-
Upload README.md with huggingface_hub
-
Phase 1: pure synthetic KiCad pre-training (5000 samples). CompF1=0.304, exp6 CompF1=0.119
-
Phase 2: Baseline + Synthetic Data, CompF1=0.156 (highest), JointF1=0.011
-
Phase 2: Anti-Overfit + Synthetic Data, CompF1=0.126, JointF1=0.011
-
Upload README.md with huggingface_hub
-
Upload retrained V9-Pure final model weights with LFS tracking
-
Update README with V8-Fixed results
-
V8-Fixed best weights: s1600, easy100 NED=0.7760, easy50 NED=0.8257
-
V5 f32: replace collapsed old model with diverse V5
-
V5: LLM-only LoRA r=8, frozen projector, 90% diversity (no collapse)
-
Upload 2 files
-
initial commit
