Module phi3 

Microsoft Phi-3 model implementation

See Phi model details at:

• Phi-3 Model

The Phi series are decoder-only transformers designed for code and language tasks. Key characteristics:

• Decoder-only transformer architecture
• RoPE embeddings
• Layer normalization
• QK normalization
• Mixed activation functions
• Improved context window handling

References:

• Hugging Face Implementation
• Alternative Implementation

Structs

Config

Model

RotaryEmbedding