THETA Bench
Collection
THETA Bench simulation policies (3,003 segments), selected real G1 adaptations (91 demos), and training data. • 15 items • Updated
THETA Bench Psi0: Simulation training, 3,003 segments
The validated final policy checkpoint is available in this repository.
This is a THETA training-result repository. It does not substitute an upstream pretrained policy for a THETA-trained checkpoint.
| Setting | Value |
|---|---|
| Training stage | Simulation training, 3,003 segments |
| Target optimizer updates | 40000 |
| Per-GPU batch / GPUs / global batch | 16 / 8 / 128 |
| Gradient accumulation | 1 |
| Conditions per global batch | 18 |
| Dataset revision | 8b2cd31e107b64cb13f812ea217a63a20845c78a |
The simulation pool contains 1,200 successful L1/L2 demonstrations and 1,803 extracted L0 prefixes, spanning 18 conditions. The 3,003 segments are not 3,003 independent demonstrations.
Use the model's native THETA adapter and model-specific dependencies. This repository does not claim compatibility with arbitrary Transformers or simulation loaders. No evaluation score is claimed by checkpoint publication.
Use the pinned Psi0 native code and THETA adapter, run directory run and final checkpoint step. Change cwd to exported run for dataset_statistics.json resolution. Before invoking Psi0Model.from_pretrained, set psi.models.psi0.QWEN3VL_VARIANT to the absolute bundled processor directory (Qwen/Qwen3-VL-2B-Instruct@89644892e4d85e24eaac8bacfd4f463576704203). This overrides the native unpinned processor/config identifier without loading upstream VLM weights. The trained safetensors strictly loads both VLM and action_header; native token embedding tying is retained. Keep the native one-camera preprocessing, instruction case, state/action normalization, RTC and H30 configuration.
Training uses independent model optimizers and shared GPU execution through MPS. Publication is performed by a CPU uploader after final checkpoint validation.