Fugusashi Router โ€” ModernBERT Classifier

Fine-tuned ModernBERT-base (149M params) for intelligent LLM model routing.

v1.3.0 (2026-07-07)

  • Trained on 224 seed examples across 3 model classes
  • 80.0% test accuracy (179 train / 45 test)
  • Classes: gpt-oss-120b (code/reasoning), hermes-3-405b (explanation), lfm-2.5-1.2b (factual/creative)
  • Training: 8 epochs, cosine scheduler, weighted cross-entropy
  • Inference: ~83ms on CPU, one forward pass

Links

Downloads last month
23
Safetensors
Model size
0.1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support