Skip to content
D-CSIL

Model Registry · HELIX family

HELIX-56M

H-E-R stack · HER×4 (12 layers) · d_model 512 · 8 heads · rotation-HSL (post-bakeoff) · 4 experts

STATUS PRIVATEVAL PPL 48.59PROJECT: HELIX
HELIX-56M — TRAINING LOSS (DEMO DATA)60,000 STEPS · MIN 3.95
4.05.26.57.89.10k15k30k45k60k

Benchmarks

MetricValueNotes
WT-103 test PPL (token / word)27.29 / 48.59docs/RESULTS_56M_WT103.md
Matched transformer control (57.36M, same protocol)25.10 / 44.82Colab A100, same data/steps — transformer leads by 0.084 nats (~8–9%)
Tokens/sec (train)8,000–9,000operator-reported
Transformer comparison runColabHELIX_Transformer_Comparison_56m.ipynb on Drive, 2026-06

Known Limitations

  • The controlled, matched-protocol comparison shows HELIX trailing a same-size dense transformer by ~8–9% perplexity (0.084 nats test) — the first measured point of the architecture's O(L)-inference tax, later confirmed larger (15.0%) at 127M.
  • Loss-curve anchor points are approximate (full step log not in registry yet); PPL and throughput are as recorded/reported.
  • Base LM only; no instruction tuning on this card (see the chat-adapt/SFT lineage in the HELIX project page).
  • Not publicly released — served through a custom, Apple Silicon-only inference engine behind an invite-only gate. There is no public checkpoint, API, or download.