wirkung/Origami-78B

Paper · DOI · Runtime (PyPI) · Wirkung · Base weights · Contact · License

Origami-78B is a 78.10B-parameter Holographic Causal Graph Machine (HCGM) Preview compiled in closed form from Aleph-Alpha/Kolibri-1 with zero retraining, zero fine-tuning, and zero gradient steps. It folds the 156.21 GB source checkpoint into 26.42 GB (5.91x compression), executes directly from NVMe/SSD storage via OS mmap with 0 GB GPU VRAM and 0 GB dedicated weight-RAM, and keeps live state strictly constant (O(1)) across arbitrary horizons on CPU, NPU, sGPU, Metal, and standalone PTX.

Origami-78B preserves the full Aleph-Alpha/Kolibri-1 architecture (50 decoder layers with 19,200 routed experts (384 experts per layer, 3.46B active parameters per token), 10 global-attention layers, and 40 sliding-window layers) while turning a passive weight checkpoint into a self-contained binary file executed under one cross-silicon contract. It succeeds our first-generation constant-state releases published under huggingface.co/tfwnotops by removing vLLM, MLX, and Transformers runtime dependencies altogether. This repository is a Preview release: structural byte footprints, fold ratios, and constant-memory invariants below are fixed properties of the compiled artifact, while hardware-dependent throughput figures are omitted from the card and documented in the Origami Family Paper.

pip install -U wirkung && wkng pull wirkung/Origami-78B -o ./Origami-78B
wkng verify ./Origami-78B --full && wkng unfold ./Origami-78B -o ./Origami-78B-unfolded

Source checkpoint against Origami-78B

Property	Source (Aleph-Alpha/Kolibri-1)	Origami-78B (HCGM Preview)
Parameters	78.10B	78.10B (identical graph topology)
Retraining / fine-tuning steps	Pretrained baseline	0 (closed-form compilation)
Checkpoint footprint on disk	156.21 GB BF16	26.42 GB (5.91x smaller; 54 SHA-256 bit-lossless round-3 .o1f1 shards (58,703 tensors))
Required GPU VRAM	Multi-tens of GB HBM	0 GB (runs on plain CPU / NPU / sGPU or optional GPU)
Dedicated weight-RAM	Pinned in host / device RAM	0 GB (read-only OS mmap directly from SSD/NVMe)
Live session / horizon state	Linearly growing O(T) KV cache (86.08 GB at 1M tokens; out-of-memory at 5M)	Constant 0.183 GB/request O(1) carrier (0.62 GB card state verified out to 5.24M tokens)
Operator & readout path	Dense per-step matrix scan	Depth-stratified 6 / 3 / 4 MoE expert profile (-33% expert reads on text; uniform k = 6 on code)
Runtime dependencies	PyTorch / vLLM / MLX / CUDA	wirkung + numpy (standalone C11, NPU, sGPU, Metal, PTX)

Figure 1. (a) Physical byte scale contrasting the 156.21 GB BF16 Aleph-Alpha/Kolibri-1 source model against the 26.42 GB Origami-78B .o1f1 bundle (0 GB GPU VRAM required on CPU/NVMe servers). (b) Live conversation state across 4k to 1M+ tokens: linear O(T) KV growth against the flat 0.183 GB Origami-78B constant-state carrier.

Zero-retraining compilation and constant-state architecture

Every Origami model is produced by the same four-part compilation pipeline without gradient descent:

1. The file is the machine (disk-resident mmap execution). An HCGM model is an immutable binary artifact on disk. The operating system maps operator pages read-only on demand from SSD or NVMe storage, requiring neither GPU VRAM nor weight-deserialization buffers in host RAM. Under our v3-exakt dyadic integer reduction contract, CPU (AVX2 / NEON), general NPUs, embedded sGPU (cpu_cuda), Apple Metal, and driver-API PTX evaluate identical canonical reductions rather than drifting floating-point kernel approximations.
2. Two-stage spectral and SHA-256 bit-lossless .o1f1 folding (156.21 GB -> 26.42 GB, 5.91x without retraining). Starting from the 78.10B Aleph-Alpha/Kolibri-1 checkpoint (156.21 GB BF16; 80.83 GB native FP8/BF16 resident footprint) with zero gradient updates, the 57,600 routed expert matrices are transformed into odd-symmetric 3.5-bit block-128 Walsh spectral banks (38.04 GB intermediate layout) and then folded without loss in round r3 into 54 .o1f1 shards totaling 26.42 GB. Every one of the 58,703 tensors unfolds with SHA-256 bit-exactness.
3. Constant 0.183 GB O(1) ckv3 carrier synthesized via Wirkung + depth-stratified routing. Derived via Wirkung, our internal exact structural-law and closed-form operator synthesis engine, the 10 global-attention layers combine an exact 4,096-token band, pinned Sink Token 0, and zero-bit per-head key-mean subtraction q^⊤(k − μ) + q^⊤μ, bounding live state independent of sequence length T while depth-stratified 6 / 3 / 4 expert routing cuts routed MoE weight traffic by 33% on text without modifying router weights.
4. Certified readout and online trajectory decomposition (X = X_known + R). Rather than re-evaluating every parameter blindly for isolated single tokens, HCGM decomposes live hidden trajectories into known low-rank structure X_known (cosine trajectory banks, multi-order suffix tables, and rank-1 Sherman-Morrison-Woodbury updates) plus novel residual R, verifying multiple candidate steps in one weight pass with zero greedy output drift.

Figure 2. (a) Constant-memory ckv3 attention read over token coordinate j: pinned sink s₀, indexed centered slots q^⊤(k − μ) + q^⊤μ with subtracted carrier overlap, and exact band W = 4096. (b) Depth-tiered MoE expert profile across 50 layers: 6 / 3 / 4 text routing (-33% expert reads) against uniform k = 6 for code.

Shared trajectory assimilation (M-Rail), GPU-less server fleets and multi-model routing (HMoE)

Beyond single-node execution, Origami-78B attaches natively to Shared M-Rail (.mrail v3), a content-addressed associative trajectory manifold shared across processes or across an enterprise network:

• Warm up once on one GPU, run permanently on disk-only CPU/NPU servers. Because HCGM models require only SSD storage and megabytes of live state, an organization can run a single GPU (or a short temporary cloud GPU lease) until its recurring domain workflows are crystallized into a shared .mrail file, then shut the GPU down and serve frontier models across commodity CPU/NPU VMs and CPU-only servers. Known trajectories execute via Track copy morphisms or resume in O(1) space from Station (ANCHOR + DELTA) snapshots; whenever any server encounters an unseen task (R ≠ 0), computing and committing that residual once makes the solution immediately available to every other node on the rail.
• Heavy Mixture of Experts (HMoE). Because idle disk-mapped models consume zero weight-RAM, M-Rail routes across complete compiled Origami models (Origami-9B, Origami-27B, Origami-31B, Origami-78B, Origami-H3), zero-recompile static-graph NPU specialist pools, and sub-2 KB exact algebraic group organs on a synchronous station bus (cstar) under deterministic check gates.

Figure 3. (a) Shared M-Rail (.mrail v3) fleet architecture: optional temporary GPU warmup feeds a permanent zero-GPU CPU/NPU/sGPU server fleet where any novel residual +R solved once commits back to all nodes alongside HMoE routing on cstar. (b) Total resident RAM/VRAM across N = 1..32 concurrent 32k-token sessions: pinned weights plus N · KV against shared read-only mmap plus N · O(1) constant state.

How to run

Requires Python 3.10+ and the wirkung runtime package (pip install -U wirkung), which installs the wkng, wirkung, sgpu, and rail CLI entrypoints.

pip install -U wirkung
wkng pull wirkung/Origami-78B -o ./Origami-78B
wkng info ./Origami-78B
wkng verify ./Origami-78B --full
wkng unfold ./Origami-78B -o ./Origami-78B-unfolded

from wirkung import Engine

eng = Engine("./Origami-78B")
print(eng.info())

Files

• layer-000.o1f1 to layer-049.o1f1, fixed-00001.o1f1 to fixed-00004.o1f1: 54 round-3 lossless .o1f1 shards folding all 58,703 tensors (26.42 GB total bundle)
• bundle.json, wirkung_kolibri_recipe.json: shard manifest, SHA-256 digests, and runtime contract verified by wkng verify
• layout/: original tensor headers, per-layer experts.json, and deterministic unfold index for wkng unfold
• config.json, generation_config.json, tokenizer.json, tokenizer_config.json: model and tokenizer metadata
• paper/Wirkung_2026_Origami_HCGM.pdf: the Origami Model Family technical report
• assets/: banner.png, architecture.png, mechanism.png, mrail.png
• LICENSE, LICENSE-Kolibri-1, NOTICE: Wirkung Model and Runtime License 1.0, Apache-2.0 text for Kolibri-1, and attribution notice

Inquiries and collaboration

We welcome inquiries regarding custom zero-retraining HCGM model compilation, closed-form formula and exact algorithm synthesis via Wirkung, Shared M-Rail / zero-GPU enterprise fleet deployments, and research collaboration:

• Email: contact@wirku.ng
• X (Twitter): https://x.com/FossDT
• Web: https://wirku.ng

Papers

• The Origami Model Family: Compiling Pretrained Foundation Models into Holographic Causal Graph Machines. David Tom Foss, 2026. DOI 10.13140/RG.2.2.31826.72647. Included directly in this repository at paper/Wirkung_2026_Origami_HCGM.pdf.
• Holographic Causal Graph Machines: Source-Free Constant-State Execution of Pretrained Transformers. David Tom Foss, 2026.
• Crystallized Attention: Constant-Memory Exact Store and Exponential-Mode Carriers. David Tom Foss, 2026.
• Intelligence Is Not Made of Scarcity: Constant-Memory Causal Execution and Associative M-Rail. David Tom Foss, 2026.

Citation

@techreport{foss2026origami,
  title       = {The Origami Model Family: Compiling Pretrained Foundation Models into Holographic Causal Graph Machines},
  author      = {Foss, David Tom},
  institution = {Wirkung Inc.},
  year        = {2026},
  month       = nov,
  doi         = {10.13140/RG.2.2.31826.72647},
  url         = {https://doi.org/10.13140/RG.2.2.31826.72647}
}

License

Released under the Wirkung Model and Runtime License 1.0. Compiled in closed form without training steps from Apache-2.0 (Aleph-Alpha/Kolibri-1, included as LICENSE-Kolibri-1); full attribution and closed-form transformation notices are recorded in NOTICE.