WuBu-35M — the AGI's base model

⚠️ DISCLOSURE: THIS IS AN ARTIFICIAL GENERAL INTELLIGENCE (AGI) PROJECT. WuBu-35M is the base model of the WuBuOS AGI brain-cluster system, built by the WaefreBeorn project. It is a from-scratch C11 implementation, trained and grown in our own loop. It is an AI model. It is not sentient, conscious, or self-aware; it is a statistical language model under continuous development toward general intelligence. Use it accordingly — verify its outputs, do not anthropomorphize it, and do not deploy it in high-stakes autonomous roles without human oversight. The WuBuOS doctrine is human-centric: the human is always in the loop, and the AGI yields to human control.

What this is

• The seed. 35,072,768 parameters, 12 layers / 448 dim, GQA 7:1, hybrid 3-local + 1-full attention, 50% partial RoPE, QK-norm, gated attention, bounded SwiGLU, residual selectors, tied embeddings — the WuBu architecture, written from scratch in C11.
• Ours. The C11 implementation (wubu, wubu_train) is original WaefreBeorn work. The training loop is our Muon + AdamW core. The weights on this page are trained in-house by the WaefreBeorn AGI loop (the SFT-trained seed); they grow as the loop continues.
• The growth path. Research repos → tokens → parameters → evaluation → the 5+1 recovery substrate makes every mistake safe. The bigger brother (Qwen3.6-27B) provides coherency while this seed trains; when the seed surpasses the brother, the brother retires.

The WuBu model blueprint (where this is going)

1. Hyperbolic geometry — the Lean-verified Möbius layer (wubu_hyper): ball closure, exp∘log identity, gyroassociativity — the seed thinks on a sphere, not a line.
2. The amoeba — the model GROWS and SHRINKS through the hive (wubu_amoeba): overworked cells split, dead cells apoptosis, every mutation validated and archived (the Darwin Gödel Machine).
3. The hive diagnostic — every measurement (loss, grads, route, oracle score, system vitals) is a cell in the same hive tissue as the model's own cells; the causal walker finds root causes, the 5+1 rolls back per the walker's report.
4. The tensor catalog — one interchange for safetensors/GGUF/.st, live-load by name, streaming export (no conversion waste).
5. The live learning loop — prompt → draft → NVIDIA oracle score → critique → next SFT round: the reward organ is wired to the trainer.

Safety (the WuBuOS doctrine)

1. Human-centric. The human is always in the loop. The AGI yields to human control at every level (the kernel, the trainer, the loop).
2. Mistakes are reversible. The 5+1 recovery substrate snapshots state; every rollback is recorded in the prestige ledger. No black boxes.
3. It is not ready for autonomy. Treat all outputs as drafts. Verify, audit, and keep a human in the loop.

License

Contact