You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Dexas Policy Checkpoints

This repository contains deployment-oriented checkpoint bundles for Dexas baku, pi05, and OpenPI JAX policies.

Included bundles

baku_h200_full/

  • epoch_0039.pt
  • epoch_0049.pt
  • epoch_0099.pt
  • epoch_0149.pt
  • epoch_0199.pt
  • epoch_0249.pt
  • args.json
  • stats.json
  • stats.pkl

pi05_full_train100/

  • metrics.jsonl
  • 1000/model.safetensors
  • 1000/metadata.pt
  • 1000/assets/dexas/norm_stats.json
  • 2000/model.safetensors
  • 2000/metadata.pt
  • 2000/assets/dexas/norm_stats.json
  • 3000/model.safetensors
  • 3000/metadata.pt
  • 3000/assets/dexas/norm_stats.json
  • 4000/model.safetensors
  • 4000/metadata.pt
  • 4000/assets/dexas/norm_stats.json
  • 10000/model.safetensors
  • 10000/metadata.pt
  • 10000/assets/dexas/norm_stats.json
  • 15000/model.safetensors
  • 15000/metadata.pt
  • 15000/assets/dexas/norm_stats.json
  • 20000/model.safetensors
  • 20000/metadata.pt
  • 20000/assets/dexas/norm_stats.json

openpi_jax_rgbfix_kp1000_r3/

  • pi0_dexas_easy_mode_v3_abs/: wandb_id.txt and steps 2000, 3000, 4000, 10000, 20000
  • pi0_dexas_easy_mode_v3_delta/: wandb_id.txt and steps 2000, 3000, 4000, 10000, 20000
  • pi05_dexas_easy_mode_v3_abs/: wandb_id.txt and steps 2000, 3000, 4000, 10000, 20000
  • pi05_dexas_easy_mode_v3_delta/: wandb_id.txt and steps 2000, 3000, 4000, 10000, 20000
  • OpenPI JAX Orbax inference bundles; each step includes _CHECKPOINT_METADATA, assets//norm_stats.json, and params/**.
  • train_state/ is intentionally omitted.

baku_scaled_20260419/

  • epoch_0034.pt
  • epoch_0044.pt
  • epoch_0059.pt
  • epoch_0119.pt
  • epoch_0179.pt
  • args.json
  • stats.json
  • stats.pkl
  • scaled BAKU run with hidden_size=384, depth=12, batch_size=512.

openpi_jax_inference_bundle_v1/

  • v1/: steps 20000, 25000, 30000, 40000, 49999
  • v2/: steps 20000, 25000, 30000, 40000, 49999
  • v3/: steps 20000, 25000, 30000, 40000, 49999
  • v4/: steps 20000, 25000, 30000, 40000, 49999
  • v5/: steps 20000, 25000, 30000, 40000, 49999
  • v6/: steps 20000, 25000, 30000, 40000, 49999
  • OpenPI JAX Orbax inference bundles; each step includes _CHECKPOINT_METADATA, assets//norm_stats.json, and params/**.
  • train_state/ is intentionally omitted.
  • wandb_id.txt is intentionally omitted.
  • Only steps that were actually uploaded in the current batch are listed above.

Notes

  • openpi_jax_inference_bundle_v1 is an anonymous OpenPI JAX Orbax inference-only bundle; train_state/ and wandb_id.txt are intentionally omitted.
  • pi05 is uploaded as a deployment/download bundle; optimizer.pt is intentionally omitted.
  • baku checkpoints are stored on a 5-epoch cadence, so the requested 100 milestone maps to epoch_0099.pt.
  • openpi_jax_rgbfix_kp1000_r3 is uploaded as an OpenPI JAX Orbax inference-only bundle from h200_cpu; train_state/ is intentionally omitted.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support