GGUF Metadata Reader
Inspect a local GGUF file’s header and metadata in your browser, then download a JSON summary. Your model file stays on your device.
Select a file or try the synthetic sample. No upload occurs.
Metadata
| Key | Value |
|---|
What this reader checks
The tool parses little-endian GGUF v2/v3 header fields and scalar metadata. Flat arrays are summarized without decoding their values; nested arrays are not supported. Strings above 1 MiB are omitted. Inspection stops at 64 MiB of header/metadata, 100,000 metadata entries or one million elements per array.
It reads bounded portions of the local file rather than loading a complete model into memory. It does not validate the tensor directory, tensor payloads, array contents, model authenticity or runtime compatibility. A successful read is not proof that the model is complete or runnable. The sample contains metadata only and is not a runnable model.
From metadata to a memory estimate
Look for architecture, layer count and attention settings when present. Metadata keys differ across architectures; absence of a key does not prove a file is broken. Split GGUF files describe the selected shard, not the complete download. File size is not total runtime VRAM use.
Use the VRAM calculator for an explicit memory estimate and the KV-cache calculator to explore context costs. These estimates still need confirmation in your runtime.
Format reference: official GGUF specification.