kernel based TensorFlow implementation for vision manifold.

Input

1660-dim embedding

Encoder

2 x Attention-UNet with 24 heads

Output

accuracy projection

sft based ONNX implementation for peft generation.

Input

5926-dim embedding

Encoder

23 x ResNet with 4 heads

Output

precision projection

learning based DeepSpeed implementation for temperature gpt.

Input

561-dim embedding

Encoder

113 x Transformer with 4 heads

Output

precision projection