TinyRustLM does not currently load GGUF. It loads a custom little-endian SLM1 container with a fixed v1 header, embedded tokenizer, hashed tensor directory, aligned tensor payloads, and f32/q8_0/q4_0 storage.
Admission is layered
Runtime parsing validates bounds and required tensors. The native packer adds source, provenance, and quality gates. The browser currently renders—but does not cryptographically enforce—the manifest sidecar.