[RyanPCo]

Opt-in EMA of the trainable weights, using RDT's schedule: decay 1 - (1 + n)^-0.75, capped at 0.9999.

• egomimic/pl_utils/ema.py: WeightEMA, an fp32 shadow copy updated with one fused _foreach_lerp_ per optimizer step.
• ModelWrapper: updates the EMA once per optimizer step and validates on the EMA weights, restoring the raw weights afterwards; the raw backup is kept on CPU during validation. Checkpoints store the EMA under ema, and state_dict keeps the raw weights even when a checkpoint is saved mid-validation. A checkpoint without an EMA starts one from its current weights.
• load_checkpoint_weights passes the EMA through, so mode=eval evaluates the EMA weights.

Enable with +model.ema.enabled=true (optional: +model.ema.power, +model.ema.max_decay).

Tested: unit tests only (tests/unit/test_ema.py, CPU).
Not yet tested on GPU:

• trainer=debug run: checkpoint has an ema key, and Train/ema_decay is logged
• resume continues num_updates
• mode=eval matches the EMA val metrics
• peak memory at validation on the 1B config

🤖 Generated with Claude Code

https://claude.ai/code/session_01Jk3skP294hqzoTA37vu4xK

[RyanPCo]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• history: learned null token for dropped and episode-start-padded history #715
• mecka private: fixed-budget operator-scaling ladder generator #714
• train: hardened WSD defined in-config, W&B experiment tag, mode=validate on EMA weights #713
• val: exact-index metric frames, pad-masked metrics with thirds, per-loader eval RNG #712
• mecka private: mono converter, opscale configs and generators, IDs kept out of the repo #709
• data/rdt: keypoint RDT recipe, pose-glitch masking, hardened staging #708
• pl_model: RDT-style weight EMA, validated in place of raw weights #701 👈 (View in Graphite)
• data: stage episodes to node-local disk; never read them from /workspace #707
• pi: keep per-sample intrinsics so overlays use the episode camera #690 : 2 other dependent PRs (#691 , #716 )
• rdt: long-range visual memory; image memory window and history stride in seconds #689
• norm stats: cache sampled rows per episode so splits share them #687 : 1 other dependent PR (#688 )
• data: exclude episodes by hash; scan Mecka egomotion for failed head tracking #686
• eval: log validation videos to wandb #684 : 1 other dependent PR (#685 )
• slurm: stop cancelled jobs from leaking whole GPUs #682 : 1 other dependent PR (#683 )
• algo: RDT (Robotics Diffusion Transformer) on the HPT chassis #680 : 1 other dependent PR (#681 )
• pi: initialize from PaliGemma with a scratch action expert #677 : 1 other dependent PR (#678 )
• perf: vectorized image augs and a torch.compile knob for HPT and Pi; Pi clips gradients like openpi #676
• loader: one zarr read per key, not one per keymap entry #674
• eval: build the pinned-episode video batches only where they are used #673
• checkpoints: last.ckpt as a symlink, not a second full copy #672
• trainHydra: metric_frames_per_episode: auto, and warn on empty prompts #671
• transforms: per-step wrist framing for proprio history #670
• viz: one keypoint renderer, the horizon trace #669
• configs: history and backbone-LR knobs on the keypoint wrist HPT model #668
• trainer: ddp_bf16_compress, DDP with a bf16 gradient all-reduce #667
• eval: overlay videos play at the source rate under DDP #666
• hpt: build the policy on CPU and let Lightning place it #664 : 1 other dependent PR (#665 )
• hpt: never re-initialise pretrained stems; hash-check them; mask text padding #645 : 1 other dependent PR (#646 )
• obs: proprio history (K proprio steps, one token each, learned time embedding) #650
• loss: mask the repeat-last action padding out of the HPT loss #649
• resnet: keep a frozen backbone in eval mode; give backbones their own LR group #652
• augs: per-sample image augmentation for HPT train-mode augs #651
• eval: deterministic validation (seeded prompts + flow sampling); log-only grad-norm MAD #644
• data: mecka_fold_flagship_top3_hpt_6d, the 3-operator flagship train split #643
• val: subsampled metric loaders + pinned contiguous video loaders #642
• infra: mecka 6D / keypoint recipes, 300M models, defaults, and the rollout revert #640
• eval: layout-aware frame-invariant metrics, and a train-split viz head #659
• data: stop letting one bad episode or a rotation channel take a run down #658
• norm: guard degenerate channels, horizon pooling, and no rot6d normalization #657
• keypoints: the 144-D wrist-first hand-keypoint action becomes the human default #656
• mecka: fix the double-mirrored LEFT wrist frame, and correct existing zarrs #655
• 6d: PI goes 6D-only; Eva and Human gain the continuous-rotation modes #654
• 6d: continuous rotation representation, bimanual layouts, shared norm helpers #653
• train: give each Slurm slurm job gets its own cache dir to avoid conflict #638 : 1 other dependent PR (#639 )
• write git version into the run dir for reproducibility #637
• train: don't load PI base weights on resume and verify that checkpoint path exists #636
• configs: declare proprio and actions dims once per HPT model config #634 : 1 other dependent PR (#635 )
• cache norm stats for data configs to minimize recomputation #633
• import cloud packages lazily to support local training #631 : 1 other dependent PR (#632 )
• cache dataset resolver for faster training initialization #630
• validate episode hashes in data config before training #628
• modify pi camera map to match data #629
• add tests to ensure training works between every model and data vendor #626
• run CI tests on every PR #625
• add env variables for user-specific paths #624
• build: set pyproject and uv as the dependency managers #623
• fix(configs): fix paths for data hydra yaml configs #622
• refactor(algo): delete ACT #627 : 1 other dependent PR (#621 )
• tests: create dedicated test folder with unit and integration tests #620
• add pyright type checker #619
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.