Gpu-Memory on Noureddine RAMDIhttps://ramdi.fr/tags/gpu-memory/Recent content in Gpu-Memory on Noureddine RAMDIHugoenSat, 23 May 2026 20:41:27 +0000MegaTrain: RAM-centric training architecture for 100B+ parameter LLMs on a single GPUhttps://ramdi.fr/github-stars/megatrain-ram-centric-training-architecture-for-100b-parameter-llms-on-a-single-gpu/Mon, 04 May 2026 10:23:01 +0000https://ramdi.fr/github-stars/megatrain-ram-centric-training-architecture-for-100b-parameter-llms-on-a-single-gpu/MegaTrain enables training 100B+ parameter LLMs on a single GPU by offloading all parameters to CPU RAM and streaming layers to GPU. Supports HuggingFace models and multi-GPU data parallelism without NCCL.