Skip to content
D-CSIL

Model Registry · HELIX family

HELIX-200M

Hybrid H-E-R stack · 28 layers · d_model 1024 · HarmonicStateLayer + ElasticAttentionGate + ResonantExpertRouting (MoE top-2 of 4)

STATUS TRAININGVAL PPL 39.4PROJECT: HELIX
HELIX-200M — TRAINING LOSS (DEMO DATA)120,140 STEPS · MIN 2.82
2.84.45.97.48.90k30k60k90k120k

Benchmarks

MetricValueNotes
Train PPL @ step 120,14016.85run log
Val PPL @ step 120k39.4held-out validation
Tokens/sec (train)~2,900sustained, BF16

Known Limitations

  • −Mid-training checkpoint; every number here will move.
  • −External baselines (GPT-2 small 37.5 PPL target) not yet crossed at this step.

Experiments Involving This Model