/**
 * WebGPU Quantization Support
 * Dequantization shaders for GGML quantization formats
 */
import { Tensor } from '../tensor.js';
import { GGMLType } from '../../../types/model.js';
/**
 * Get the block size for a quantization type
 */
export declare function getBlockSize(type: GGMLType): number;
/**
 * Get bytes per block for a quantization type
 */
export declare function getBytesPerBlock(type: GGMLType): number;
/**
 * Check if a type requires dequantization
 */
export declare function requiresDequantization(type: GGMLType): boolean;
/**
 * Dequantize Q4_0 data to f32
 */
export declare function dequantizeQ4_0(quantizedData: Uint8Array, numElements: number): Promise<Tensor>;
/**
 * Dequantize Q8_0 data to f32
 */
export declare function dequantizeQ8_0(quantizedData: Uint8Array, numElements: number): Promise<Tensor>;
/**
 * Dequantize Q4_K data to f32 using CPU (correct implementation)
 * Based on ggml-quants.c dequantize_row_q4_K
 */
export declare function dequantizeQ4_K(quantizedData: Uint8Array, numElements: number): Promise<Tensor>;
/**
 * Dequantize data based on type
 */
export declare function dequantize(quantizedData: Uint8Array, numElements: number, type: GGMLType): Promise<Tensor>;
/**
 * Quantize f32 to Q8_0 (for testing)
 */
export declare function quantizeToQ8_0(data: Float32Array): Uint8Array;
/**
 * Reset all cached pipelines (useful after shader updates)
 */
export declare function resetQuantPipelines(): void;
//# sourceMappingURL=index.d.ts.map