Verifying Downloaded Weights in CI

Verify artifact identity before tests or deployment. The useful result is a method that preserves evidence, exposes constraints, and makes the next decision reviewable. This note describes an engineering workflow rather than a claim about a particular organization or production event.

Establish the boundary

Pin adapter version and expected digest. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

Download into a temporary directory

Download into a temporary directory. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

File size and sha256 before loading

Compare file size and sha256 before loading. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

Reject redirects to unexpected hosts

Reject redirects to unexpected hosts. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

Cache only verified files

Cache only verified files. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

A minimal load smoke test and clean up

Run a minimal load smoke test and clean up. Keep the relevant source version, configuration, and stable identifier together while this work is performed. That gives a reviewer a direct path from an observed outcome to the input and rule that produced it. Treat automatic checks and human judgment as separate controls, then record both in the release or incident note.

Practical check

Complete bash CI script using curl and sha256sum or shasum. The example below is intentionally small enough to run as a preflight check before a longer release, migration, or build workflow.

curl -fsSLO https://your-instance.example/download/qwen7b-ticket-triage
shasum -a 256 qwen7b-ticket-triage

Evidence for the next decision

Keep the input digest, outcome, and reviewer note with the resulting artifact or postmortem. Embedding access tokens in the script or logs. When the current evidence is incomplete, preserve the failing example, define the missing validation, and defer promotion until the result can be compared fairly.

Make the process idempotent. Repeating the same step on unchanged input should not append metadata, discard extra detail, or silently change ownership. Idempotence makes retries safe and exposes hidden state when outputs differ.

Review results by failure category rather than a single blended number. A small number of high-risk failures can outweigh an improvement on common cases, particularly when the output feeds a release or escalation workflow.

Keep accepted and rejected examples together. A concise rejected sample with its expected outcome is a durable regression test and prevents the same edge condition from returning as an unexplained surprise.

Use a separate holdout set for release decisions. Training records can guide implementation, but they cannot demonstrate that the process behaves correctly on new wording, new sources, or a changed input order.