divagr18/QwenPU

Qwen3.5-0.8B writes a Bend program; the same model type-checks and runs it inside its own weights. No compiler, no interpreter, no tool call.

Qwen3.5-2B with a SIMT GPU inside its weights, 45 compiled layers in front of Qwen’s own, using its Gated DeltaNet layers as registers and memory.

An NES CPU compiled into the weights of a stock Qwen3.5 model

A small proof-of-concept language model (not an LLM) that aims to incorporate latent-space prediction, internal state using recurrent trace units, and byte-by-byte output, built with MLX.

Pure C++ implementation of several models for real-time chatting on your computer (CPU & GPU)

Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results

Run the full 2.78-trillion-parameter Kimi K3 model, DeepSeek V4.1 Flash or GLM-5.3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C inference engine.

NPU Ollama - An Ollama/OpenAI compatible API for Intel OpenVINO compatible computers

Perplexity open source garden for inference technology