V3 Migration Guide#

V1 to V2 migration

If you are still on V1 please see the V2 Migration Guide first.

What has changed?#

  • We have simplified and improved moderations.

  • All Schemas were moved to a single place and should now be imported from genai.schema module.

  • Some schemas were renamed for better readability.

Moderations#

  • Stigma (ModerationStigma class) has been replaced by Social Bias (ModerationSocialBias class).

  • Implicit Hate (ModerationImplicitHate class) has been replaced by Social Bias (ModerationSocialBias class).

  • You can now set a different threshold for input and output moderations.

  • Text generation response moderation was renamed to moderations:

    TextChatStreamCreateResponse.moderation -> TextChatStreamCreateResponse.moderations TextGenerationStreamCreateResponse.moderation -> TextGenerationStreamCreateResponse.moderations TextGenerationResult.moderation -> TextGenerationResult.moderations

❌ Old Way

from genai.schema import ModerationHAP, ModerationStigma, ModerationImplicitHate, ModerationParameters

for response in client.text.generation.create(
    model_id="google/flan-t5-xl",
    inputs=["What is a molecule?", "What is NLP?"],
    moderations=ModerationParameters(
        hap=ModerationHAP(input=True, output=True, threshold=0.8),
        stigma=ModerationStigma(input=True, output=True, threshold=0.8),
        implicit_hate=ModerationImplicitHate(input=True, output=True, threshold=0.8),
    )
):
    for result in response.results:
        moderation = result.moderation

✅ New Way

from genai.schema import (
    ModerationHAP,
    ModerationHAPInput,
    ModerationHAPOutput,
    ModerationParameters,
    ModerationSocialBias,
    ModerationSocialBiasInput,
    ModerationSocialBiasOutput,
)

for response in client.text.generation.create(
    model_id="google/flan-t5-xl",
    inputs=["what is a molecule?", "what is nlp?"],
    moderations=moderationparameters(
        hap=moderationhap(
            input=moderationhapinput(enabled=true, threshold=0.8),
            output=moderationhapoutput(enabled=true, threshold=0.8),
        ),
        social_bias=moderationsocialbias(
            input=moderationsocialbiasinput(enabled=true, threshold=0.8),
            output=moderationsocialbiasoutput(enabled=true, threshold=0.8),
        ),
    )
):
    for result in response.results:
        moderation = result.moderations

Imports#

  • It is no longer to import schemas from submodules, use from genai.schema import ...

❌ Old Way

from genai.text.generation import DecodingMethod

✅ New Way

from genai.schema import DecodingMethod

📝 Other changes#

  • Deprecate TuningType enum; use values from client.tune.types() method.

  • Following schemas or their properties were renamed:

    • UserPromptResult -> PromptResult

    • PromptsResponseResult -> PromptResult

    • UserResponseResult -> UserResult

    • UserCreateResultApiKey -> UserApiKey

    • PromptRetrieveRequestParamsSource -> PromptListSource

    • BaseMessage.file_ids -> BaseMessage.files