Package: Rgguf 0.1.0

Rgguf: Read 'GGUF' Model Files into 'Rfmalloc'-Backed Arrays

Reads 'GGUF' model files, the file format used by the 'llama.cpp' project (<https://github.com/ggerganov/llama.cpp>) to store large language model tensors and metadata, and exposes their tensors directly as 'Rfmalloc'-backed, file-backed ALTREP matrices and arrays. Quantized tensor types supported by the vendored 'gguflib' parser (<https://github.com/antirez/gguf-tools>) are dequantized on demand. Metadata key-value pairs and the tensor directory are exposed as ordinary R lists and data frames. A minimal writer is included to create 'GGUF' files containing 32-bit floating point tensors, primarily to support round-trip testing.

Authors:Sounkou Mahamane Toure [aut, cre], Salvatore Sanfilippo [cph], Georgi Gerganov [cph]