admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-08-08 17:39:44 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(g05): publish the 27D base checkpoint

Browse Source
This commit is contained in:
Pepijn
2026-07-29 12:35:49 +02:00
parent d94fef7b40
commit e6ec909cc7
1 changed files with 25 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/g05.mdx
+25 -1
View File
@@ -33,12 +33,29 @@ ActionCodec metadata, statistics, exact prompt template, source revision, and
license. Loading rejects a different head, horizon, processor mode, or license. Loading rejects a different head, horizon, processor mode, or
normalization contract. normalization contract.
The converted checkpoints are private under the LeRobot organization and pinned
to the revisions validated for this integration:
| Repository | Revision | Contract |
| ------------------------ | ------------------------------------------ | ---------------------------- |
| `lerobot/g05_base` | `a74ab9a849a8e5220e2ec3c0ac94e1bcadb000d4` | 27D base, R1 Lite processor |
| `lerobot/g05_libero` | `cea5cb93bf98264d5b8b498c64b18b5b768aa1ed` | 20D LIBERO |
| `lerobot/g05_robotwin20` | `f3405ee0f995a7ee9bdb6992353445cae32efe54` | 20D RoboTwin |
| `lerobot/g05_so101` | `6379367ea401bb8ff1f954848721e140d8260060` | 20D SO-101 (`so100` profile) |
The named `atomic_4` adapter is intentionally separate from LIBERO. Its raw state The named `atomic_4` adapter is intentionally separate from LIBERO. Its raw state
is EEF relative xyz+quaternion, base xyz+quaternion, and two gripper positions is EEF relative xyz+quaternion, base xyz+quaternion, and two gripper positions
(16D). Its action is EEF delta xyz+rpy, gripper, four mobile-base commands, and a (16D). Its action is EEF delta xyz+rpy, gripper, four mobile-base commands, and a
control-mode flag (12D). This needs G0.5's 27D whole-body layout; a released 20D control-mode flag (12D). This needs G0.5's 27D whole-body layout; a released 20D
LIBERO checkpoint is rejected. LIBERO checkpoint is rejected.
`lerobot/g05_base` supplies those 27D model weights, both action heads, the
ActionCodec tokenizer, and the released six-step R1 Lite processor/statistics
contract. OpenGalaxea's base release does not contain an `atomic_4` processor or
Atomic-4 dataset statistics. Loading the base weights for Atomic-4 therefore
requires an `atomic_4` `G05Config` plus statistics computed from the target
Atomic-4 dataset; reusing the R1 Lite statistics would be incorrect.
## Install ## Install
Install LeRobot's small config dependency: Install LeRobot's small config dependency:
@@ -58,7 +75,7 @@ git -C GalaxeaVLA checkout b34966f387dd2ae0f003143b81494afd9213e613
export PYTHONPATH="/path/to/GalaxeaVLA/src:${PYTHONPATH}" export PYTHONPATH="/path/to/GalaxeaVLA/src:${PYTHONPATH}"
``` ```
The LeRobot organization hosts the prepared LIBERO, RoboTwin, and SO-101 The LeRobot organization hosts the prepared base, LIBERO, RoboTwin, and SO-101
checkpoints privately. Authenticate with `hf auth login` before loading them. checkpoints privately. Authenticate with `hf auth login` before loading them.
SO-100 and SO-101 share the released `so100` embodiment token and six-joint SO-100 and SO-101 share the released `so100` embodiment token and six-joint
right-arm contract. A missing left-wrist camera is zero-padded exactly as in the right-arm contract. A missing left-wrist camera is zero-padded exactly as in the
@@ -211,6 +228,13 @@ recipe-driven path. It produced finite total loss `4.38332`, including non-zero
`ce_loss=4.04244` and `fm_loss=0.340881`, with 15.58 GiB peak allocated CUDA `ce_loss=4.04244` and `fm_loss=0.340881`, with 15.58 GiB peak allocated CUDA
memory. memory.
The converted private `g05_base` checkpoint strict-loaded all 946 mapped tensors
with no missing, duplicate, unexpected, or shape-mismatched keys. Author/LeRobot
parity was exact for images, tokens, and masks; normalized state/action inputs
matched within `1.2e-7`, and postprocessing within `1.8e-6`. A System 2 smoke
generated a native `Subtask:` trace and a finite `[1, 32, 27]` action chunk in
the same pass, using 11.02 GiB peak allocated memory.
A 50-episode LIBERO/RoboTwin success-rate comparison additionally requires the A 50-episode LIBERO/RoboTwin success-rate comparison additionally requires the
matching simulator, task assets, reset seeds, and author evaluator; no task-level matching simulator, task assets, reset seeds, and author evaluator; no task-level
benchmark number is claimed until that separate gate runs. benchmark number is claimed until that separate gate runs.