feat: expose HarmBlockMethod to external callers

[telpirion]

This enum is required for the SafetySettings() class to allow users to toggle harmful content blocking.

Without exposing this type, the following code won't work (as of version 1.72.0).

generative_models.SafetySetting(
        category=generative_models.HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
        method=generative_models.HarmBlockMethod.SEVERITY,
        threshold=generative_models.HarmBlockThreshold.BLOCK_ONLY_HIGH,
    ),

Found here.

[Ark-kun]

It should be exposed via SafetySettings.HarmBlockMethod. Is it not?

[telpirion]

It should be exposed via SafetySettings.HarmBlockMethod. Is it not?

Ah, now I see it. It is accessed through SafetySetting (no 's').

Question: why is HarmBlockThreshold and HarmCategory exposed directly in the generative_models module but not HarmBlockThreshold?

[Ark-kun]

Question: why is HarmBlockThreshold and HarmCategory exposed directly in the generative_models module but not HarmBlockThreshold?

Historical reasons. Initially we tried to use safety_settings: dict[HarmCategory, HarmBlockThreshold] instead of safety_settings: list[SafetySetting]. So SafetySetting was not exposed and HarmBlockThreshold and HarmCategory we exposed directly. Then HarmBlockMethod was added and we had to expose SafetySetting and switch to safety_settings: list[SafetySetting]. Now there are 4 classes related to safety. To avoid the generative_models module crowding we decided not to expose HarmBlockMethod and any subsequent classes, choosing to group them under SafetySetting.

Maybe we're wrong and we should just expose HarmBlockMethod the same way...

[telpirion]

Okay, makes sense. If you decide not to expose HarmBlockMethod in this module, feel free to delete this PR. Thanks!