Module deepseek 

DeepSeek Model V2 - Clean implementation using solid abstractions

This implements the DeepSeek/DeepSeek-V2 architecture including:

• DeepSeek-7B, DeepSeek-67B, DeepSeek-Coder, DeepSeek-Math
• Uses unified Tensor type from tensor_core
• Supports MoE (Mixture of Experts) architecture

Structs

DeepSeekAttention

DeepSeek attention mechanism

DeepSeekConfig

DeepSeekLayer

DeepSeek transformer layer

DeepSeekMLP

DeepSeek MLP (feed-forward network)

DeepSeekModelV2

Main DeepSeek model implementation