning models

Injecting doubts in CoT of reasoning models