[RyanPCo]

Adds DINOv3Stem, a pretrained DINOv3 ViT image stem in egomimic/models/hpt_nets.py that returns patch tokens at any aspect ratio. RDT (#680) uses it as its image encoder.

Behaviour changes

None. It is a new class, and no existing config uses it.

What it does

• Same call convention as ResNet: (B, T, N, 3, H, W) images in [0, 1] in, (B, patches, output_dim) out. CLS and register tokens are dropped.
• Loaded through timm, not transformers. The pinned transformers (4.53.2) has no DINOv3 class (it needs >= 4.56), and the timm/*_dinov3 hub repos are ungated where facebook/dinov3-* is gated. Default model: vit_base_patch16_dinov3.lvd1689m.
• Any aspect ratio. DINOv3 is RoPE-only, so image_size may be [H, W] (multiples of the patch size, otherwise a ValueError), with no position interpolation. The stem exposes grid_size.
• The stem owns the resize and the checkpoint's mean / std, so a config must not normalize these images in its augs.
• freeze_backbone freezes the tower, keeps it in eval mode and runs it under no_grad. backbone_parameters() returns the tower's parameters for a separate LR group; proj stays in the main group.

Pretrained-weight verification

pretrained_reference_state_dict reads the checkpoint back from the local HF cache (local_files_only=True, never the network), so the pretrained-weight hash check (#645) covers this tower as it covers the Qwen stems. An uncached checkpoint or a random tower is skipped rather than fetched. Against the real timm/vit_base_patch16_dinov3.lvd1689m file, the key sets match (162 tensors) and every value is equal.

Tests

tests/unit/test_dinov3_stem.py runs on a random-init tiny tower (no network, no weights). It covers:

• patch-token shapes, including 2 frames on the time axis;
• a frozen tower staying in eval mode with no gradients;
• a fine-tuned tower getting gradients;
• rejection of an off-grid size;
• the verifier passing on a matching cached checkpoint and raising once the tower is perturbed;
• an uncached checkpoint being skipped without a network call.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UkAZwERXcFKjueZQKfRF6J

[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

• vision: DINOv3Stem, a timm DINOv3 ViT image stem (patch tokens, any aspect ratio) #679 👈 (View in Graphite)
• loader: image history (a past front frame, gap in seconds) and the episode fps in every sample #678
• pi: initialize from PaliGemma with a scratch action expert #677 : 1 other dependent PR (#680 )
• 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.

[RyanPCo]

Folded into #680 to keep the stack under Graphite's 50-branch cap; the changes are unchanged there.