# Free offline structural preflight for Diffusion/VQ-BeT LeRobot checkpoints — feedback welcome > Source: > Published: 2026-09-24 04:48:57+00:00 Sharing a small **AI-generated community utility** for inspecting a local Diffusion/VQ-BeT LeRobot saved-processor checkpoint before a full runtime setup. It is free and uses only Python’s standard library. **Source, usage and limits:** [diffusion-pusht-processor-checkpoint/docs/PREFLIGHT.md at main · perigot827/diffusion-pusht-processor-checkpoint · GitHub](https://github.com/perigot827/diffusion-pusht-processor-checkpoint/blob/main/docs/PREFLIGHT.md) After reviewing and downloading `preflight.py` as described in the docs, run: ``` python3 preflight.py /path/to/your/local/checkpoint ``` The script reads local checkpoint structure, JSON metadata and bounded SafeTensors headers. It makes no network requests and does not modify checkpoint files. Findings include missing required files, invalid JSON and unsafe file references. The repository’s lightweight checkpoint bundle intentionally omits learned weights, so a missing-weights finding on that bundle is expected. **Limits:** this is not a migration repair, runtime compatibility test, inference test, performance assessment or robot-safety check. A structural PASS does not prove that a model will load or work. This is not an official Hugging Face or LeRobot release. The current version passes 14 automated tests. It has not had human code review, and those tests are not evidence of independent adoption or time savings. If you already have a relevant local checkpoint, did a finding change your next debugging step? False positives, unclear output and “it added nothing” are useful feedback. An optional reply with the policy family, format/version, diagnostic **codes only** and your next step is enough. Please do not post checkpoint files, raw logs, local paths, credentials, email addresses or other private information. There is no purchase, registration or paid offer requested here.