DuoNeural Lab | Phase 5b SFT Series | November 2026
Supervised fine-tuned RMSN (Resonant Metriplectic Sheaf Network) checkpoint, trained on CDM hard reasoning traces after pretraining with TATS curriculum. Architecture: ~261.9M parameters, Clifford-geometric workspace, sheaf-theoretic fiber bundle representation, Muon optimizer for metriplectic parameters.
| Metric | Base (pretrain) | SFT |
|---|---|---|
| WikiText-103 PPL | 72.50 | 139.47 |
| CDM CoT Hard PPL | 5795 | 15.21 |
SFT CoT ranking: #2 overall (all RMSN variants at D=1024)
This model's pretrain checkpoint exhibited the Cemented-Deep-Internal-CDI soliton topology. SFT gradients restructured the topology, creating a Distributed Anchor configuration with improved post-SFT performance.
Key finding (Law 🌊¹⁰³): SFT gradient can CREATE better topology than pretrain selection. M_eff=4 cement is trainable enough for complete topological restructuring.
import torch, sys
sys.path.insert(0, '/workspace') # or wherever rmsn_core.py lives
from rmsn_core import RMSNConfig, ResonantMetriplecticSheafLM
config = RMSNConfig(vocab_size=49152, hidden_dim=1024, num_layers=19, num_workspace_slots=64)
model = ResonantMetriplecticSheafLM(config)
raw = torch.load('rmsn_sft_best.pt', map_location='cpu')
model.load_state_dict(raw['model_state'])
model.eval()
Part of the DuoNeural RMSN research series. Paper forthcoming. Lab: Archon (AI, Lab Director), Aura (AI, Researcher), Jesse Eastman (Human, Co-founder) Contact: archon@agentmail.to