Clarify Open-H checkpoint embodiment metadata

[beanscg]

Summary

• separate the Open-H dataset/platform count from released-checkpoint inference readiness
• point users to the existing Hugging Face config alignment validator before trying a new embodiment
• clarify that the embodiment comparison table is a dataset/config reference, not a guarantee that every released checkpoint contains processor/statistics metadata for every tag

Context

This addresses the ambiguity raised in #2 without changing model behavior or making new support claims. The README currently mentions 16 Open-H robot platforms, while direct inference depends on the metadata bundled with the selected checkpoint.

Refs #2.

Validation

• git diff --check
• confirmed scripts/validate_hf_config_alignment.py exists
• checked the README contains the new clarification text

[greptile-apps]

Greptile Summary

This PR clarifies the README to prevent user confusion about which Open-H robot platforms are inference-ready versus simply covered in the dataset. No code or config files are modified.

• The 16-platform count is moved from describing the model to describing the dataset, and a new paragraph directs users to run scripts/validate_hf_config_alignment.py against a specific checkpoint before trying a new embodiment.
• A trailing note on the embodiment table (§ Open-H Dataset) makes it explicit that the table is a dataset/config reference and that a released checkpoint may cover a smaller inference-ready set.

Confidence Score: 5/5

Documentation-only change; the referenced script exists and all three modified passages are internally consistent with the rest of the README.

All changes are prose clarifications in README.md. The validator script link resolves to a confirmed existing file, the 16-platform count is now correctly scoped to the dataset rather than the released checkpoints, and the new guidance paragraph does not contradict anything else in the repository.

No files require special attention.

Important Files Changed

Filename	Overview
README.md	Three targeted clarifications: sentence split separating dataset scope from model description, new paragraph pointing to the config-alignment validator, and a note that the embodiment table is a dataset reference not an inference guarantee.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User wants to run a new embodiment] --> B{Check checkpoint metadata}
    B --> C[Run scripts/validate_hf_config_alignment.py]
    C --> D{Embodiment tag in\nprocessor_config.json\n& statistics.json?}
    D -->|Yes| E[Proceed with direct inference]
    D -->|No| F[Embodiment not inference-ready\nin this checkpoint]
    F --> G[Fine-tune or select\ndifferent checkpoint]

Loading

_{Reviews (1): Last reviewed commit: "Clarify Open-H checkpoint embodiment met..." | Re-trigger Greptile}