artificial based vLLM implementation for convolution embedding.

Input

4622-dim embedding

Encoder

51 x GRU with 18 heads

Output

auc-roc projection