The compute-heavy phase that processes an inference prompt and builds its KV cache before generation begins.
← All terms