Discussion on "Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase"

Aman Karki · 2026-09-27T16:40:53.617Z

My first pull request to llama.cpp was 95 lines of CUDA. Georgi Gerganov, the guy who started the project, merged it the next day. The second one was even smaller. Nine new lines of code, and the chan

Discussion on "Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase" | Hashnode

Search Hashnode

Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase

Responses