1DeepSeek-V3: Achieving Efficient LLM Scaling with 2,048 GPUs (opens in new tab)(arxiv.org)7qtwhat10mo ago1