Module training 

RWKV7 training (byte-level) implemented in Rust.

This module trains a RWKV7 model that is weight-compatible with the SIMD inference path in crate::rwkv7::Model by exporting the same safetensors keys & shapes.

Design goals:

• Correctness first: math matches src/rwkv7/model.rs
• GPU acceleration via libtorch (tch) when available
• Small-model friendliness (<10M params) with stable defaults

Structs

TrainConfig

TrainReport

Functions

cuda_wkv_available

Check if CUDA WKV kernel is available

train_enwik8