llama_split_mode.md

Here's a concise overview:

1. Enumeration LlamaSplitMode:
   • Purpose: This enum defines the modes for splitting a Llama model across multiple GPUs, dictating how the model's computational load is distributed.
   • Values:
     • none: Indicates usage of a single GPU for the model.
     • layer: Specifies that layers and Key-Value (KV) pairs are split across multiple GPUs.
     • row: Designates that rows of the model are distributed across different GPUs.