An unnecessarily tiny implementation of GPT-2 in NumPy.

GPT in a QR Code ; The actual most atomic way to train and inference a GPT in pure, dependency-free JS/Python.

Minimal GPT (~350 lines with a simple task to test it)

Like picoGPT but for BERT.

No description available

picoGPT in Rust

J port of "An unnecessarily tiny implementation of GPT-2 in NumPy."

No description available

No description available

A pico-level implementation of Generative Pre-Trained Transformer(GPT) from scratch.

Try out here. Initially, you need to download the model weights. Once downloaded, the weights can be loaded from local storage on next visits.

A fun little play with transformers and text generation

A simple small GPT model for educative purpose that can be trained on CPU

From-scratch GPT in pure Julia. Every gradient by hand. No AD, no GPU, no magic.

No description available

Small scale GPT-like full transformer stack, runnable on any PC

Smaller than nano! :)

A hands-on fork of NanoGPT with FlashAttention-2 CUDA kernels, INT8/INT4 GPTQ quantization, paged KV-cache reuse, and continuous batching, turning a tiny Shakespeare model into a full-speed GPU LLM inference demo.

Rewrite of nanoGPT

60 lines code for a simple GPT

No description available

A tiny implementation of BERT in NumPy, inspired by jaymody/picoGPT

PicoGPT and others

No description available

No description available

GPT2 in numpy. A worse version of the original picogpt.

A tiny llm for learning training process

GPT in a QR Code ; The actual most atomic way to train and inference a GPT in pure, dependency-free JS/Python.

No description available

No description available