Enums and Constants

This page documents the enums and constants available in the ZETIC Melange iOS SDK.

ModelMode

Controls the inference strategy for general (non-LLM) models.

import ZeticMLange

Value	Description
`RUN_AUTO`	Default. Balanced speed and accuracy (SNR > 20dB).
`RUN_SPEED`	Maximizes inference speed with minimum latency.
`RUN_ACCURACY`	Maximizes precision based on SNR scores.

Usage

let model = try ZeticMLangeModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    modelMode: RUN_SPEED
)

LLMModelMode

Controls the inference strategy for LLM models.

Value	Description	Status
`RUN_SPEED`	Most aggressive quantization for minimum latency.	Available
`RUN_AUTO`	Balanced speed and accuracy across benchmark datasets.	Paused
`RUN_ACCURACY`	Highest precision quantization.	Paused

Usage

let model = try ZeticMLangeLLMModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    version: VERSION,
    modelMode: RUN_SPEED
)

LLMDataSetType

Specifies the benchmark dataset for accuracy evaluation in LLM Accurate mode.

Value	Description
`MMLU`	Massive Multitask Language Understanding
`TRUTHFULQA`	TruthfulQA benchmark
`CNN_DAILYMAIL`	CNN/DailyMail summarization
`GSM8K`	Grade School Math 8K

Usage

let model = try ZeticMLangeLLMModel(
    personalKey: PERSONAL_KEY,
    name: MODEL_NAME,
    version: VERSION,
    modelMode: RUN_ACCURACY,
    dataSetType: MMLU
)

LLMKVCacheCleanupPolicy

Controls how the LLM engine handles a full KV cache.

Value	Description
`CLEAN_UP_ON_FULL`	Clears the entire context when the KV cache is full. (Default)
`DO_NOT_CLEAN_UP`	Keeps the context without cleanup when the KV cache is full.

When using DO_NOT_CLEAN_UP, calling run() again without calling cleanUp() may cause unexpected behavior.

Enums and Constants

ModelMode

Usage

LLMModelMode

Usage

LLMDataSetType

Usage

LLMKVCacheCleanupPolicy

See Also

On this page