Lightweight Llama3 Inference Engine – CUDA C
Hey, recently I took inspiration from llama.cpp, ollama, and many other similar tools that enable inference of LLMs locally, and I just finished building a Llama inference engine for the 8B model in CUDA C. I recently wanted to explore my newly founded interest in CUDA programming and my passion for machine learning. This project only makes use of the native CUDA runtime api and cuda_fp16. The inference takes place in fp16, so it requires around 17-18GB of VRAM (~16GB for model params and some more for intermediary caches). It doesn’t use cuBLAS or any similar libraries since I wanted to be…
In plain words
Lightweight Llama3 Inference Engine is a CUDA C implementation for running the Llama 8B model locally on GPUs with at least 17-18GB of VRAM. It uses only the native CUDA runtime API and half-precision floating point, without relying on optimized libraries like cuBLAS. Designed for developers interested in low-level GPU programming and machine learning, it reads model files from HuggingFace in safetensor format. While less optimized than existing inference engines, it prioritizes educational value and direct exposure to CUDA fundamentals.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
Hey, recently I took inspiration from llama.cpp, ollama, and many other similar tools that enable inference of LLMs locally, and I just finished building a Llama inference engine for the 8B model in CUDA C. I recently wanted to explore my newly founded interest in CUDA programming and my passion for machine learning. This project only makes use of the native CUDA runtime api and cuda_fp16. The inference takes place in fp16, so it requires around 17-18GB of VRAM (~16GB for model params and some more for intermediary caches). It doesn’t use cuBLAS or any similar libraries since I wanted to be exposed to the least amount of abstraction. Hence, it isn’t as optimized as a cuBLAS implementation or other inference engines like the ones that inspired the project. ## *A brief overview of the implementation* I used CUDA C. It reads a .safetensor file of the model that you can pull from HuggingFace. The actual kernels are fairly straightforward for normalizations, skip connections, RoPE, and activation functions (SiLU). For GEMM, I got as far as implementing tiled matrix multiplication with vectorized retrieval for each thread. The GEMM kernel is also written in such a way that the second matrix is not required to be pre-transposed while still achieving coalesced memory access to HBM. Feel free to have a look at the project repo and try it out if you’re interested. If you like what you see, feel free to star the repo too! I highly appreciate any feedback, good or constructive.
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 16d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, January 2025
the whole month →- IM
Hello! I'm Byran. I spent the past ~6 months engineering a laptop from scratch. It's fully open-source on GH at: https://github.com/Hello9999901/laptop
Dev tools · 2025 · byran.ee
- TITetris in a PDF▲1,289
I realized that the PDF engines of modern desktop browsers (PDFium and PDF.js) support JavaScript with enough I/O primitives to make a basic game like Tetris. It was a bit tricky to find a union of features that work in both engines, but in the end it turns out that showing/hiding annotation "fields" works well to make monochrome pixels, and keyboard input can be achieved by typing in a text input box. All in all it's quite janky but a nice reminder of how general purpose PDF scripting can be. The linked PDF is all ASCII so you can just open it in a text editor, or have a look at…
Life & fun · 2025 · th0mas.nl
Create lifelike, personalized AI avatars from text prompts
AI · 2025 · jogg.ai
