nanoeuler
A GPT-2-class LLM built from scratch in C/CUDA without any external ML libraries.
github.comBuilt with
UnknownBuild evidence
Strong
This is a well-documented GitHub repository featuring complete source code for a custom training pipeline, including C and CUDA kernels, verified by a full-model gradient check.
Shipped
2h agonanoeuler is a research-oriented LLM implementation written entirely in C and CUDA. It features a hand-written backpropagation engine, byte-level BPE tokenizer, and a custom FlashAttention implementation, supporting both pretraining and supervised fine-tuning pipelines.
Timeline
Teaser
Video
Playable
Product
Loading…


