Moonshot Launches Kimi Linear Model: Processing Long Contexts 2.9 Times Faster

The Moonshot team has launched the Kimi Linear model, achieving a technological breakthrough in the AIGC field. The model uses a hybrid linear attention architecture, improving the speed of processing long contexts by 2.9 times and decoding speed by 6 times. Its performance surpasses the traditional Softmax attention mechanism, showing excellent results particularly in scenarios such as context processing and reinforcement learning.