Function loadLinear

loadLinear(
    weights: Weights,
    path: string,
    config: ModelConfig,
    audit: WeightAudit,
    representations?: TrellisRepresentations,
): AnyLinear
Quantized when .scales exists (MLX-quantized checkpoints), else dense bf16/f16/f32. Both flavors carry the optional ADDITIVE .bias term (qwen2 qkv, starcoder2, …).
Parameters
- weights: Weights
- path: string
- config: ModelConfig
- audit: WeightAudit
- Optionalrepresentations: TrellisRepresentations
Returns AnyLinear
- Defined in packages/inference/src/layers/loading.ts:28