Mostik.ai – latent communication between AI modelsmostik.ai 2 pointszX41ZdbW5 days ago1 commentSaveHideCopy link On HNComments−arpperzhao5dConjecture: Does GLM only handle the prefill stage, compress its output hidden vectors (trained?), and then send them to Qwen for decoding?
Comments
Conjecture: Does GLM only handle the prefill stage, compress its output hidden vectors (trained?), and then send them to Qwen for decoding?