LiquidAI's LFM2.5 is a hybrid: most layers mix with a short gated convolution and only a few carry full attention, so the state a token needs is small and constant where the convolution runs. These are those checkpoints in the CMF container — a single memory-mapped file read by cortiq, a Rust binary with no ML framework under it. GPU via Vulkan/Metal/DX12 with a CPU fallback; NVIDIA, AMD, Intel and Apple silicon read the same file.

All three are 4-bit tiled with ladder scales (q4tp), quantized straight from the bf16 checkpoints.

Steady-state decode, cortiq bench --core, single stream, cortiq 0.5.99+ (the whole-token graph learned this family's short-conv mixer and its sigmoid-routed MoE in 0.5.98/0.5.99 — earlier versions decode it an order of magnitude slower).

On a discrete card the whole token runs as one submitted graph, conv ring and expert routing included; the MoE's greedy output is token-identical to the CPU path. On Apple silicon the engine measures both arms at startup and picks; for the 2.6B the host arm wins there, and CMF_GPU=0 pins it.

Speaks the OpenAI API, so anything that talks to OpenAI talks to it:

Each .cmf ships a .sha256 beside it.