hashfita — Chronobreaker Project
Bitcoin hash analysis system: RingTransformer neural network on MLX, OpenCL SHA-256 feature extraction, QUIC distributed workers, .NET/C# compute layer.
hashfita — Chronobreaker
A Bitcoin hash analysis system that attempts to predict nonce locations using a custom transformer neural network (RingTransformer) trained via reinforcement on SHA-256 ring features extracted by OpenCL GPU kernels. Targets Apple Silicon (M3 Pro) as primary compute platform.
Repository: /Users/alefita/workdir/hashfita/
Version: v0.1.0
Python: 3.13
Package Manager: uv (pyproject.toml)
Architecture
| Layer | Technology | Purpose |
|---|---|---|
| Model | MLX (mlx.nn) | RingTransformer — 272k params, 4D RoPE |
| Training | MLX autograd + MuonClip | Custom optimizer with Newton-Schulz ortho |
| Feature extraction | PyOpenCL | SHA-256 carrier/ring/trace extraction on GPU |
| Nonce validation | PyOpenCL | Double SHA-256 batch validation |
| Networking | aioquic (QUIC) | Distributed scheduler-worker protocol |
| Workers | .NET/C# + Python IPC | Hardware-adaptive compute (NVIDIA/AMD/Apple) |
Compute Stack
- MLX ≥ 0.31.1 — Primary framework (model, training, autograd, optimizers)
- PyOpenCL ≥ 2024.1 — GPU kernel execution for SHA-256 operations
- NumPy ≥ 1.26 — Bridge for PyOpenCL buffer transfer only
- No PyTorch — Zero imports, zero references
Key Components
| Component | File | Purpose |
|---|---|---|
| RingTransformer | src/chronobreaker/model/transformer.py | Neural network: 17 rings → nonce prediction |
| AttentionBlock4D | src/chronobreaker/model/attention.py | Multi-head attention with 4D RoPE |
| MuonClip | src/chronobreaker/train/muonclip_mlx.py | Newton-Schulz + QK-Clip optimizer |
| ReinforcementTrainer | src/chronobreaker/train/trainer.py | 3-pass training: forward → GPU validation → autograd |
| OpenCL Context | src/chronobreaker/core/opencl_context.py | Kernel loading, buffer management, GPU dispatch |
| Holographic Projector | src/chronobreaker/holographic/projector.py | PHIN Scatter candidate generation |
Related Pages
- ring-transformer — Full architecture deep-dive
- muonclip-optimizer — Optimizer internals
- mlx-opencl-bridge — Buffer transfer patterns
- opencl-kernel-design — Kernel inventory and loading
- gpu-async-pipelines — Async architecture
Status
- ✅ RingTransformer training functional
- ✅ OpenCL feature extraction pipeline
- ✅ QUIC distributed protocol
- ⚠️ 3-copy buffer chain (OpenCL→numpy→MLX) — known bottleneck
- ⚠️ No TDR recovery
- ⚠️ float32 exclusively — bfloat16 opportunity
- 🔜 .NET Silk.NET.OpenCL migration pending