A stage with two or more layers keeps one flat copy and unpacks at most two layers. A sliding window is a softmax mask. JSON and the position seed change only the pick. DeepGEMM's 128-channel scale stays out of the GEMM. The four docs list what was added and what was left out.
Layer-sharded peer-to-peer inference. Each device downloads only its own
layers from the Hub via safetensors byte ranges, so no device holds the
whole model -- this pools memory, which DaisyChain-Train explicitly cannot.
Loads any Llama-style or GPT-2-style .safetensors repo. Verified INT8 units,
WGSL kernels and their exact gates carried over from DaisyChain-Web unchanged.
47 checks: bit-exact split equivalence (both families, with mutation checks),
wire protocol refusals, ring routing, safetensors/dtype/BPE loader.
