AlpaJudge

AlpaJudge is a family of learned judges for evaluating autonomous-driving reasoning. This repository provides one reference-based checkpoint and one reference-free checkpoint.

License/Terms of Use:

Model weights: The model weights are released under the OpenMDW-1.1 license.

Source code: Apache License 2.0, as provided in the AlpaJudge GitHub repository.

Checkpoints

Variant	Folder	Inputs	Output
AlpaJudge-RB	reference-based/	Question, reference answer, and candidate answer; no images	One alignment probability in [0, 1]
AlpaJudge-RF	reference-free/	Front-wide/front-tele frames, ego meta-action history and future, and a candidate rationale	Five rubric probabilities plus the aggregated action, component, and overall scores

Both variants are available from nvidia/alpajudge in the subfolders shown above.

Download

Authenticate with an account that can access the repository, then download only the variant you need:

from pathlib import Path

from huggingface_hub import snapshot_download

variant = "reference-based"  # or "reference-free"
repo_dir = Path(
    snapshot_download(
        "nvidia/alpajudge",
        allow_patterns=f"{variant}/*",
    )
)
checkpoint = repo_dir / variant

Use the alpajudge inference package to load either subfolder. See the usage guide for setup, complete RB and RF examples, input contracts, score definitions, and limitations.

Support

📣 Usage questions: post on the Alpamayo NV Developer Forum.

🐛 Code bugs / documentation issues / feature requests: file a GitHub issue using the appropriate template (Bug report, Documentation request, or Feature request) at https://github.com/NVlabs/alpajudge/issues/new/choose . The relevant NVIDIA responder is auto-assigned.

🚨 Security vulnerabilities: please use NVIDIA's Vulnerability Disclosure Program. Do not file public issues.