Retrieval Tickets โ€” Contrastive Encoders

Contrastive encoders (ResNet50 + MLP head) trained on LeRobot LIBERO datasets for use in retrieval_tickets RETRIEVAL_SWITCH mode.

Files

Path Suite Size
libero_goal/contrastive_encoder.pt LIBERO goal (10 tasks) ~293 MB
libero_spatial/contrastive_encoder.pt LIBERO spatial (10 tasks) ~293 MB
libero_object/contrastive_encoder.pt LIBERO object (10 tasks) ~293 MB

Each: ~25.6M params, ResNet50 backbone with MLP projection head, trained ~500 epochs on LeRobot LIBERO frames with InfoNCE triplet loss + same-camera positive sampling.

Usage

# Pull all 3 (default in retrieval_tickets)
bash scripts/setup_encoder.sh

# Or single suite
bash scripts/setup_encoder.sh goal

After download:

encoders/libero_<suite>/contrastive_encoder.pt

Used in retrieval-switch evaluation:

"encoder_ckpt_path": "encoders/libero_goal/contrastive_encoder.pt"
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. ๐Ÿ™‹ Ask for provider support