A Proof of Concept for the LEMA (Layer-wise Efficient Memory Abstraction) framework. Enables stable fine-tuning of Llama-2-7B on consumer-grade hardware (16GB VRAM) through layer-wise weight streaming and triple-buffer memory virtualization.
machine-learning deep-learning pytorch kaggle memory-efficiency fine-tuning llm llama2 ai-infrastructure low-resource-ai vram-optimization low-resource-computing lema lema-framework
-
Updated
Feb 18, 2026 - Jupyter Notebook