tinyengram

Open research codebase reproducing DeepSeek-style Engram memory-injection architecture on Qwen-3 and Stable Diffusion, comparing against LoRA fine-tuning with training experiments and vision extensions.