pi0.5 Build Block Tower Reward Recap - Mixed
Fine-tuned pi0.5 checkpoint for block tower building using mixed reward recap semantics with advantage conditioning and dropout.
Experiment
- Config name:
pi05_build_block_tower_recap_mixed - Run type: experiment
- Objective: train block tower recap with mixed advantage conditioning (flat prompt, no subtask hierarchy) to compare against positive-only and subtask variants
Dataset
- 6 HuggingFace datasets:
villekuosmanen/build_block_towerplus 5 DAgger rounds (1.0.0 through 1.4.0)
Uploaded Checkpoints
30000: intermediate checkpoint, SHA-256341308cd5f8be0236d26cde04a1a7494c9428da7e5559996578b7ed55fb6dabf40000: intermediate checkpoint, SHA-256ec86899d8566a7c0e4f90839616f113fb0401d4288180d020d9d0f15f5972ef249999: final checkpoint, SHA-2568fbeda24b6593304cb0a6b42a5ec9dd342cdb5f0e1b25a074687785761eccf9f
Checkpoints are stored as params-only artifacts under checkpoints/<step>/params/.
Assets
assets/contains normalization stats and dataset metadata used by this run.
W&B
Repo Structure
checkpoints/30000/params/
checkpoints/30000/assets/
checkpoints/40000/params/
checkpoints/40000/assets/
checkpoints/49999/params/
checkpoints/49999/assets/
README.md
TRAINING_LOG.md