Downloads · 30 days
0
bdcube-block-diffusion-original-4xh100-run
Alexander1211/bdcube-block-diffusion-original-4xh100-run
bdcube-block-diffusion-original-4xh100-run is a machine learning model from Alexander1211. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Portable training bundle for the successful BDCube 4xH100 Block Diffusion original run.
Access
Public
Updated Apr 16, 2026
Repo size
37.1 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors35.9 GB · 96%
At a glance
- Access
- Public
- Created
- Apr 16, 2026
- Updated
- Apr 16, 2026
- SHA
- a9f847ac
Model card
From the Hugging Face model README
bdcube-block-diffusion-original-4xh100-run
Portable training bundle for the successful BDCube 4xH100 Block Diffusion original run.
Included assets
- Full checkpoint set.
- Train/val/sample/geometry logs.
- Original run manifests and resume inputs.
- Step-30000 trainer resume state.
- Geometry eval outputs and sampled generations.
- Resolved configs, dataset summary, and run metadata.
- Portable resume manifests in
portable_manifests/.
External repos
- Pair dataset repo:
Alexander1211/TRELLIS-500K_train_pairs_clean_qwen35_v5_native - Eval assets repo:
Alexander1211/bdcube-eval-assets-4xh100
Notes
- The portable resume manifests rewrite
pair_pathrelative to the pair dataset repo root. - New-server training code should not require local raw
mesh_pathexistence before usingpair_path. logs/latest_fixed20_metrics_summary.jsonis preserved for history butlogs/geometry_eval.jsonlis the authoritative metric log.
- Created
- Apr 16, 2026
- Updated
- Apr 16, 2026