SemWAM RoboTwin Full v1

SemWAM checkpoint for RoboTwin. This repository contains an intermediate model snapshot and the configuration and normalization statistics needed for inference.

Checkpoint

  • Optimizer step: 238,910
  • Completed epochs at this snapshot: 10
  • Status: intermediate checkpoint; the full training run continues.
  • Checkpoint format: v2
  • Weight file: checkpoints/weights/step_238910.pt

The selected weights, saved inference contract, matching normalization statistics, finite tensor values and complete save at this step were verified before publication. Publication does not establish closed-loop benchmark performance.

Files

File Purpose
checkpoints/weights/step_238910.pt Current model weights, including the frozen semantic compressor
inference_config.json Saved inference contract
config.yaml Inference configuration reconstructed from the saved contract
dataset_stats.json Matching action and state normalization statistics
checkpoint_manifest.json File identities and checksums
model_card_summary.json Machine-readable model summary
trainer_state.reference.json Step and inference reference

This repository branch retains only the latest published checkpoint.

This package contains inference files only. Training data identities, private paths and initialization provenance, optimizer state and RNG state are omitted. Inference architecture, frozen encoder checksums and numerical normalization statistics are retained.

Inference

Use the SemWAM evaluation entrypoint with the selected v2 weight. Restore the saved model contract before constructing the model and use the matching normalization statistics. Required external frozen assets are identified by checksum in checkpoint_manifest.json; the semantic compressor is included in the checkpoint.

Downloads last month
47
Video Preview
loading