[RyanPCo]

Qwen 3.5 exists only from transformers 5.2.0, so the 4.53.2 pin cannot
stay; pin 5.17.0 (latest stable). The pin exists for openpi's
transformers_replace patch, whose files are verbatim copies of
transformers sources, so the two move together: the port and the
pi0.5-side 5.x fixes are in the submodule bump (openpi a385735 on
ryanco/transformers-5, not yet pushed).

egomimic side:

• from_pretrained(torch_dtype=) -> dtype= in the Qwen3 text stem and
  the two embedding-process scripts.
• hermetic_env no longer patches transformers.utils.hub._is_offline_mode
  unconditionally: 5.x dropped it and calls huggingface_hub.is_offline_mode(),
  which reads the constant we already patch.
• pi_unavailable also requires config.json in the HF cache: 5.x resolves
  the tokenizer class through the model config, so the tokenizer files alone
  are no longer enough (the cached google/paligemma-3b-mix-224 entry was
  missing it; fetched once online).
• New tests/unit/test_transformers_upgrade.py: the installed version
  matches the pyproject pin, transformers.models.qwen3_5 imports, and
  the openpi patch is actually installed (tuple-returning GemmaRMSNorm,
  the AdaRMS config fields, check(), embed_scale == 1).

Resolved transitive moves: tokenizers 0.21.4 -> 0.23.2, huggingface-hub
0.36.2 -> 1.31.0, safetensors 0.8.0 unchanged. torch, numpy, datasets,
timm and lightning are untouched.

The shared emimic venv is NOT upgraded by this branch; until it is, run
this branch against /coc/flash7/rco3/EgoVerse-wt/venvs/transformers-upgrade.

Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com

[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

• FlowVLA: backbone-neutral chassis and a ResNet+text backbone #661
• QwenVLA: a Qwen 3.5 VLM policy with a layer-wise cross-attention flow head #660
• hpt: Qwen 3.5-0.8B VLM stem pair (one joint image+text forward) #648
• deps: transformers 5.17.0 (Qwen 3.5), with openpi's patch ported #647 👈 (View in Graphite)
• hpt: cache frozen Qwen text features per unique prompt #646
• hpt: never re-initialise pretrained stems; hash-check them; mask text padding #645 : 1 other dependent PR (#664 )
• 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.