公开交接包：V7b11 长窗口 tokenizer + realcam64 因果相机 DiT。仓库根不是工作目录，而是 tar 分卷（约 60GB，不压缩）。单文件超过 Hugging Face 上限，所以拆成多个 GPSToken-handoff.tar.part-。 拼完后先读 README.md 和.cursor/skills/gpstoken-migrate/SKILL.md。续训 DiT：export WANDBAPIKEY=... 然后 bash ops/resumedit.sh。本包不含任何密钥。 分卷校验和在 GPSToken-handoff.tar.sha256。
Independent publisher
Richard Zhang
Richard-ZZZZZ
Models in Library2
Datasets in Library1
Models on Hugging Face16
Followers—
Compare the published evidence
This profile connects the 3 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Reported licenses: other. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Models
This directory is the handoff for continuing particle-flow JEPA on a new machine. Open it as the project and follow.cursor/skills/jepa-particle-migration/SKILL.md. Locked Direct results stay on the 8-frame 144p GPSToken contract: PushT F0G0 85.67±0.58, TwoRoom F0G0 100.00±0.00. The long-window tokenizer to use for a longer canvas is V7b11 (weights/tracktokenizer/v7b11/ckpt00004000.pt), not a drop-in replacement for those checkpoints. Official task HDF5 files are not in this tree. They were moved to s3://lq-vision-usw2/fsx/ruichzhang/code/INTACT-JEPA/data/ on 2026-10-06. The public LeWM Hugging Face datasets listed in code/INTACT-JEPA/docs/INSTALL.md are the replacement when that bucket is…
Datasets
这是从 B200 集群打包上传的 particle-flow 全量可继续开发快照（含粒子流 Tokenizer、粒子流 DiT 世界模型、粒子流 JEPA 契约 A/B 世界模型、以及下游视频分割 VOS 与点追踪读出器）。 - Python 3.12，torch 2.8.0+cu128。 - 完整依赖锁文件在 env/requirements.lock（含 dmcontrol、mujoco、stableworldmodel、hydra-core、wandb、einops 等所有模块共用的依赖）。 - 在新机器上创建虚拟环境至 env/.venv（详见 env/README.md）： - V7b11（66 帧长窗口，11 帧分块，longchunk=11, longlayout=fixed）：gpstoken/ckpts/tokenizer/track-tok-v7b11-step4000/ckpt00004000.pt（SHA256 cc62f1b73b34487273aa09938d2bf82b05fd5c1de69d7ef440229c99bfa81555）。 - V7（≤17 帧 2D/3D 底座）：gpstoken/ckpts/tokenizer/track-tok-f2-selfcond-v7-step4500/ckpt00004500.pt。 - 3D 渲染与深度头：U3 gpstoken/experiments/gs3d/unified/U3/models/ckpt008000.pt；V6j…