Instructions to use World-Snapshot/WSM_v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use World-Snapshot/WSM_v1 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("World-Snapshot/WSM_v1", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("World-Snapshot/WSM_v1", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]WSM v1 TunnelingNets
This repository stores the official WSM v1 TunnelingNet training runs in three release tiers.
- Direct inference files. Recommended checkpoints, resolved run contracts, and active inference sidecars retain their original paths. These files can be used without unpacking an archive.
- Continued-training materials. Trainer states, final and intermediate checkpoints, statistics, logs, curves, and other retained training records are stored in 7 deterministic, uncompressed TAR part(s), each below 15 GiB.
- Excluded artifacts. Redundant per-step visualizations and local runtime metadata are not released.
To download only the directly usable inference tier:
hf download World-Snapshot/WSM_v1 --repo-type model --exclude "continued_training_materials/*" --local-dir WSM_v1
To obtain the complete release and restore the continued-training material:
hf download World-Snapshot/WSM_v1 --repo-type model --local-dir WSM_v1
python WSM_v1/restore_wsm_v1.py --parts-dir WSM_v1/continued_training_materials --output WSM_v1
The restored training layout is unchanged:
WSM_v1/
βββ train_log/
βββ train_log_sparse/
βββ train_log_custom/
Release contents
- Dense runs: 1148
- Sparse runs: 7
- Custom runs: 3
- Total selected runs: 1158
- Direct inference files: 3824
- Continued-training files after restoration: 14448
Dense selection follows configs/wsm_v1_recommended_dense_channel_heads.json. Sparse and custom selection follows the recommended_checkpoint fields in their modality files. Reviewed dense Heads carrying a quality warning remain included because they are the documented checkpoint for a unique task. Unselected ablations and superseded runs are omitted.
TAR is intentionally uncompressed: it reduces repository file count without adding expensive compression around model states, NPZ files, and images that are already compressed. All archive members retain their original relative paths. Per-step pseudo-visualization, wsg-reader-visualization, generic visuals directories, and top-level sparse step previews are omitted.
Source directories are never modified. Local paths, user/host identifiers, emails, and credential-shaped strings are removed from uploaded text and supported PyTorch states. See RELEASE_MANIFEST.json for the exact selection, packaging, and privacy policy.
License
In short: You are welcome to use, study, modify, and extend this project
for noncommercial purposes, including research, experimentation, development,
training, evaluation, education, and personal projects. Commercial use
requires separate permission from Rutgers; contact
compliance@research.rutgers.edu and reference Rutgers Docket #2027-020. The
LICENSE file contains the governing terms.
The source code, documentation, and Rutgers-owned model artifacts in this repository, including trained or fine-tuned weights, checkpoints, adapters, and other model-weight files, are provided under the WSM Non-commercial Research License. Third-party components remain subject to their respective licenses.
- Downloads last month
- -