berkus/ANE - ANE

Commit Graph

Author	SHA1	Message	Date
maderix	cb474e1537	Add dynamic weight training pipeline — 110ms/step without recompilation Dynamic weight pipeline that eliminates the ~3.7s recompile-every-10-steps bottleneck. Weights are passed via IOSurface spatial dimension instead of baked as constants, so kernels compile once at startup (345ms) and run indefinitely without exec() restart. Key components: - training_dynamic/ — full pipeline (config, IO, MIL generators, train loop) - 9 dynamic kernels shared across all 12 layers - Vocab compaction 32K→9.2K for faster classifier - Vectorized cross-entropy with vDSP/NEON - Adam optimizer with gradient clipping + cosine LR schedule - Checkpoint save/resume - test_dynamic_matmul.m — validates dynamic weight matmul vs cblas - test_weight_patch.m — tests weight update via IOSurface - dashboard.py — updated with --dynamic flag for v2 pipeline support, improved step regex parsing, --scratch/--lr/--accum CLI args Performance: 110ms/step steady-state (no recompile overhead) ane_fwd=21 ane_bwd=28 io_fwd=12 io_bwd=15 silu=10 cls=13 rms=5 ms	2026-03-03 04:34:55 -08:00
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

Author

SHA1

Message

Date

maderix

cb474e1537

Add dynamic weight training pipeline — 110ms/step without recompilation

Dynamic weight pipeline that eliminates the ~3.7s recompile-every-10-steps
bottleneck. Weights are passed via IOSurface spatial dimension instead of
baked as constants, so kernels compile once at startup (345ms) and run
indefinitely without exec() restart.

Key components:
- training_dynamic/ — full pipeline (config, IO, MIL generators, train loop)
  - 9 dynamic kernels shared across all 12 layers
  - Vocab compaction 32K→9.2K for faster classifier
  - Vectorized cross-entropy with vDSP/NEON
  - Adam optimizer with gradient clipping + cosine LR schedule
  - Checkpoint save/resume

- test_dynamic_matmul.m — validates dynamic weight matmul vs cblas
- test_weight_patch.m — tests weight update via IOSurface

- dashboard.py — updated with --dynamic flag for v2 pipeline support,
  improved step regex parsing, --scratch/--lr/--accum CLI args

Performance: 110ms/step steady-state (no recompile overhead)
  ane_fwd=21 ane_bwd=28 io_fwd=12 io_bwd=15 silu=10 cls=13 rms=5 ms

2026-03-03 04:34:55 -08:00

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

2 Commits