wirkung — Holographic Causal Graph Machine (HCGM) & sGPU Runtime

Sovereign zero-dependency native execution engine for the Origami model family (Origami-9B, Origami-27B, Origami-31B, Origami-H3, Origami-78B) and the sGPU (cpu_cuda) CUDA-on-CPU + M-Rail operator runtime.

• Author: David Tom Foss (Wirkung Inc. · david@wirku.ng)
• Zero Framework Dependencies: Runs directly on CPU servers and Apple Silicon with numpy as the sole Python dependency — zero PyTorch, zero CUDA toolkit, zero vLLM, zero transformers, zero external tokenizer libraries.
• Capabilities:
  • Zero-unpack .o3f / .hpack SIMD contractions and .o1f1 lossless round-1/2/3 unfolding
  • Constant-memory WY Gated DeltaNet + pinned-sink Ward carriers and exact Cauchy-Schwarz vocabulary pruning
  • Continuous M-Rail (X = X_known + R) trajectory assimilation and synchronous .mrail v3 fleet merging
  • Drop-in OpenAI & DeepSeek DSH HTTP/SSE + XML/JSON tool-calling server (wkng serve)
  • Standalone sGPU (sgpu / cpuda / rail) CUDA Driver/Runtime/cuBLAS/Math execution on CPU servers

Quickstart

pip install wirkung
# or: pip install wkng

wkng doctor
sgpu doctor
wkng pull wirkung/Origami-9B -o ./Origami-9B
wkng run ./Origami-9B -p "Explain Walsh-Hadamard incoherence in two sentences." -n 96 --chat
wkng serve ./Origami-9B --port 8080

License

Licensed under the Wirkung Model and Runtime License Agreement (LICENSE). Free for non-commercial research, personal evaluation, and academic use in compiled Object form; commercial or production deployment of Origami models, wkng, or sgpu requires a commercial license from Wirkung Inc. (contact@wirku.ng).