计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
package-apis【效率与协作】Explains VLA-Adapter Prismatic/VLA APIs, robot constants, checkpoint layouts, LoRA merge, and HF conversion utilities.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
This sub-skill documents the native package layout and public API surface; it
does not implement or vendor that package. The generated skill directory
contains only this documentation, references, and the read-only checkpoint
layout validator. The native prismatic/, vla-scripts/,
experiments/robot/, and conversion scripts remain in the separate checkout.
Before inspecting any native source or installed package, set the absolute root and enter it:
export VLA_ADAPTER_REPO_ROOT=/abs/path/to/VLA-Adapter
cd "$VLA_ADAPTER_REPO_ROOT"
python -m pip install -e "$VLA_ADAPTER_REPO_ROOT"
External prerequisites include the native checkout, its base Python package
dependencies, a compatible checkpoint, and (only for conversion or model
execution) the matching CUDA/Hugging Face stack. This sub-skill is not
self-contained and does not provide load, load_vla, action-head,
projector, LoRA-merge, or HF-conversion implementations. It does not load
weights or execute a model.
Use this sub-skill to read references/api-reference.md and references/prismatic-apis-and-conversion.md, map native entrypoints and checkpoint files, or run the local layout validator. Treat the API signatures and conversion inventory as reference facts, not copyable implementations.
Run the skill-local validator by absolute skill path:
python "$VLA_ADAPTER_SKILL_ROOT/sub-skills/package-apis/scripts/check_checkpoint_layout.py" \
--checkpoint "$VLA_ADAPTER_REPO_ROOT/outputs/CHECKPOINT" --help
The validator only inspects local files. For native operations, use the native
entrypoints from <absolute-repo-root> after separately confirming their
prerequisites; this adapter never invokes them.