The open-weight exemption you describe reverses the logic of pre-release testing. A proprietary developer can respond to a test result by delaying or changing a release. Published weights cannot be pulled back in the same way. The framework skips the models for which publication makes any failure detected afterward effectively irreversible.
The open-weight exemption you describe reverses the logic of pre-release testing. A proprietary developer can respond to a test result by delaying or changing a release. Published weights cannot be pulled back in the same way. The framework skips the models for which publication makes any failure detected afterward effectively irreversible.