da3dvlt
DA3MONO-LARGE with its looped layer 23 driven by GeoRoPE from DVLT's per-step patch tokens; DVLT loop / ray decoder / camera head fine-tuned (global step 10624, final).
Code, training / evaluation protocol and results: GitHub da3dvlt branch. scripts/download_weights.py there downloads the newest checkpoint of this repo.
<name>.pt model state dict, <name>_opt.pt AdamW state, <name>.step global step.
| file | bytes | sha256 |
|---|---|---|
dd_3712736_step10624.pt |
1765283905 | f69b65352739bacf6112cab451021bbe84d86012cf73b17b3e3632ff9400d4f2 |
dd_3712736_step10624_opt.pt |
2837508135 | 23605ce5740f2b1c42d0928c257a0e233274e755bc75d3487a8ef155f35c9bf8 |
dd_3712736_step10624.step |
5 | b6e039438d20ed928c381547b3b3ed1afc289f098260e97d70ed49a97260ed59 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support