cad-models-evaluated

IN premise — summaries/2026/08/24/shi-2024-context-aware-decoding-s4-results.md

Created 2026-08-25T02:58:35+00:00

CAD is evaluated on OPT (13B/30B), GPT-Neo (2.7B/20B), LLaMA (13B/30B), and FLAN-T5 (XL 3B / XXL 11B), spanning decoder-only and encoder-decoder architectures.

Summary

CAD has been tested across a wide range of language models, from small (2.7B parameters) to large (30B), and across both major architectural families (decoder-only and encoder-decoder). This breadth matters because it means the results aren't an artifact of one particular model design or scale, so the findings generalize across the current landscape of open models.