admin/any4lerobot
1
0
Fork 0
mirror of https://github.com/Tavish9/any4lerobot.git synced 2026-07-24 12:35:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add generic converter adapter hooks

Browse Source 
Co-authored-by: Codex <codex@openai.com>
This commit is contained in:
Tavish
2026-06-21 15:44:14 +08:00
parent 7a8642edfc
commit 65296e75cb
3 changed files with 71 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
generic_converter/README.md
+12 -4
View File
@@ -43,7 +43,13 @@ Optional attributes:
Required methods:
- `load_tasks(self) -> list[ConversionTask]`
- `load_subset(self, task: ConversionTask) -> Iterable[Sequence[dict]]`
- `load_subset(self, task: ConversionTask) -> Iterable[Any]`
Optional hooks:
- `create_dataset(self, task: ConversionTask)`
- `save_episode(self, dataset, episode_data, task) -> bool`
- `get_episode_length(self, episode_data) -> int`
`run_converter` reads `adapter.output_path` and calls `adapter.load_tasks()`
without arguments. Store paths, task manifests, or other adapter options on the
@@ -53,9 +59,11 @@ Use `adapter.temp_output_path` when building task-level temporary output paths.
`load_subset` receives the full `ConversionTask`, not just an input path. Use
`task.input_path` for raw data and `task.metadata` for dataset-specific values
such as language instructions. Each yielded episode must be a sequence of frame
dictionaries accepted by `LeRobotDataset.add_frame`; each frame should include
the LeRobot `task` field when language tasks are needed.
such as language instructions. By default, each yielded episode must be a
sequence of frame dictionaries accepted by `LeRobotDataset.add_frame`; each
frame should include the LeRobot `task` field when language tasks are needed.
Adapters that need custom dataset classes or extra per-episode arguments can
override the optional hooks.
## ConversionTask