This AI Research Introduces Flash-Decoding: A New Artificial Intelligence Approach Based on FlashAttention to Make Long-Context LLM Inference Up to 8x Faster