DEEPSEEK PAPER

Summary from the TLDR AI newsletter

One of the best coding models of last year was the DeepSeek LLM. It got close to GPT-3.5 on many benchmarks (even if it is likely 3x the size). Information about model training, token counts, model architecture, and more has been released in a technical report.