Product · mentioned 6 times
Named project that is a version of the Qwen language model with a computer-like circuit built into its architecture.
Tap a name to go there · tap a line to see why they are linked
Startup Insider· 8 Nov
Philipp Müller says Qwen's architecture contains a mechanism called Gated Delta Net — a memory state the model was not designed to use this way — which was configured to act as a CPU inside the model, with register accesses, arithmetic and jumps, and that Qwen was then taught in post-training to use that part of itself like a tool.