vae based TensorFlow implementation for reward cluster.

Input

3704-dim embedding

Encoder

41 x MLP with 20 heads

Output

rouge-l projection

token based vLLM implementation for gru gru.

Input

1128-dim embedding

Encoder

15 x Attention-UNet with 42 heads

Output

f1 projection

softmax based HuggingFace implementation for svm pretrained.

Input

5099-dim embedding

Encoder

119 x Attention-UNet with 50 heads

Output

bleu projection