xberg 1.1.5

High-performance document intelligence library for Rust. Extract text, metadata, and structured data from PDFs, Office documents, images, and 107 formats and 371 programming languages via tree-sitter code intelligence with async/sync APIs.
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
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138

//! Candle-based VLM OCR backends.
//!
//! Pure-Rust transformer OCR via the `xberg-candle-ocr` crate. This module
//! holds the `OcrBackend + Plugin` impls and the per-model configuration
//! plumbing; model code itself lives in `xberg-candle-ocr::models`.

mod config;

// Shared `ExtractedDocument` construction (metadata, tables, detected_languages)
// used by every candle-* backend below, so the same logic isn't pasted per
// backend (issue #179).
//
// Gated on the union of the concrete backends that consume it, NOT on the umbrella
// `candle-ocr` feature. The umbrella can be enabled on its own -- `xberg-cli`'s
// `candle-ocr` feature profile does exactly that -- which compiles this module with
// no backend to call into it, and every item here becomes dead code under
// `-D warnings`. Same union the device-preference helpers below use. ~keep
#[cfg(any(
    feature = "candle-trocr",
    feature = "candle-paddleocr-vl",
    all(
        not(target_arch = "wasm32"),
        any(feature = "candle-glm-ocr", feature = "candle-deepseek-ocr")
    )
))]
mod ocr_result;

#[cfg(feature = "candle-trocr")]
pub mod trocr_backend;

#[cfg(feature = "candle-paddleocr-vl")]
pub mod paddleocr_vl_backend;

#[cfg(all(feature = "candle-glm-ocr", not(target_arch = "wasm32")))]
pub mod glm_ocr_backend;

#[cfg(feature = "candle-paddleocr-vl")]
pub(crate) mod model_stager;

#[cfg(all(feature = "candle-deepseek-ocr", not(target_arch = "wasm32")))]
pub mod deepseek_ocr_backend;

pub use config::{
    CandleDevicePreference, CandleModelId, CandleOcrConfig, CandleTrocrVariant, DeepseekOcrBackendOptions,
    GlmOcrBackendOptions, GlmOcrLayoutMode, GlmOcrTaskKind, PaddleOcrVlBackendOptions, PaddleOcrVlTaskKind,
    TrocrBackendOptions,
};

#[cfg(feature = "candle-trocr")]
pub use trocr_backend::TrocrBackend;

#[cfg(feature = "candle-paddleocr-vl")]
pub use paddleocr_vl_backend::PaddleOcrVlBackend;

#[cfg(all(feature = "candle-glm-ocr", not(target_arch = "wasm32")))]
pub use glm_ocr_backend::GlmOcrBackend;

#[cfg(all(feature = "candle-deepseek-ocr", not(target_arch = "wasm32")))]
pub use deepseek_ocr_backend::DeepseekOcrBackend;

#[cfg(any(
    feature = "candle-trocr",
    feature = "candle-paddleocr-vl",
    all(
        not(target_arch = "wasm32"),
        any(feature = "candle-glm-ocr", feature = "candle-deepseek-ocr")
    )
))]
use crate::core::config::{AccelerationConfig, ExecutionProviderType, OcrConfig};
#[cfg(any(
    feature = "candle-trocr",
    feature = "candle-paddleocr-vl",
    all(
        not(target_arch = "wasm32"),
        any(feature = "candle-glm-ocr", feature = "candle-deepseek-ocr")
    )
))]
use xberg_candle_ocr::DevicePreference;

/// Resolve a candle [`DevicePreference`] from the centralised acceleration
/// config plus a validated backend-specific device override.
///
/// Precedence (highest first):
/// 1. The typed backend option's device (when present) — an explicit per-call override.
/// 2. `OcrConfig.acceleration.provider` — the central config that already
///    drives layout-detection and embeddings.
/// 3. `DevicePreference::Auto`.
///
/// The mapping from [`ExecutionProviderType`] (ORT-flavoured) to
/// [`DevicePreference`] (candle-flavoured) is:
/// - `Auto`     -> `DevicePreference::Auto`
/// - `Cpu`      -> `DevicePreference::Cpu`
/// - `Cuda`     -> `DevicePreference::Cuda`
/// - `CoreMl`   -> `DevicePreference::Metal` (Apple Neural Engine + GPU runs on Metal in candle)
/// - `TensorRt` -> `DevicePreference::Cuda` (TensorRT runs on CUDA hardware; candle has no separate TRT path)
#[cfg(any(
    feature = "candle-trocr",
    feature = "candle-paddleocr-vl",
    all(
        not(target_arch = "wasm32"),
        any(feature = "candle-glm-ocr", feature = "candle-deepseek-ocr")
    )
))]
pub(crate) fn resolve_device_preference(
    config: &OcrConfig,
    explicit: Option<CandleDevicePreference>,
) -> DevicePreference {
    if let Some(explicit) = explicit {
        return explicit.into();
    }

    if let Some(accel) = &config.acceleration {
        return device_preference_from_acceleration(accel);
    }

    DevicePreference::Auto
}

/// Map an [`AccelerationConfig`] to the candle [`DevicePreference`] taxonomy.
///
/// Lifted out of `resolve_device_preference` so the mapping is independently
/// testable and reusable from future candle backends.
#[cfg(any(
    feature = "candle-trocr",
    feature = "candle-paddleocr-vl",
    all(
        not(target_arch = "wasm32"),
        any(feature = "candle-glm-ocr", feature = "candle-deepseek-ocr")
    )
))]
fn device_preference_from_acceleration(accel: &AccelerationConfig) -> DevicePreference {
    match accel.provider {
        ExecutionProviderType::Auto => DevicePreference::Auto,
        ExecutionProviderType::Cpu => DevicePreference::Cpu,
        ExecutionProviderType::Cuda | ExecutionProviderType::TensorRt => DevicePreference::Cuda,
        ExecutionProviderType::CoreMl => DevicePreference::Metal,
    }
}