Crate oxillama_gguf 

oxillama-gguf

GGUF v3 binary format parser and tensor loader for OxiLLaMa.

This crate provides complete parsing of the GGUF file format, including:

• Binary header validation (magic, version)
• Typed key-value metadata extraction
• Tensor info parsing (name, shape, quantization type, offset)
• Memory-mapped tensor data access (via mmap feature)
• Full file loading with mmap or read-to-memory

Supported GGUF Versions

• Version 2 (legacy)
• Version 3 (current standard)

Quick Start

use oxillama_gguf::GgufModel;

let model = GgufModel::load("model.gguf").unwrap();
println!("Architecture: {}", model.architecture().unwrap());
println!("Tensors: {}", model.file.header.tensor_count);

Re-exports

pub use error::GgufError;

pub use error::GgufResult;

pub use header::GgufHeader;

pub use metadata::MetadataStore;

pub use metadata::MetadataValue;

pub use parser::GgufFile;

pub use reader::BinaryReader;

pub use reader_core::align_up;

pub use reader_core::parse_gguf;

pub use reader_core::ParsedGguf;

pub use source::SliceSource;

pub use source::Source;

pub use tensor_info::TensorInfo;

pub use tensor_info::TensorStore;

pub use types::GgufTensorType;

pub use types::GgufValueType;

pub use source::FileSource;std

pub use source::ReadSource;std

pub use loader::GgufModel;std

pub use safetensors::SafetensorsConverter;std

pub use quantize_on_load::QuantPlan;std

pub use quantize_on_load::QuantTarget;std

pub use resume::checkpoint_path_for;std

pub use resume::compute_fingerprint;std

pub use resume::compute_fingerprint_with_probe;std

pub use resume::load_checkpoint;std

pub use resume::save_checkpoint;std

pub use resume::validate_checkpoint;std

pub use resume::PrefixFingerprint;std

pub use resume::ResumeCheckpoint;std

pub use resume::ResumeHandle;std

pub use schema::validate_schema;std

pub use schema::SchemaValidator;std

pub use schema::SchemaViolation;std

pub use sharded::ShardedGgufModel;std

pub use streaming::StreamingGgufParser;std

pub use streaming::TensorInfoIter;std

pub use writer::GgufWriter;std

Modules

error

Error types for the GGUF parser.

header

GGUF file header parsing.

http_source

HTTP Range-request backed crate::source::Source for remote GGUF loading.

loaderstd

GGUF file loader with memory mapping support.

metadata

GGUF metadata key-value store with typed access.

parser

Complete GGUF file parser.

quantize_on_loadstd

Quantize-on-load: convert F16/F32 tensors to Q4_0 or Q8_0 while loading.

reader

Binary reader utility for safe parsing of GGUF byte streams.

reader_core

Core GGUF parse logic generic over any Source.

resumestd

Partial-download resume support for GGUF files.

safetensorsstd

Safetensors binary-format import bridge.

schemastd

Pluggable metadata schema validators.

shardedstd

Sharded GGUF model support.

source

Abstract byte-stream source for no_std-compatible GGUF parsing.

streamingstd

Streaming / lazy GGUF parser.

tensor_info

GGUF tensor information and storage.

types

Core types for the GGUF format — quantization type IDs, value types, and constants.

writerstd

GGUF v3 binary writer — serialize models to the GGUF format.