Andy Huang
|
e113fae683
|
feat: implement ANE SDK for general-purpose neural engine development
- Implement modular ANE-MIL layer library (Linear, Conv2D, Softmax, LayerNorm, etc.)
- Add Sequential model container with automated activation surface chaining (ping-ponging)
- Implement optimized 'Weights-as-Tensors' pattern across all SDK layers for zero-recompile weight updates
- Add comprehensive automated regression testing suite (regression_test.py)
- Standardize verification for legacy Transformer training and new modular SDK components
- Update README.md and roadmap to reflect SDK capabilities and usage instructions
- Refactor hardcoded paths and unify checkpoint naming conventions for stability
|
2026-03-03 15:35:55 +11:00 |
maderix
|
4d67db1bdb
|
stories110M: 12-layer ANE training with dashboard, 107ms/step
- Scale to full stories110M (109M params, 12 layers) with real TinyStories data
- vDSP-vectorized cross-entropy (110ms→14ms), NEON fp16 IO, async dW
- TUI dashboard: loss curve, ANE/CPU power, CPU/memory graphs, text generation
- Split into modular headers: config, io, mil, cpu_ops
|
2026-03-01 03:14:39 -08:00 |