gliformer 0.1.0__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (108) hide show
  1. gliformer/__init__.py +96 -0
  2. gliformer/backbones/__init__.py +161 -0
  3. gliformer/backbones/deberta_2d.py +898 -0
  4. gliformer/backbones/flash_deberta.py +231 -0
  5. gliformer/backbones/qwen3.py +483 -0
  6. gliformer/backbones/qwen3_5.py +366 -0
  7. gliformer/config.py +3141 -0
  8. gliformer/encoders/__init__.py +56 -0
  9. gliformer/encoders/audio.py +424 -0
  10. gliformer/encoders/base.py +634 -0
  11. gliformer/encoders/media.py +311 -0
  12. gliformer/encoders/omni.py +879 -0
  13. gliformer/encoders/text.py +122 -0
  14. gliformer/encoders/vision.py +425 -0
  15. gliformer/gliformer.py +2139 -0
  16. gliformer/layers/__init__.py +123 -0
  17. gliformer/layers/anchor_layer.py +970 -0
  18. gliformer/layers/anchor_modeling.py +341 -0
  19. gliformer/layers/anchor_normalization.py +335 -0
  20. gliformer/layers/anchor_relations.py +258 -0
  21. gliformer/layers/anchored_scorer.py +78 -0
  22. gliformer/layers/attention.py +136 -0
  23. gliformer/layers/attention_bias.py +601 -0
  24. gliformer/layers/groups.py +1222 -0
  25. gliformer/layers/mlp.py +55 -0
  26. gliformer/layers/pair_rep.py +136 -0
  27. gliformer/layers/pooling.py +77 -0
  28. gliformer/layers/position.py +1383 -0
  29. gliformer/layers/positional_layer.py +66 -0
  30. gliformer/layers/rnn.py +29 -0
  31. gliformer/layers/rotary.py +59 -0
  32. gliformer/layers/structuring_relations.py +404 -0
  33. gliformer/model.py +3414 -0
  34. gliformer/outputs.py +135 -0
  35. gliformer/processing/__init__.py +5 -0
  36. gliformer/processing/_structuring_alignment.py +1462 -0
  37. gliformer/processing/_structuring_hierarchy.py +892 -0
  38. gliformer/processing/collator.py +241 -0
  39. gliformer/processing/decoder.py +133 -0
  40. gliformer/processing/formatting.py +574 -0
  41. gliformer/processing/label_augmentation.py +496 -0
  42. gliformer/processing/mappings.py +285 -0
  43. gliformer/processing/pdf.py +589 -0
  44. gliformer/processing/processor.py +2258 -0
  45. gliformer/processing/schema.py +1115 -0
  46. gliformer/processing/structuring_decoder.py +792 -0
  47. gliformer/processing/structuring_processor.py +992 -0
  48. gliformer/processing/structuring_types.py +647 -0
  49. gliformer/tasks/__init__.py +711 -0
  50. gliformer/tasks/anchored_extraction.py +176 -0
  51. gliformer/tasks/audio/__init__.py +11 -0
  52. gliformer/tasks/audio/decoder.py +61 -0
  53. gliformer/tasks/audio/model.py +332 -0
  54. gliformer/tasks/audio/processor.py +161 -0
  55. gliformer/tasks/box_ops.py +141 -0
  56. gliformer/tasks/classification/__init__.py +10 -0
  57. gliformer/tasks/classification/decoder.py +77 -0
  58. gliformer/tasks/classification/model.py +123 -0
  59. gliformer/tasks/classification/processor.py +164 -0
  60. gliformer/tasks/classification/scorer.py +134 -0
  61. gliformer/tasks/count/__init__.py +2 -0
  62. gliformer/tasks/count/decoder.py +33 -0
  63. gliformer/tasks/count/model.py +69 -0
  64. gliformer/tasks/count/processor.py +56 -0
  65. gliformer/tasks/embedding/__init__.py +2 -0
  66. gliformer/tasks/embedding/decoder.py +23 -0
  67. gliformer/tasks/embedding/model.py +237 -0
  68. gliformer/tasks/embedding/processor.py +116 -0
  69. gliformer/tasks/joint_relex/__init__.py +2 -0
  70. gliformer/tasks/joint_relex/decoder.py +259 -0
  71. gliformer/tasks/joint_relex/model.py +1398 -0
  72. gliformer/tasks/joint_relex/processor.py +349 -0
  73. gliformer/tasks/losses.py +193 -0
  74. gliformer/tasks/matcher.py +291 -0
  75. gliformer/tasks/media.py +611 -0
  76. gliformer/tasks/media_processor.py +308 -0
  77. gliformer/tasks/ner/__init__.py +2 -0
  78. gliformer/tasks/ner/decoder.py +144 -0
  79. gliformer/tasks/ner/model.py +180 -0
  80. gliformer/tasks/ner/processor.py +590 -0
  81. gliformer/tasks/open_relex/__init__.py +11 -0
  82. gliformer/tasks/open_relex/decoder.py +444 -0
  83. gliformer/tasks/open_relex/model.py +734 -0
  84. gliformer/tasks/open_relex/processor.py +637 -0
  85. gliformer/tasks/span_decoder.py +324 -0
  86. gliformer/tasks/span_processor.py +378 -0
  87. gliformer/tasks/structuring/__init__.py +11 -0
  88. gliformer/tasks/structuring/decoder.py +189 -0
  89. gliformer/tasks/structuring/model.py +1073 -0
  90. gliformer/tasks/structuring/processor.py +13 -0
  91. gliformer/tasks/vision/__init__.py +13 -0
  92. gliformer/tasks/vision/decoder.py +84 -0
  93. gliformer/tasks/vision/model.py +1081 -0
  94. gliformer/tasks/vision/processor.py +219 -0
  95. gliformer/training.py +513 -0
  96. gliformer/utils.py +18 -0
  97. gliformer-0.1.0.dist-info/METADATA +408 -0
  98. gliformer-0.1.0.dist-info/RECORD +108 -0
  99. gliformer-0.1.0.dist-info/WHEEL +5 -0
  100. gliformer-0.1.0.dist-info/top_level.txt +2 -0
  101. gliformer_eval/__init__.py +1 -0
  102. gliformer_eval/eval_box_accuracy.py +105 -0
  103. gliformer_eval/eval_classification.py +483 -0
  104. gliformer_eval/eval_ner.py +414 -0
  105. gliformer_eval/eval_qa.py +684 -0
  106. gliformer_eval/eval_relex.py +1105 -0
  107. gliformer_eval/eval_structuring.py +1310 -0
  108. gliformer_eval/eval_sts.py +402 -0
gliformer/__init__.py ADDED
@@ -0,0 +1,96 @@
1
+ try:
2
+ import wcwidth
3
+
4
+ if not hasattr(wcwidth, "wcswidth"):
5
+ wcwidth.wcswidth = lambda text: sum( # type: ignore[attr-defined]
6
+ max(getattr(wcwidth, "wcwidth", lambda _: 1)(char), 0)
7
+ for char in str(text)
8
+ )
9
+ if not hasattr(wcwidth, "wcwidth"):
10
+ wcwidth.wcwidth = lambda char: 1 # type: ignore[attr-defined]
11
+ except Exception:
12
+ pass
13
+
14
+ from .config import (
15
+ GLiFormerConfig,
16
+ GLiFormerAudioConfig,
17
+ TaskHeadConfig,
18
+ AnchorHeadConfig,
19
+ BaseHeadConfig,
20
+ NERHeadConfig,
21
+ ClassificationHeadConfig,
22
+ MediaClassificationHeadConfig,
23
+ ImageClassificationHeadConfig,
24
+ AudioClassificationHeadConfig,
25
+ SetPredictionHeadConfig,
26
+ GLiFormerLayoutConfig,
27
+ GLiFormerOmniConfig,
28
+ GLiFormerTextConfig,
29
+ GLiFormerVisionConfig,
30
+ ObjectDetectionHeadConfig,
31
+ SegmentationHeadConfig,
32
+ AudioSegmentationHeadConfig,
33
+ JointRelexHeadConfig,
34
+ OpenRelexHeadConfig,
35
+ StructuringModeConfig,
36
+ StructuringHeadConfig,
37
+ CountHeadConfig,
38
+ EmbeddingHeadConfig,
39
+ )
40
+ from .model import (
41
+ BaseGLiFormerModel,
42
+ GLiFormerAudioModel,
43
+ GLiFormerAudioOutput,
44
+ GLiFormerLayoutModel,
45
+ GLiFormerLayoutOutput,
46
+ GLiFormerOmniModel,
47
+ GLiFormerOmniOutput,
48
+ GLiFormerModel,
49
+ GLiFormerOutput,
50
+ GLiFormerTextModel,
51
+ GLiFormerTextOutput,
52
+ GLiFormerVisionModel,
53
+ GLiFormerVisionOutput,
54
+ resolve_gliformer_model_class,
55
+ )
56
+ from .processing.processor import GLiFormerProcessor
57
+ from .processing.pdf import GLiFormerPDFProcessor, PDFTableProcessor
58
+ from .processing.decoder import GLiFormerDecoder
59
+ from .processing.collator import (
60
+ BaseGLiFormerDataCollator,
61
+ GLiFormerAudioDataCollator,
62
+ GLiFormerDataCollator,
63
+ GLiFormerLayoutDataCollator,
64
+ GLiFormerOmniDataCollator,
65
+ GLiFormerTextDataCollator,
66
+ GLiFormerVisionDataCollator,
67
+ resolve_gliformer_collator_class,
68
+ )
69
+ from .processing.schema import GLiFormerSchema
70
+ from .processing.formatting import StructuringOutputFormatter, FieldType
71
+ from .processing.mappings import (
72
+ BaseClassMapping,
73
+ CatClassMapping,
74
+ ExtractionItemMapping,
75
+ ExtractionClassMapping,
76
+ StructuringItemMapping,
77
+ StructuringClassMapping,
78
+ OpenRelexItemMapping,
79
+ OpenRelexClassMapping,
80
+ VisionItemMapping,
81
+ VisionClassMapping,
82
+ BatchClassesMapping,
83
+ )
84
+ from .gliformer import (
85
+ BaseGLiFormer,
86
+ GLiFormer,
87
+ GLiFormerAudio,
88
+ GLiFormerLayout,
89
+ GLiFormerOmni,
90
+ GLiFormerText,
91
+ GLiFormerVision,
92
+ )
93
+ from .training import GLiFormerTrainer
94
+
95
+ # Backward compat alias
96
+ RelationsHeadConfig = JointRelexHeadConfig
@@ -0,0 +1,161 @@
1
+ from collections.abc import Iterable
2
+ from dataclasses import dataclass
3
+
4
+ from transformers import AutoConfig, AutoModel
5
+
6
+ from .deberta_2d import (
7
+ Deberta2DConfig,
8
+ Deberta2DModel,
9
+ LayoutDebertaConfig,
10
+ LayoutDebertaModel,
11
+ )
12
+ from .flash_deberta import (
13
+ ATTN_KERNELS,
14
+ flash_kernels_available,
15
+ is_flash_kernel,
16
+ normalize_attn_kernel,
17
+ )
18
+ from .qwen3 import GLiFormerQwen3Model, Qwen3BidirectionalModel
19
+ from .qwen3_5 import (
20
+ QWEN3_5_AVAILABLE,
21
+ GLiFormerQwen3_5Model,
22
+ GLiFormerQwen3_5TextModel,
23
+ Qwen3_5BidirectionalModel,
24
+ Qwen3_5Config,
25
+ Qwen3_5TextConfig,
26
+ )
27
+
28
+ try:
29
+ AutoConfig.register(LayoutDebertaConfig.model_type, LayoutDebertaConfig)
30
+ AutoModel.register(LayoutDebertaConfig, LayoutDebertaModel)
31
+ except ValueError:
32
+ # Transformers raises when a notebook/process imports this module twice.
33
+ pass
34
+
35
+ try:
36
+ AutoModel.register(GLiFormerQwen3Model.config_class, GLiFormerQwen3Model, exist_ok=True)
37
+ except ValueError:
38
+ pass
39
+
40
+ if QWEN3_5_AVAILABLE:
41
+ try:
42
+ AutoModel.register(Qwen3_5Config, GLiFormerQwen3_5Model, exist_ok=True)
43
+ AutoModel.register(Qwen3_5TextConfig, GLiFormerQwen3_5TextModel, exist_ok=True)
44
+ except ValueError:
45
+ pass
46
+
47
+
48
+ @dataclass(frozen=True)
49
+ class BackboneSpec:
50
+ name: str
51
+ model_class: type
52
+ config_class: type | None = None
53
+ aliases: tuple[str, ...] = ()
54
+ model_class_names: tuple[str, ...] = ()
55
+
56
+
57
+ BACKBONE_REGISTRY: dict[str, BackboneSpec] = {}
58
+
59
+
60
+ def register_backbone(
61
+ name: str,
62
+ model_class: type,
63
+ config_class: type | None = None,
64
+ aliases: Iterable[str] = (),
65
+ model_class_names: Iterable[str] = (),
66
+ ) -> BackboneSpec:
67
+ normalized = normalize_backbone_type(name)
68
+ alias_tuple = tuple(normalize_backbone_type(alias) for alias in aliases)
69
+ spec = BackboneSpec(
70
+ name=normalized,
71
+ model_class=model_class,
72
+ config_class=config_class,
73
+ aliases=alias_tuple,
74
+ model_class_names=tuple(model_class_names),
75
+ )
76
+ keys = (normalized, *alias_tuple)
77
+ seen = set()
78
+ for key in keys:
79
+ if key in seen or key in BACKBONE_REGISTRY:
80
+ raise ValueError(f"Backbone {key!r} is already registered")
81
+ seen.add(key)
82
+ for key in keys:
83
+ BACKBONE_REGISTRY[key] = spec
84
+ return spec
85
+
86
+
87
+ def normalize_backbone_type(backbone_type: str | None) -> str:
88
+ return (backbone_type or "auto").replace("-", "_").lower()
89
+
90
+
91
+ def get_backbone(backbone_type: str | None) -> BackboneSpec | None:
92
+ normalized = normalize_backbone_type(backbone_type)
93
+ if normalized == "auto":
94
+ return None
95
+ try:
96
+ return BACKBONE_REGISTRY[normalized]
97
+ except KeyError as exc:
98
+ raise ValueError(
99
+ f"Unknown GLiFormer backbone_type {backbone_type!r}. "
100
+ f"Available backbones: {', '.join(available_backbones())}"
101
+ ) from exc
102
+
103
+
104
+ def available_backbones(include_auto: bool = False) -> tuple[str, ...]:
105
+ names = sorted({spec.name for spec in BACKBONE_REGISTRY.values()})
106
+ if include_auto:
107
+ return ("auto", *names)
108
+ return tuple(names)
109
+
110
+
111
+ register_backbone(
112
+ "deberta_2d",
113
+ LayoutDebertaModel,
114
+ LayoutDebertaConfig,
115
+ aliases=("layout_deberta",),
116
+ model_class_names=("LayoutDebertaModel", "Deberta2DModel"),
117
+ )
118
+ register_backbone(
119
+ "qwen3",
120
+ GLiFormerQwen3Model,
121
+ GLiFormerQwen3Model.config_class,
122
+ aliases=("qwen3_bidirectional",),
123
+ model_class_names=("GLiFormerQwen3Model", "Qwen3BidirectionalModel"),
124
+ )
125
+ if QWEN3_5_AVAILABLE:
126
+ register_backbone(
127
+ "qwen3_5",
128
+ GLiFormerQwen3_5Model,
129
+ Qwen3_5Config,
130
+ aliases=("qwen3_5_bidirectional",),
131
+ model_class_names=("GLiFormerQwen3_5Model", "Qwen3_5BidirectionalModel"),
132
+ )
133
+ register_backbone(
134
+ "qwen3_5_text",
135
+ GLiFormerQwen3_5TextModel,
136
+ Qwen3_5TextConfig,
137
+ model_class_names=("GLiFormerQwen3_5TextModel",),
138
+ )
139
+
140
+ __all__ = [
141
+ "ATTN_KERNELS",
142
+ "BACKBONE_REGISTRY",
143
+ "BackboneSpec",
144
+ "available_backbones",
145
+ "Deberta2DConfig",
146
+ "Deberta2DModel",
147
+ "flash_kernels_available",
148
+ "get_backbone",
149
+ "GLiFormerQwen3_5Model",
150
+ "GLiFormerQwen3_5TextModel",
151
+ "GLiFormerQwen3Model",
152
+ "is_flash_kernel",
153
+ "LayoutDebertaConfig",
154
+ "LayoutDebertaModel",
155
+ "normalize_attn_kernel",
156
+ "normalize_backbone_type",
157
+ "QWEN3_5_AVAILABLE",
158
+ "Qwen3_5BidirectionalModel",
159
+ "Qwen3BidirectionalModel",
160
+ "register_backbone",
161
+ ]