llama-rs 0.16.0

A high-performance Rust implementation of llama.cpp - LLM inference engine with full GGUF support
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39

//! Model profile diagnostics.
//!
//! Given a GGUF or SafeTensors model file, compare its tensor inventory
//! and metadata against a declarative architecture profile and report
//! the deltas. The tool operates purely on tensor *descriptors*
//! (name, shape, dtype) and metadata — tensor data is never read, so
//! the classifier runs in milliseconds even on 50 GB files.
//!
//! The diagnostic path is independent of `ModelLoader`: a broken
//! loader must not block diagnostics. Profile ↔ loader consistency is
//! enforced by CI fixture tests.

pub mod compare;
pub mod error;
pub mod generate;
pub mod model_source;
pub mod profile;
pub mod render;
pub mod report;
pub mod shape;
pub mod types;

pub use compare::{compare, ReportContext};
pub use generate::generate_profile;
pub use error::DiagnosticError;
pub use model_source::{GgufSource, ModelSource, SafetensorsSource};
pub use profile::{
    find_profile_for_architecture, list_builtin_profile_names, load_builtin_profile,
    load_profile_file, Profile, SymbolSource,
};
pub use render::{render_json, render_text, MAX_TENSOR_LIST};
pub use report::{
    ArchitectureInfo, DiagnosticReport, FormatInfo, Hypothesis, HypothesisTriggers,
    MetadataDelta, MetadataDeltas, MissingTensor, ProfileRef, ResolvedSymbol,
    ShapeComparisonSkipped, ShapeMismatch, Summary, TensorPattern, UnexpectedTensor, Verdict,
    Warning, SCHEMA_VERSION,
};
pub use shape::{ShapeExpr, SymbolTable};
pub use types::{MetadataBundle, MetadataValue, TensorDtype, TensorInventory, TensorRecord};