Skip to content
OpenCatalogcurated by FLOSSK
AI & Machine Learning

RWKV

RNN-meets-transformer linear-attention LM architecture running with O(n) memory—unique open line for long-context and embedded inference.

Why it is included

Non-transformer open architecture with active community ports (CUDA, Metal, Web).

Best for

Experimenters wanting recurrent-style LLMs without full KV cache growth.

Strengths

  • Linear time
  • Embedded friendly
  • Multiple runtimes

Limitations

  • Ecosystem smaller than Llama for tooling

Good alternatives

Transformer LMs (Llama-class) · Linear-attention research stacks

Related tools