Function ffn

gpt2.py:24–31

(x, c_fc, c_proj)

def ffn(x, c_fc, c_proj):  # [n_seq, n_embd] -> [n_seq, n_embd]
    # project up
    a = gelu(linear(x, **c_fc))  # [n_seq, n_embd] -> [n_seq, 4*n_embd]

    # project back down
    x = linear(a, **c_proj)  # [n_seq, 4*n_embd] -> [n_seq, n_embd]

    return x


def attention(q, k, v, mask):  # [n_q, d_k], [n_k, d_k], [n_k, d_v], [n_q, n_k] -> [n_q, d_v]