Package smile.llm.llama
Class Attention
java.lang.Object
smile.llm.llama.Attention
Multi-head attention. It caches key and value information, applying rotary
embeddings, and performing linear transformations.
-
Constructor Summary
-
Method Summary
-
Constructor Details
-
Attention
Constructor.- Parameters:
args
- the model configuration parameters.
-
-
Method Details
-
forward
Forward pass through the attention module.- Parameters:
x
- the input tensor.startPos
- the starting position for attention caching.cis
- the precomputed frequency tensor.mask
- the attention mask tensor.- Returns:
- the output tensor.
-