111 kB

Ctrl+K

1 contributor

History: 31 commits

patdev

Use FP32 master LoRA weights for stable L4 distillation

cf3b558 verified about 13 hours ago

__pycache__
Add v7 distillation FP8 sparsity DINOv3 multiview symmetry MoE stack about 22 hours ago
dinov3_bridge.py

2.21 kB
Add v7 distillation FP8 sparsity DINOv3 multiview symmetry MoE stack about 22 hours ago
distill_core.py

3.84 kB
Fix sparse batch-size handling in v7 few-step distillation about 21 hours ago
distill_dinov3.py

2.44 kB
Add v7 distillation FP8 sparsity DINOv3 multiview symmetry MoE stack about 22 hours ago
distill_flow.py

8.18 kB
Harden v7 distillation with BF16 and non-finite guards about 13 hours ago
fastgen_ss_bridge.py

2.84 kB
Add exact FastGen rectified-flow bridge for AniGen SS about 22 hours ago
fp8_sparse.py

3 kB
Fix v7 ModelOpt FP8 and AutoQuantize API for 2026 release about 22 hours ago
geometry_moe.py

2.53 kB
Fix Geometry MoE top-k weight indexing about 22 hours ago
lora.py

4.1 kB
Use FP32 master LoRA weights for stable L4 distillation about 13 hours ago
moe_inject.py

3.34 kB
Place injected Geometry MoE experts on AniGen device and dtype about 21 hours ago
moe_train.py

3.67 kB
Add Geometry-Inductive MoE shared-teacher training loop about 21 hours ago
qat_recovery.py

4.26 kB
Add selective FP8 QAT shared-teacher recovery about 21 hours ago
sparse24_recovery.py

5.29 kB
Add exact 2:4 shared-teacher recovery with final re-prune about 21 hours ago
symmetry_loss.py

1.5 kB
Integrate flow equivariance symmetry loss for v7 distillation about 21 hours ago
teacher_hybrid.py

2.42 kB
Add v7 distillation FP8 sparsity DINOv3 multiview symmetry MoE stack about 22 hours ago