Where did you get that GPT3 has 12288 size token embeddings? I thin that's the internal or output size of the token inside the transformer layers, not in the em

Where did you get that GPT3 has 12288 size token embeddings? I thin that's the internal or output size of the token inside the transformer layers, not in the embedding table.