gpt_neox.layers.8/cfg.json

{"expansion_factor": 64, "normalize_decoder": true, "num_latents": 0, "k": 32, "multi_topk": false, "d_in": 768}