compute_qkv: query [batch, length_q, channels] memory [batch, length_m,...
In ifrit98/transformR: An R Implementation of Transformer Model with Self Attention

query [batch, length_q, channels] memory [batch, length_m, channels] (optional, usually RNN hidden states) return [batch, length, depth] (q, k ,v) tensors

compute_qkv(
  query,
  memory = NULL,
  key_depth = 64L,
  value_depth = 64L,
  q_filter_width = 1L,
  kv_filter_width = 1L,
  q_padding = "valid",
  kv_padding = "valid",
  vars_3d_num_heads = 0L
)