Skip to content

Prepare EgoScale Stage 1 DLC full training #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
Apexcmgd wants to merge 65 commits into
base: main
Choose a base branch
from
Draft

Prepare EgoScale Stage 1 DLC full training #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
65 commits
Select commit Hold shift + click to select a range
a57cb32
Migrate cotrain full-data config updates
wudi7012 Jul 14, 2026
9509d3d
ready for unified action space design
wudi7012 Jul 15, 2026
aa15880
Add unified 80D action mappings
wudi7012 Jul 15, 2026
59e4550
Map cotrain RLDS data into unified 80D space
wudi7012 Jul 15, 2026
5aa44f3
Mask unified action training and evaluation
wudi7012 Jul 15, 2026
86c2be9
Audit unified actions on real RLDS data
wudi7012 Jul 15, 2026
5b0728b
Make unified 80D mandatory for cotrain configs
wudi7012 Jul 15, 2026
30c240e
move to baiducloud and add docs
wudi7012 Jul 15, 2026
ab4f99d
env config
wudi7012 Jul 16, 2026
ef6fd8b
prepare for baiducloud training
wudi7012 Jul 16, 2026
2cdade4
add doc
wudi7012 Jul 16, 2026
31f8e30
pre for training
wudi7012 Jul 17, 2026
d7b690c
fix dataset
wudi7012 Jul 17, 2026
f3b718f
update doc
wudi7012 Jul 18, 2026
b28a546
add assets
wudi7012 Jul 18, 2026
e31fed8
add norm, clean config
wudi7012 Jul 18, 2026
e92c916
fix eval bc
wudi7012 Jul 18, 2026
d56e2dc
finish ego norm
wudi7012 Jul 19, 2026
3753d26
Add parallel-gripper staged training workflow
Jul 20, 2026
edef0c4
Fix fresh-clone DSW environment setup
Jul 20, 2026
358b790
Make DSW setup idempotent
Jul 20, 2026
2f955b7
Speed up DSW dependency setup
Jul 20, 2026
0580754
Route locked DSW installs through Alibaba mirror
Jul 20, 2026
3400181
Add low-memory params-only checkpoints
Jul 20, 2026
d23f6e0
Add B200 9999 validation evaluation
wudi7012 Jul 21, 2026
e61cf18
fix evaluation
wudi7012 Jul 21, 2026
2d88258
Add legacy32 real-only action-space ablation
wudi7012 Jul 21, 2026
7ab1b89
Use batch 512 for legacy32 four-GPU control
wudi7012 Jul 21, 2026
f0fc981
Document branch checkout for legacy32 jobs
wudi7012 Jul 21, 2026
d6cb08a
Add Piper30 Aliyun legacy replay config
wudi7012 Jul 22, 2026
8bf2e81
Align Aliyun replay with global batch 32
wudi7012 Jul 22, 2026
dc0670a
Revert "Align Aliyun replay with global batch 32"
wudi7012 Jul 22, 2026
6aae966
Add Piper comparison experiments with Aliyun recipe
wudi7012 Jul 23, 2026
829d57b
Support configurable training parameter initialization
wudi7012 Jul 24, 2026
9d8f09f
Use aligned EgoVerse action chunks for stage one
Jul 27, 2026
529dc62
Fix TensorFlow rank assertion test
Jul 27, 2026
10813f9
Add EgoMimic stage two adapter
Jul 27, 2026
7aaa4d1
Correct EgoMimic groceries as bimanual
Jul 27, 2026
85f744a
Shard long EgoMimic demos for multihost
Jul 27, 2026
cafd2bf
Use disjoint EgoMimic human validation
Jul 27, 2026
aac693c
Document EgoMimic stage two smoke
Jul 27, 2026
1083c54
Support finite full-data norm statistics
Jul 27, 2026
5609563
Clean up norm statistics lint
Jul 27, 2026
6649834
Add all-task EgoMimic stage two recipe
Jul 27, 2026
4f36266
Document all-task EgoMimic smoke
Jul 27, 2026
b8ad99b
add gemma config
wudi7012 Jul 28, 2026
38677c2
doc update
wudi7012 Jul 30, 2026
1dd59df
fix doc
wudi7012 Jul 31, 2026
5eea8d5
add new piper30 val
wudi7012 Aug 3, 2026
7cbee66
update doc
wudi7012 Aug 3, 2026
b4d65ca
Merge remote-tracking branch 'origin/wudi' into dev/shengjunhe
Aug 4, 2026
79ac2fc
Add EgoVerse RL2 and Gemma stage initialization
Aug 4, 2026
72d1bc3
Preserve strict staged checkpoint inheritance
Aug 4, 2026
d23d3a2
Add self-collected aligned Stage 2 data contract
Aug 4, 2026
63b11f0
Finalize self-collected Stage 2 data loading
Aug 5, 2026
26a1358
Prepare Stage 1 DLC full training
Aug 6, 2026
805aa5d
Keep rerun stub link directory
Aug 6, 2026
d6e3ce0
feat: rebuild aligned stage2 dataset pipeline
Aug 7, 2026
68bda99
feat: unify stage2 aligned actions as relative SE3
Aug 10, 2026
9b0db70
Fix Stage 3 Piper DLC fine-tuning
Aug 12, 2026
7d22425
Allow relocated RLDS roots in preflight
Aug 12, 2026
e8fa3a8
Add robust DLC Stage 3 smoke entrypoint
Aug 12, 2026
a70aeb5
Mirror DLC smoke logs to console
Aug 12, 2026
ebaf24f
Add robust DLC Stage 3 formal entrypoint
Aug 12, 2026
65a3ce8
Add staged Stage 3 and Stage 4 training configs
Aug 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
7 changes: 4 additions & 3 deletions .gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,4 @@
# Data directories.
assets/
checkpoints/
data/
wandb/
Expand Down Expand Up @@ -170,5 +169,7 @@ cython_debug/
# new
uv.lock
Miniforge3-Linux-x86_64.sh
third_party/rerun-stub/

third_party/rerun-stub/*
!third_party/rerun-stub/dist/
third_party/rerun-stub/dist/*
!third_party/rerun-stub/dist/.gitkeep
3 changes: 3 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,8 @@
# openpi

> Project-specific staged Ego-to-robot training and Alibaba Cloud DSW/DLC instructions:
> [docs/egoscale_staged_training.md](docs/egoscale_staged_training.md).

openpi holds open-source models and packages for robotics, published by the [Physical Intelligence team](https://www.physicalintelligence.company/).

Currently, this repo contains three types of models:
Expand Down
Loading
Loading