Projects

Poorly Optimized Language Model

Institute of Poorly Optimized GPU Code presents its first creation: "Poorly Optimized Language mode". As a first victim of research we've chosen Qwen3 models. Repository will showcase progress of building model using PyTorch and then optimizing its inference with methods like KV-caching, Speculative decoding and so on. After sufficient progress blog post will be created showcasing what was achieved.