# Interface: ModelConfig

## Properties

### expert\_dtype

• `Optional` **expert\_dtype**: `string`

Some MoEs store their experts at a narrower precision than the rest of the model and declare
it here, *outside* `quantization_config` (e.g. DeepSeek-V4 is `quant_method: "fp8"` for
attention but `expert_dtype: "fp4"` for the experts, which dominate the parameter count).

#### Defined in[[expertdtype.defined-in]]

[packages/hub/src/lib/parse-safetensors-metadata.ts:724](https://github.com/huggingface/huggingface.js/blob/main/packages/hub/src/lib/parse-safetensors-metadata.ts#L724)

___

### quantization

• `Optional` **quantization**: [`MlxQuantizationConfig`](MlxQuantizationConfig)

Current MLX format, optionally with per-module overrides keyed by module name.

#### Defined in[[quantization.defined-in]]

[packages/hub/src/lib/parse-safetensors-metadata.ts:716](https://github.com/huggingface/huggingface.js/blob/main/packages/hub/src/lib/parse-safetensors-metadata.ts#L716)

___

### quantization\_config

• `Optional` **quantization\_config**: [`MlxQuantizationConfig`](MlxQuantizationConfig) \| [`QuantizationConfig`](QuantizationConfig)

#### Defined in[[quantizationconfig.defined-in]]

[packages/hub/src/lib/parse-safetensors-metadata.ts:717](https://github.com/huggingface/huggingface.js/blob/main/packages/hub/src/lib/parse-safetensors-metadata.ts#L717)

___

### text\_config

• `Optional` **text\_config**: `Pick`\<[`ModelConfig`](ModelConfig), ``"expert_dtype"`` \| ``"quantization"`` \| ``"quantization_config"``\>

#### Defined in[[textconfig.defined-in]]

[packages/hub/src/lib/parse-safetensors-metadata.ts:718](https://github.com/huggingface/huggingface.js/blob/main/packages/hub/src/lib/parse-safetensors-metadata.ts#L718)