Downloads · 30 days
0
sihc-group5-flagship-ckpt
TongtongLiang/sihc-group5-flagship-ckpt
sihc-group5-flagship-ckpt is a machine learning model from TongtongLiang. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
40 blocks, width 1024, five stages of eight blocks, direct 4×4 subpatch embedding, 16×16 workspace, 32 context tokens from zero-based block index 8. Velocity prediction with DINOv3-L/16 REPA at block 8.
Access
Public
Updated Oct 1, 2026
Repo size
31 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt30.9 GB · 99%
At a glance
- Access
- Public
- Created
- Sep 17, 2026
- Updated
- Oct 1, 2026
- SHA
- e5a8fac0
Model card
From the Hugging Face model README
SiHC-XL: sublayer flagship with REPA
40 blocks, width 1024, five stages of eight blocks, direct 4×4 subpatch embedding, 16×16 workspace, 32 context tokens from zero-based block index 8. Velocity prediction with DINOv3-L/16 REPA at block 8.
Sampling: primary EMA, Heun-50, CFG 2.4 in clean-time interval [0.1, 0.9], balanced 50K ImageNet-256 samples and JiT reference statistics.
This collection retains earlier epoch-400/480 copies. The selected paper checkpoint is linked in the Paper flagship section below.
checkpoints/ contains full resume files. evaluation_50k/ contains existing evaluation records. HANDOFF.md and orphan logs are historical operational records.
Files and configuration
| Checkpoint | Epoch | Format | Architecture / status |
|---|---|---|---|
| checkpoints/step_00500400.pt | 400 | resume | sihc_sublayer_5x8_d1024_b4_ctx32s8 |
| checkpoints/step_00600480.pt | 480 | resume | sihc_sublayer_5x8_d1024_b4_ctx32s8 |
Machine-readable checkpoint catalog records revisions, file sizes, configuration and architecture classification.
Code and reproduction guide · Research archive · Paper flagship XL
The DINO teacher is needed only for REPA training. Follow the model-specific loader; blockwise and sublayerwise weights are different architectures.
Paper flagship
The selected SiHC-XL + REPA checkpoint is hosted by our coauthor: step_00650520.pt. Its recorded primary-EMA Heun-50 evaluation is FID 1.71491, IS 299.5535, CFG 2.4, interval [0.1, 0.9], 50,000 samples. Historical checkpoints above retain their own identities and metrics.
- Created
- Sep 17, 2026
- Updated
- Oct 1, 2026