@wetron/gguf

Browser GGUF metadata, tensor descriptor, and weight payload parser.

import { parseGguf } from '@wetron/gguf';

const graph = parseGguf(bytes);

parseGguf supports GGUF v2 and v3. It reads every metadata key-value pair and tensor info entry. The returned ModelGraph starts with a GGUF metadata node and groups standardized tensor names into embedding, attention, feed-forward, state-space, transformer-block, and output stages. Tensor names, shapes, and GGML quantization types are available through initializers and tensorShapes.

For little-endian files with available weights, ModelGraph.weights.source.get(name) returns a zero-copy view of one tensor's encoded payload. @wetron/core decodes GGML scalar types and Q4_0 when the weight panel requests values. Other quantization types remain available as raw payload bytes and descriptors. Big-endian files expose descriptors without weight previews.