---
title: "Discussion on \"Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase\" | Hashnode"
description: "Discussion on \"Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase\". A 95-line pooling kernel, a 9-line fix for a silent CPU fallback, and how ggml decides where your ops run"
url: "https://hashnode.com/posts/llama-cpp-two-cuda-prs/6ab9471561ca1c18dd7082e6"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Two CUDA PRs into llama.cpp, and what they taught me about a 100K-star codebase](https://amankarki.hashnode.dev/llama-cpp-two-cuda-prs?utm_source=hashnode&utm_medium=feed)

A 95-line pooling kernel, a 9-line fix for a silent CPU fallback, and how ggml decides where your ops run

Sep 27 · 11 min read · amankarki.hashnode.dev  
Read the article

-   [#cuda](https://hashnode.com/tag/cuda)
-   [#llama-cpp](https://hashnode.com/tag/llama-cpp)
-   [#cpp](https://hashnode.com/tag/cpp)
-   [#opensource](https://hashnode.com/tag/opensource)
-   [#gpu](https://hashnode.com/tag/gpu)

## Written by

[Aman Karki](https://hashnode.com/@amankarki)

@amankarki · India

  
[View profile](https://hashnode.com/@amankarki)

C++/CUDA Engineer · LLM Inference & Systems

C++/CUDA engineer building LLM inference and GPU code from scratch: verbum.cpp, Lattice, RAAG. 2 CUDA PRs merged into llama.cpp. I write about inference, GPU performance, and measuring things honestly.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.