IMPORTANT: In July 2025, we stopped releasing new versions of KTX modules, and we removed the KTX libraries from the Firebase Android BoM (v34.0.0). If you use KTX APIs from the previously released KTX modules, we recommend that you migrate your app to use KTX APIs from the main modules instead. For details, see the FAQ about this initiative.

OnDeviceConfig

@PublicPreviewAPI
class OnDeviceConfig

Configuration for on-device AI model inference.

Summary

Public companion properties
`OnDeviceConfig`	`IN_CLOUD` A default configuration that only uses in-cloud inference.

Public constructors
`OnDeviceConfig( mode: InferenceMode, maxOutputTokens: Int?, temperature: Float?, topK: Int?, seed: Int?, candidateCount: Int, modelOption: OnDeviceModelOption?, enableThinking: Boolean? )`

Public properties
`Int`	`candidateCount` The number of generated responses to return.
`Boolean?`	`enableThinking` Whether to enable thinking mode for the on-device model.
`Int?`	`maxOutputTokens` The maximum number of tokens to generate in the response.
`InferenceMode`	`mode` The `InferenceMode` to use for the model.
`OnDeviceModelOption?`	`modelOption` Configuration for the on-device model selection and performance.
`Int?`	`seed` The seed to use for generation to ensure reproducibility.
`Float?`	`temperature` A parameter controlling the degree of randomness in token selection.
`Int?`	`topK` The `topK` parameter changes how the model selects tokens for output.

Public companion properties

IN_CLOUD

val IN_CLOUD: OnDeviceConfig

A default configuration that only uses in-cloud inference.

Public constructors

OnDeviceConfig

OnDeviceConfig(
    mode: InferenceMode,
    maxOutputTokens: Int? = null,
    temperature: Float? = null,
    topK: Int? = null,
    seed: Int? = null,
    candidateCount: Int = 1,
    modelOption: OnDeviceModelOption? = null,
    enableThinking: Boolean? = null
)

Public properties

candidateCount

val candidateCount: Int

The number of generated responses to return. See GenerationConfig for more detail. By default it's set to 1.

enableThinking

val enableThinking: Boolean?

Whether to enable thinking mode for the on-device model.

maxOutputTokens

val maxOutputTokens: Int?

The maximum number of tokens to generate in the response. See GenerationConfig for more detail.

mode

val mode: InferenceMode

The InferenceMode to use for the model.

modelOption

val modelOption: OnDeviceModelOption?

Configuration for the on-device model selection and performance.

seed

val seed: Int?

The seed to use for generation to ensure reproducibility. See GenerationConfig for more detail.

temperature

val temperature: Float?

A parameter controlling the degree of randomness in token selection. See GenerationConfig for more detail.

topK

val topK: Int?

The topK parameter changes how the model selects tokens for output. See GenerationConfig for more detail.