Hakureirm/rwkv7-0.1b-hf

Retire the optional kernels, carry the tokenizer, drop dead config keys

max_position_embeddings 4096: the training context of the source checkpoint (RWKV is recurrent; this only sizes generation defaults)

Sync remote code from rwkv7-hub (canonical generated build)

Shape the triton imports so the remote-code import scanner exempts them (they only execute under the same availability guard)

Ship the implementation as remote code (auto_map): stock transformers loads this repo directly; verified CPU + Apple MPS + CUDA

RWKV-7 0.1B, converted from BlinkDL/rwkv-7-world for the in-tree transformers implementation

initial commit