Downloaded models often corrupted — need checksum validation and more robust download handling

[hakabe]

Description:

Frequently, models downloaded directly through the GPT4All app become corrupted and fail to load properly. This issue occurs inconsistently — occasionally, re-downloading the same model fixes the problem, but other times multiple attempts are needed before the model works.

When attempting to load a corrupted model, the app throws various errors, suggesting that the download was incomplete or damaged.

Steps to Reproduce:

Open the GPT4All app.

Navigate to the Model tab and download any model (e.g., mistral, orca-mini, etc.).

Wait for the download to complete.

Attempt to load the downloaded model.

Expected Behavior:

Models should download completely and load successfully every time. The app should verify download integrity before allowing a model to be loaded.

Actual Behavior:

Some model downloads appear to complete successfully but are actually corrupted.

Loading these models results in runtime or initialization errors.

Re-downloading sometimes fixes the issue, but the behavior is inconsistent.

Environment:

App version: GPT4All v3.10.0

OS: Windows 10 Pro

Internet connection: Stable 1Gbps Fibre

Suggested Improvements:

Implement MD5/SHA checksum validation to ensure file integrity after each download.

Add automatic retry or resume capability for incomplete downloads.

Provide clear user feedback (e.g., “Verifying model integrity…”).

Offer a manual integrity check or “redownload” option in the UI.