mteb 2.5.3__py3-none-any.whl → 2.5.4__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 (102) hide show
  1. mteb/_create_dataloaders.py +10 -15
  2. mteb/_evaluators/any_sts_evaluator.py +1 -4
  3. mteb/_evaluators/evaluator.py +2 -1
  4. mteb/_evaluators/image/imagetext_pairclassification_evaluator.py +5 -6
  5. mteb/_evaluators/pair_classification_evaluator.py +3 -1
  6. mteb/_evaluators/retrieval_metrics.py +17 -16
  7. mteb/_evaluators/sklearn_evaluator.py +9 -8
  8. mteb/_evaluators/text/bitext_mining_evaluator.py +23 -16
  9. mteb/_evaluators/text/summarization_evaluator.py +20 -16
  10. mteb/abstasks/_data_filter/filters.py +1 -1
  11. mteb/abstasks/_data_filter/task_pipelines.py +3 -0
  12. mteb/abstasks/_statistics_calculation.py +18 -10
  13. mteb/abstasks/_stratification.py +18 -18
  14. mteb/abstasks/abstask.py +27 -21
  15. mteb/abstasks/aggregate_task_metadata.py +1 -9
  16. mteb/abstasks/aggregated_task.py +3 -16
  17. mteb/abstasks/classification.py +10 -4
  18. mteb/abstasks/clustering.py +18 -14
  19. mteb/abstasks/clustering_legacy.py +8 -8
  20. mteb/abstasks/image/image_text_pair_classification.py +5 -3
  21. mteb/abstasks/multilabel_classification.py +20 -16
  22. mteb/abstasks/pair_classification.py +18 -9
  23. mteb/abstasks/regression.py +3 -3
  24. mteb/abstasks/retrieval.py +12 -9
  25. mteb/abstasks/sts.py +6 -3
  26. mteb/abstasks/task_metadata.py +20 -16
  27. mteb/abstasks/text/bitext_mining.py +36 -25
  28. mteb/abstasks/text/reranking.py +7 -5
  29. mteb/abstasks/text/summarization.py +8 -3
  30. mteb/abstasks/zeroshot_classification.py +5 -2
  31. mteb/benchmarks/benchmark.py +2 -2
  32. mteb/cache.py +20 -18
  33. mteb/cli/_display_tasks.py +2 -2
  34. mteb/cli/build_cli.py +5 -5
  35. mteb/cli/generate_model_card.py +6 -4
  36. mteb/deprecated_evaluator.py +56 -43
  37. mteb/evaluate.py +35 -29
  38. mteb/filter_tasks.py +25 -26
  39. mteb/get_tasks.py +25 -27
  40. mteb/languages/language_scripts.py +5 -3
  41. mteb/leaderboard/app.py +1 -1
  42. mteb/load_results.py +12 -12
  43. mteb/models/abs_encoder.py +2 -2
  44. mteb/models/cache_wrappers/cache_backend_protocol.py +3 -5
  45. mteb/models/cache_wrappers/cache_backends/_hash_utils.py +5 -4
  46. mteb/models/cache_wrappers/cache_backends/faiss_cache.py +2 -1
  47. mteb/models/cache_wrappers/cache_backends/numpy_cache.py +30 -13
  48. mteb/models/cache_wrappers/cache_wrapper.py +2 -2
  49. mteb/models/get_model_meta.py +8 -1
  50. mteb/models/instruct_wrapper.py +11 -5
  51. mteb/models/model_implementations/andersborges.py +2 -2
  52. mteb/models/model_implementations/blip_models.py +8 -8
  53. mteb/models/model_implementations/bm25.py +1 -1
  54. mteb/models/model_implementations/clip_models.py +3 -3
  55. mteb/models/model_implementations/cohere_models.py +1 -1
  56. mteb/models/model_implementations/cohere_v.py +2 -2
  57. mteb/models/model_implementations/dino_models.py +23 -23
  58. mteb/models/model_implementations/emillykkejensen_models.py +3 -3
  59. mteb/models/model_implementations/jina_clip.py +1 -1
  60. mteb/models/model_implementations/jina_models.py +1 -1
  61. mteb/models/model_implementations/kennethenevoldsen_models.py +2 -2
  62. mteb/models/model_implementations/llm2clip_models.py +3 -3
  63. mteb/models/model_implementations/moco_models.py +2 -2
  64. mteb/models/model_implementations/model2vec_models.py +1 -1
  65. mteb/models/model_implementations/nomic_models.py +8 -8
  66. mteb/models/model_implementations/openclip_models.py +7 -7
  67. mteb/models/model_implementations/random_baseline.py +3 -3
  68. mteb/models/model_implementations/rasgaard_models.py +1 -1
  69. mteb/models/model_implementations/repllama_models.py +2 -2
  70. mteb/models/model_implementations/rerankers_custom.py +3 -3
  71. mteb/models/model_implementations/rerankers_monot5_based.py +3 -3
  72. mteb/models/model_implementations/siglip_models.py +10 -10
  73. mteb/models/model_implementations/vlm2vec_models.py +1 -1
  74. mteb/models/model_implementations/voyage_v.py +4 -4
  75. mteb/models/model_meta.py +11 -12
  76. mteb/models/search_encoder_index/search_indexes/faiss_search_index.py +5 -5
  77. mteb/models/search_wrappers.py +22 -10
  78. mteb/models/sentence_transformer_wrapper.py +9 -4
  79. mteb/py.typed +0 -0
  80. mteb/results/benchmark_results.py +25 -19
  81. mteb/results/model_result.py +49 -21
  82. mteb/results/task_result.py +45 -51
  83. mteb/similarity_functions.py +11 -7
  84. mteb/tasks/classification/dan/dk_hate_classification.py +1 -1
  85. mteb/tasks/classification/est/estonian_valence.py +1 -1
  86. mteb/tasks/classification/multilingual/scala_classification.py +1 -1
  87. mteb/tasks/image_text_pair_classification/eng/sugar_crepe.py +1 -1
  88. mteb/tasks/retrieval/code/code_rag.py +12 -12
  89. mteb/tasks/retrieval/dan/dan_fever_retrieval.py +1 -1
  90. mteb/tasks/retrieval/dan/tv2_nordretrieval.py +2 -2
  91. mteb/tasks/retrieval/dan/twitter_hjerne_retrieval.py +2 -2
  92. mteb/tasks/retrieval/nob/norquad.py +2 -2
  93. mteb/tasks/retrieval/nob/snl_retrieval.py +2 -2
  94. mteb/tasks/retrieval/tur/tur_hist_quad.py +1 -1
  95. mteb/types/_result.py +2 -1
  96. mteb/types/statistics.py +9 -3
  97. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/METADATA +1 -1
  98. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/RECORD +102 -101
  99. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/WHEEL +0 -0
  100. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/entry_points.txt +0 -0
  101. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/licenses/LICENSE +0 -0
  102. {mteb-2.5.3.dist-info → mteb-2.5.4.dist-info}/top_level.txt +0 -0
@@ -104,7 +104,7 @@ dinov2_training_datasets = set(
104
104
 
105
105
 
106
106
  dinov2_small = ModelMeta(
107
- loader=DINOModel, # type: ignore
107
+ loader=DINOModel,
108
108
  name="facebook/dinov2-small",
109
109
  model_type=["dense"],
110
110
  languages=["eng-Latn"],
@@ -125,7 +125,7 @@ dinov2_small = ModelMeta(
125
125
  use_instructions=False,
126
126
  training_datasets=dinov2_training_datasets,
127
127
  citation="""@misc{oquab2023dinov2,
128
- title={DINOv2: Learning Robust Visual Features without Supervision},
128
+ title={DINOv2: Learning Robust Visual Features without Supervision},
129
129
  author={Maxime Oquab and Timothée Darcet and Théo Moutakanni and Huy Vo and Marc Szafraniec and Vasil Khalidov and Pierre Fernandez and Daniel Haziza and Francisco Massa and Alaaeldin El-Nouby and Mahmoud Assran and Nicolas Ballas and Wojciech Galuba and Russell Howes and Po-Yao Huang and Shang-Wen Li and Ishan Misra and Michael Rabbat and Vasu Sharma and Gabriel Synnaeve and Hu Xu and Hervé Jegou and Julien Mairal and Patrick Labatut and Armand Joulin and Piotr Bojanowski},
130
130
  year={2023},
131
131
  eprint={2304.07193},
@@ -135,7 +135,7 @@ dinov2_small = ModelMeta(
135
135
  )
136
136
 
137
137
  dinov2_base = ModelMeta(
138
- loader=DINOModel, # type: ignore
138
+ loader=DINOModel,
139
139
  name="facebook/dinov2-base",
140
140
  model_type=["dense"],
141
141
  languages=["eng-Latn"],
@@ -156,7 +156,7 @@ dinov2_base = ModelMeta(
156
156
  use_instructions=False,
157
157
  training_datasets=dinov2_training_datasets,
158
158
  citation="""@misc{oquab2023dinov2,
159
- title={DINOv2: Learning Robust Visual Features without Supervision},
159
+ title={DINOv2: Learning Robust Visual Features without Supervision},
160
160
  author={Maxime Oquab and Timothée Darcet and Théo Moutakanni and Huy Vo and Marc Szafraniec and Vasil Khalidov and Pierre Fernandez and Daniel Haziza and Francisco Massa and Alaaeldin El-Nouby and Mahmoud Assran and Nicolas Ballas and Wojciech Galuba and Russell Howes and Po-Yao Huang and Shang-Wen Li and Ishan Misra and Michael Rabbat and Vasu Sharma and Gabriel Synnaeve and Hu Xu and Hervé Jegou and Julien Mairal and Patrick Labatut and Armand Joulin and Piotr Bojanowski},
161
161
  year={2023},
162
162
  eprint={2304.07193},
@@ -166,7 +166,7 @@ dinov2_base = ModelMeta(
166
166
  )
167
167
 
168
168
  dinov2_large = ModelMeta(
169
- loader=DINOModel, # type: ignore
169
+ loader=DINOModel,
170
170
  name="facebook/dinov2-large",
171
171
  model_type=["dense"],
172
172
  languages=["eng-Latn"],
@@ -187,7 +187,7 @@ dinov2_large = ModelMeta(
187
187
  use_instructions=False,
188
188
  training_datasets=dinov2_training_datasets,
189
189
  citation="""@misc{oquab2023dinov2,
190
- title={DINOv2: Learning Robust Visual Features without Supervision},
190
+ title={DINOv2: Learning Robust Visual Features without Supervision},
191
191
  author={Maxime Oquab and Timothée Darcet and Théo Moutakanni and Huy Vo and Marc Szafraniec and Vasil Khalidov and Pierre Fernandez and Daniel Haziza and Francisco Massa and Alaaeldin El-Nouby and Mahmoud Assran and Nicolas Ballas and Wojciech Galuba and Russell Howes and Po-Yao Huang and Shang-Wen Li and Ishan Misra and Michael Rabbat and Vasu Sharma and Gabriel Synnaeve and Hu Xu and Hervé Jegou and Julien Mairal and Patrick Labatut and Armand Joulin and Piotr Bojanowski},
192
192
  year={2023},
193
193
  eprint={2304.07193},
@@ -197,7 +197,7 @@ dinov2_large = ModelMeta(
197
197
  )
198
198
 
199
199
  dinov2_giant = ModelMeta(
200
- loader=DINOModel, # type: ignore
200
+ loader=DINOModel,
201
201
  name="facebook/dinov2-giant",
202
202
  model_type=["dense"],
203
203
  languages=["eng-Latn"],
@@ -218,7 +218,7 @@ dinov2_giant = ModelMeta(
218
218
  use_instructions=False,
219
219
  training_datasets=dinov2_training_datasets,
220
220
  citation="""@misc{oquab2023dinov2,
221
- title={DINOv2: Learning Robust Visual Features without Supervision},
221
+ title={DINOv2: Learning Robust Visual Features without Supervision},
222
222
  author={Maxime Oquab and Timothée Darcet and Théo Moutakanni and Huy Vo and Marc Szafraniec and Vasil Khalidov and Pierre Fernandez and Daniel Haziza and Francisco Massa and Alaaeldin El-Nouby and Mahmoud Assran and Nicolas Ballas and Wojciech Galuba and Russell Howes and Po-Yao Huang and Shang-Wen Li and Ishan Misra and Michael Rabbat and Vasu Sharma and Gabriel Synnaeve and Hu Xu and Hervé Jegou and Julien Mairal and Patrick Labatut and Armand Joulin and Piotr Bojanowski},
223
223
  year={2023},
224
224
  eprint={2304.07193},
@@ -253,7 +253,7 @@ webssl_dino300m_full2b = ModelMeta(
253
253
  use_instructions=False,
254
254
  training_datasets=webssl_dino_training_datasets,
255
255
  citation="""@article{fan2025scaling,
256
- title={Scaling Language-Free Visual Representation Learning},
256
+ title={Scaling Language-Free Visual Representation Learning},
257
257
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
258
258
  year={2025},
259
259
  eprint={2504.01017},
@@ -284,7 +284,7 @@ webssl_dino1b_full2b = ModelMeta(
284
284
  use_instructions=False,
285
285
  training_datasets=webssl_dino_training_datasets,
286
286
  citation="""@article{fan2025scaling,
287
- title={Scaling Language-Free Visual Representation Learning},
287
+ title={Scaling Language-Free Visual Representation Learning},
288
288
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
289
289
  year={2025},
290
290
  eprint={2504.01017},
@@ -315,7 +315,7 @@ webssl_dino2b_full2b = ModelMeta(
315
315
  use_instructions=False,
316
316
  training_datasets=webssl_dino_training_datasets,
317
317
  citation="""@article{fan2025scaling,
318
- title={Scaling Language-Free Visual Representation Learning},
318
+ title={Scaling Language-Free Visual Representation Learning},
319
319
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
320
320
  year={2025},
321
321
  eprint={2504.01017},
@@ -346,7 +346,7 @@ webssl_dino3b_full2b = ModelMeta(
346
346
  use_instructions=False,
347
347
  training_datasets=webssl_dino_training_datasets,
348
348
  citation="""@article{fan2025scaling,
349
- title={Scaling Language-Free Visual Representation Learning},
349
+ title={Scaling Language-Free Visual Representation Learning},
350
350
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
351
351
  year={2025},
352
352
  eprint={2504.01017},
@@ -377,7 +377,7 @@ webssl_dino5b_full2b = ModelMeta(
377
377
  use_instructions=False,
378
378
  training_datasets=webssl_dino_training_datasets,
379
379
  citation="""@article{fan2025scaling,
380
- title={Scaling Language-Free Visual Representation Learning},
380
+ title={Scaling Language-Free Visual Representation Learning},
381
381
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
382
382
  year={2025},
383
383
  eprint={2504.01017},
@@ -408,7 +408,7 @@ webssl_dino7b_full8b_224 = ModelMeta(
408
408
  use_instructions=False,
409
409
  training_datasets=webssl_dino_training_datasets,
410
410
  citation="""@article{fan2025scaling,
411
- title={Scaling Language-Free Visual Representation Learning},
411
+ title={Scaling Language-Free Visual Representation Learning},
412
412
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
413
413
  year={2025},
414
414
  eprint={2504.01017},
@@ -439,7 +439,7 @@ webssl_dino7b_full8b_378 = ModelMeta(
439
439
  use_instructions=False,
440
440
  training_datasets=webssl_dino_training_datasets,
441
441
  citation="""@article{fan2025scaling,
442
- title={Scaling Language-Free Visual Representation Learning},
442
+ title={Scaling Language-Free Visual Representation Learning},
443
443
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
444
444
  year={2025},
445
445
  eprint={2504.01017},
@@ -470,7 +470,7 @@ webssl_dino7b_full8b_518 = ModelMeta(
470
470
  use_instructions=False,
471
471
  training_datasets=webssl_dino_training_datasets,
472
472
  citation="""@article{fan2025scaling,
473
- title={Scaling Language-Free Visual Representation Learning},
473
+ title={Scaling Language-Free Visual Representation Learning},
474
474
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
475
475
  year={2025},
476
476
  eprint={2504.01017},
@@ -502,7 +502,7 @@ webssl_dino2b_light2b = ModelMeta(
502
502
  use_instructions=False,
503
503
  training_datasets=webssl_dino_training_datasets,
504
504
  citation="""@article{fan2025scaling,
505
- title={Scaling Language-Free Visual Representation Learning},
505
+ title={Scaling Language-Free Visual Representation Learning},
506
506
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
507
507
  year={2025},
508
508
  eprint={2504.01017},
@@ -533,7 +533,7 @@ webssl_dino2b_heavy2b = ModelMeta(
533
533
  use_instructions=False,
534
534
  training_datasets=webssl_dino_training_datasets,
535
535
  citation="""@article{fan2025scaling,
536
- title={Scaling Language-Free Visual Representation Learning},
536
+ title={Scaling Language-Free Visual Representation Learning},
537
537
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
538
538
  year={2025},
539
539
  eprint={2504.01017},
@@ -564,7 +564,7 @@ webssl_dino3b_light2b = ModelMeta(
564
564
  use_instructions=False,
565
565
  training_datasets=webssl_dino_training_datasets,
566
566
  citation="""@article{fan2025scaling,
567
- title={Scaling Language-Free Visual Representation Learning},
567
+ title={Scaling Language-Free Visual Representation Learning},
568
568
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
569
569
  year={2025},
570
570
  eprint={2504.01017},
@@ -595,7 +595,7 @@ webssl_dino3b_heavy2b = ModelMeta(
595
595
  use_instructions=False,
596
596
  training_datasets=webssl_dino_training_datasets,
597
597
  citation="""@article{fan2025scaling,
598
- title={Scaling Language-Free Visual Representation Learning},
598
+ title={Scaling Language-Free Visual Representation Learning},
599
599
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
600
600
  year={2025},
601
601
  eprint={2504.01017},
@@ -626,7 +626,7 @@ webssl_mae300m_full2b = ModelMeta(
626
626
  use_instructions=False,
627
627
  training_datasets=webssl_dino_training_datasets,
628
628
  citation="""@article{fan2025scaling,
629
- title={Scaling Language-Free Visual Representation Learning},
629
+ title={Scaling Language-Free Visual Representation Learning},
630
630
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
631
631
  year={2025},
632
632
  eprint={2504.01017},
@@ -657,7 +657,7 @@ webssl_mae700m_full2b = ModelMeta(
657
657
  use_instructions=False,
658
658
  training_datasets=webssl_dino_training_datasets,
659
659
  citation="""@article{fan2025scaling,
660
- title={Scaling Language-Free Visual Representation Learning},
660
+ title={Scaling Language-Free Visual Representation Learning},
661
661
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
662
662
  year={2025},
663
663
  eprint={2504.01017},
@@ -688,7 +688,7 @@ webssl_mae1b_full2b = ModelMeta(
688
688
  use_instructions=False,
689
689
  training_datasets=webssl_dino_training_datasets,
690
690
  citation="""@article{fan2025scaling,
691
- title={Scaling Language-Free Visual Representation Learning},
691
+ title={Scaling Language-Free Visual Representation Learning},
692
692
  author={David Fan and Shengbang Tong and Jiachen Zhu and Koustuv Sinha and Zhuang Liu and Xinlei Chen and Michael Rabbat and Nicolas Ballas and Yann LeCun and Amir Bar and Saining Xie},
693
693
  year={2025},
694
694
  eprint={2504.01017},
@@ -2,7 +2,7 @@ from mteb.models.model_meta import ModelMeta
2
2
  from mteb.models.sentence_transformer_wrapper import sentence_transformers_loader
3
3
 
4
4
  embedding_gemma_300m_scandi = ModelMeta(
5
- loader=sentence_transformers_loader, # type: ignore
5
+ loader=sentence_transformers_loader,
6
6
  name="emillykkejensen/EmbeddingGemma-Scandi-300m",
7
7
  model_type=["dense"],
8
8
  languages=["dan-Latn", "swe-Latn", "nor-Latn", "nob-Latn", "nno-Latn"],
@@ -35,7 +35,7 @@ embedding_gemma_300m_scandi = ModelMeta(
35
35
 
36
36
 
37
37
  qwen_scandi = ModelMeta(
38
- loader=sentence_transformers_loader, # type: ignore
38
+ loader=sentence_transformers_loader,
39
39
  name="emillykkejensen/Qwen3-Embedding-Scandi-0.6B",
40
40
  model_type=["dense"],
41
41
  languages=["dan-Latn", "swe-Latn", "nor-Latn", "nob-Latn", "nno-Latn"],
@@ -59,7 +59,7 @@ qwen_scandi = ModelMeta(
59
59
 
60
60
 
61
61
  mmbert_scandi = ModelMeta(
62
- loader=sentence_transformers_loader, # type: ignore
62
+ loader=sentence_transformers_loader,
63
63
  name="emillykkejensen/mmBERTscandi-base-embedding",
64
64
  model_type=["dense"],
65
65
  languages=["dan-Latn", "swe-Latn", "nor-Latn", "nob-Latn", "nno-Latn"],
@@ -121,7 +121,7 @@ class JinaCLIPModel(AbsEncoder):
121
121
 
122
122
 
123
123
  jina_clip_v1 = ModelMeta(
124
- loader=JinaCLIPModel, # type: ignore
124
+ loader=JinaCLIPModel,
125
125
  name="jinaai/jina-clip-v1",
126
126
  model_type=["dense"],
127
127
  languages=["eng-Latn"],
@@ -795,7 +795,7 @@ jina_embeddings_v4 = ModelMeta(
795
795
 
796
796
 
797
797
  jina_embeddings_v3 = ModelMeta(
798
- loader=JinaWrapper, # type: ignore
798
+ loader=JinaWrapper,
799
799
  loader_kwargs=dict(
800
800
  trust_remote_code=True,
801
801
  model_prompts={
@@ -4,7 +4,7 @@ from mteb.models.sentence_transformer_wrapper import (
4
4
  )
5
5
 
6
6
  dfm_enc_large = ModelMeta(
7
- loader=sentence_transformers_loader, # type: ignore
7
+ loader=sentence_transformers_loader,
8
8
  name="KennethEnevoldsen/dfm-sentence-encoder-large",
9
9
  model_type=["dense"],
10
10
  languages=["dan-Latn"],
@@ -39,7 +39,7 @@ dfm_enc_large = ModelMeta(
39
39
  )
40
40
 
41
41
  dfm_enc_med = ModelMeta(
42
- loader=sentence_transformers_loader, # type: ignore
42
+ loader=sentence_transformers_loader,
43
43
  name="KennethEnevoldsen/dfm-sentence-encoder-medium",
44
44
  model_type=["dense"],
45
45
  languages=["dan-Latn"],
@@ -181,7 +181,7 @@ llm2clip_training_sets = set(
181
181
  )
182
182
 
183
183
  llm2clip_openai_l_14_336 = ModelMeta(
184
- loader=llm2clip_loader, # type: ignore
184
+ loader=llm2clip_loader,
185
185
  name="microsoft/LLM2CLIP-Openai-L-14-336",
186
186
  model_type=["dense"],
187
187
  languages=["eng-Latn"],
@@ -206,7 +206,7 @@ llm2clip_openai_l_14_336 = ModelMeta(
206
206
 
207
207
  # NOTE: https://huggingface.co/microsoft/LLM2CLIP-Openai-L-14-224/discussions/1
208
208
  llm2clip_openai_l_14_224 = ModelMeta(
209
- loader=llm2clip_loader, # type: ignore
209
+ loader=llm2clip_loader,
210
210
  name="microsoft/LLM2CLIP-Openai-L-14-224",
211
211
  model_type=["dense"],
212
212
  languages=["eng-Latn"],
@@ -230,7 +230,7 @@ llm2clip_openai_l_14_224 = ModelMeta(
230
230
  )
231
231
 
232
232
  llm2clip_openai_b_16 = ModelMeta(
233
- loader=llm2clip_loader, # type: ignore
233
+ loader=llm2clip_loader,
234
234
  name="microsoft/LLM2CLIP-Openai-B-16",
235
235
  model_type=["dense"],
236
236
  languages=["eng-Latn"],
@@ -117,7 +117,7 @@ mocov3_training_datasets = set(
117
117
  )
118
118
 
119
119
  mocov3_vit_base = ModelMeta(
120
- loader=mocov3_loader, # type: ignore
120
+ loader=mocov3_loader,
121
121
  name="nyu-visionx/moco-v3-vit-b",
122
122
  model_type=["dense"],
123
123
  languages=["eng-Latn"],
@@ -141,7 +141,7 @@ mocov3_vit_base = ModelMeta(
141
141
  )
142
142
 
143
143
  mocov3_vit_large = ModelMeta(
144
- loader=mocov3_loader, # type: ignore
144
+ loader=mocov3_loader,
145
145
  name="nyu-visionx/moco-v3-vit-l",
146
146
  model_type=["dense"],
147
147
  languages=["eng-Latn"],
@@ -139,7 +139,7 @@ class Model2VecModel(AbsEncoder):
139
139
  **kwargs: Additional arguments to pass to the wrapper.
140
140
  """
141
141
  requires_package(self, "model2vec", model_name, "pip install 'mteb[model2vec]'")
142
- from model2vec import StaticModel # type: ignore
142
+ from model2vec import StaticModel
143
143
 
144
144
  self.model_name = model_name
145
145
  self.model = StaticModel.from_pretrained(self.model_name)
@@ -193,7 +193,7 @@ NOMIC_CITATION = """
193
193
  """
194
194
 
195
195
  nomic_embed_v1_5 = ModelMeta(
196
- loader=NomicWrapper, # type: ignore
196
+ loader=NomicWrapper,
197
197
  loader_kwargs=dict(
198
198
  trust_remote_code=True,
199
199
  model_prompts=model_prompts,
@@ -222,7 +222,7 @@ nomic_embed_v1_5 = ModelMeta(
222
222
  )
223
223
 
224
224
  nomic_embed_v1 = ModelMeta(
225
- loader=NomicWrapper, # type: ignore
225
+ loader=NomicWrapper,
226
226
  loader_kwargs=dict(
227
227
  trust_remote_code=True,
228
228
  model_prompts=model_prompts,
@@ -251,7 +251,7 @@ nomic_embed_v1 = ModelMeta(
251
251
  )
252
252
 
253
253
  nomic_embed_v1_ablated = ModelMeta(
254
- loader=NomicWrapper, # type: ignore
254
+ loader=NomicWrapper,
255
255
  loader_kwargs=dict(
256
256
  trust_remote_code=True,
257
257
  model_prompts=model_prompts,
@@ -279,7 +279,7 @@ nomic_embed_v1_ablated = ModelMeta(
279
279
  )
280
280
 
281
281
  nomic_embed_v1_unsupervised = ModelMeta(
282
- loader=NomicWrapper, # type: ignore
282
+ loader=NomicWrapper,
283
283
  loader_kwargs=dict(
284
284
  trust_remote_code=True,
285
285
  model_prompts=model_prompts,
@@ -334,7 +334,7 @@ nomic_modern_bert_embed = ModelMeta(
334
334
  training_datasets=nomic_training_data,
335
335
  public_training_data=None,
336
336
  citation="""@misc{nussbaum2024nomic,
337
- title={Nomic Embed: Training a Reproducible Long Context Text Embedder},
337
+ title={Nomic Embed: Training a Reproducible Long Context Text Embedder},
338
338
  author={Zach Nussbaum and John X. Morris and Brandon Duderstadt and Andriy Mulyar},
339
339
  year={2024},
340
340
  eprint={2402.01613},
@@ -446,7 +446,7 @@ m_languages = [
446
446
  ]
447
447
 
448
448
  nomic_embed_text_v2_moe = ModelMeta(
449
- loader=NomicWrapper, # type: ignore
449
+ loader=NomicWrapper,
450
450
  loader_kwargs=dict(
451
451
  trust_remote_code=True,
452
452
  model_prompts=model_prompts,
@@ -472,12 +472,12 @@ nomic_embed_text_v2_moe = ModelMeta(
472
472
  training_datasets=None, # did not look into this further
473
473
  superseded_by=None,
474
474
  citation="""@misc{nussbaum2025trainingsparsemixtureexperts,
475
- title={Training Sparse Mixture Of Experts Text Embedding Models},
475
+ title={Training Sparse Mixture Of Experts Text Embedding Models},
476
476
  author={Zach Nussbaum and Brandon Duderstadt},
477
477
  year={2025},
478
478
  eprint={2502.07972},
479
479
  archivePrefix={arXiv},
480
480
  primaryClass={cs.CL},
481
- url={https://arxiv.org/abs/2502.07972},
481
+ url={https://arxiv.org/abs/2502.07972},
482
482
  }""",
483
483
  )
@@ -120,7 +120,7 @@ def openclip_loader(model_name, **kwargs):
120
120
 
121
121
 
122
122
  CLIP_ViT_L_14_DataComp_XL_s13B_b90K = ModelMeta(
123
- loader=openclip_loader, # type: ignore
123
+ loader=openclip_loader,
124
124
  name="laion/CLIP-ViT-L-14-DataComp.XL-s13B-b90K",
125
125
  model_type=["dense"],
126
126
  languages=["eng-Latn"],
@@ -146,7 +146,7 @@ CLIP_ViT_L_14_DataComp_XL_s13B_b90K = ModelMeta(
146
146
  )
147
147
 
148
148
  CLIP_ViT_B_32_DataComp_XL_s13B_b90K = ModelMeta(
149
- loader=openclip_loader, # type: ignore
149
+ loader=openclip_loader,
150
150
  name="laion/CLIP-ViT-B-32-DataComp.XL-s13B-b90K",
151
151
  model_type=["dense"],
152
152
  languages=["eng-Latn"],
@@ -172,7 +172,7 @@ CLIP_ViT_B_32_DataComp_XL_s13B_b90K = ModelMeta(
172
172
  )
173
173
 
174
174
  CLIP_ViT_B_16_DataComp_XL_s13B_b90K = ModelMeta(
175
- loader=openclip_loader, # type: ignore
175
+ loader=openclip_loader,
176
176
  name="laion/CLIP-ViT-B-16-DataComp.XL-s13B-b90K",
177
177
  model_type=["dense"],
178
178
  languages=["eng-Latn"],
@@ -198,7 +198,7 @@ CLIP_ViT_B_16_DataComp_XL_s13B_b90K = ModelMeta(
198
198
  )
199
199
 
200
200
  CLIP_ViT_bigG_14_laion2B_39B_b160k = ModelMeta(
201
- loader=openclip_loader, # type: ignore
201
+ loader=openclip_loader,
202
202
  name="laion/CLIP-ViT-bigG-14-laion2B-39B-b160k",
203
203
  model_type=["dense"],
204
204
  languages=["eng-Latn"],
@@ -224,7 +224,7 @@ CLIP_ViT_bigG_14_laion2B_39B_b160k = ModelMeta(
224
224
  )
225
225
 
226
226
  CLIP_ViT_g_14_laion2B_s34B_b88K = ModelMeta(
227
- loader=openclip_loader, # type: ignore
227
+ loader=openclip_loader,
228
228
  name="laion/CLIP-ViT-g-14-laion2B-s34B-b88K",
229
229
  model_type=["dense"],
230
230
  languages=["eng-Latn"],
@@ -250,7 +250,7 @@ CLIP_ViT_g_14_laion2B_s34B_b88K = ModelMeta(
250
250
  )
251
251
 
252
252
  CLIP_ViT_H_14_laion2B_s32B_b79K = ModelMeta(
253
- loader=openclip_loader, # type: ignore
253
+ loader=openclip_loader,
254
254
  name="laion/CLIP-ViT-H-14-laion2B-s32B-b79K",
255
255
  model_type=["dense"],
256
256
  languages=["eng-Latn"],
@@ -276,7 +276,7 @@ CLIP_ViT_H_14_laion2B_s32B_b79K = ModelMeta(
276
276
  )
277
277
 
278
278
  CLIP_ViT_L_14_laion2B_s32B_b82K = ModelMeta(
279
- loader=openclip_loader, # type: ignore
279
+ loader=openclip_loader,
280
280
  name="laion/CLIP-ViT-L-14-laion2B-s32B-b82K",
281
281
  model_type=["dense"],
282
282
  languages=["eng-Latn"],
@@ -68,7 +68,7 @@ _common_mock_metadata = dict(
68
68
  license="mit",
69
69
  max_tokens=np.inf,
70
70
  reference=None,
71
- similarity_fn_name="cosine", # type: ignore
71
+ similarity_fn_name="cosine",
72
72
  framework=[],
73
73
  use_instructions=False,
74
74
  public_training_code=None, # No training code, as this is a random baseline
@@ -187,7 +187,7 @@ class RandomEncoderBaseline:
187
187
 
188
188
 
189
189
  random_encoder_baseline = ModelMeta(
190
- loader=RandomEncoderBaseline, # type: ignore
190
+ loader=RandomEncoderBaseline,
191
191
  name="baseline/random-encoder-baseline",
192
192
  model_type=["dense"],
193
193
  modalities=["text", "image"],
@@ -232,7 +232,7 @@ class RandomCrossEncoderBaseline:
232
232
 
233
233
 
234
234
  random_cross_encoder_baseline = ModelMeta(
235
- loader=RandomCrossEncoderBaseline, # type: ignore
235
+ loader=RandomCrossEncoderBaseline,
236
236
  name="baseline/random-cross-encoder-baseline",
237
237
  model_type=["cross-encoder"],
238
238
  modalities=["text", "image"],
@@ -4,7 +4,7 @@ from mteb.models.model_implementations.model2vec_models import Model2VecModel
4
4
  from mteb.models.model_meta import ModelMeta, ScoringFunction
5
5
 
6
6
  potion_base_8m = ModelMeta(
7
- loader=Model2VecModel, # type: ignore
7
+ loader=Model2VecModel,
8
8
  name="rasgaard/m2v-dfm-large",
9
9
  model_type=["dense"],
10
10
  languages=["dan-Latn"],
@@ -154,7 +154,7 @@ REPLLAMA_CITATION = """
154
154
  """
155
155
 
156
156
  repllama_llama2_original = ModelMeta(
157
- loader=RepLLaMAModel, # type: ignore
157
+ loader=RepLLaMAModel,
158
158
  loader_kwargs=dict(
159
159
  base_model_name_or_path="meta-llama/Llama-2-7b-hf",
160
160
  device_map="auto",
@@ -187,7 +187,7 @@ repllama_llama2_original = ModelMeta(
187
187
 
188
188
 
189
189
  repllama_llama2_reproduced = ModelMeta(
190
- loader=RepLLaMAModel, # type: ignore
190
+ loader=RepLLaMAModel,
191
191
  loader_kwargs=dict(
192
192
  base_model_name_or_path="meta-llama/Llama-2-7b-hf",
193
193
  device_map="auto",
@@ -214,7 +214,7 @@ class JinaReranker(RerankerWrapper):
214
214
 
215
215
 
216
216
  monobert_large = ModelMeta(
217
- loader=MonoBERTReranker, # type: ignore
217
+ loader=MonoBERTReranker,
218
218
  loader_kwargs=dict(
219
219
  fp_options="float16",
220
220
  ),
@@ -239,7 +239,7 @@ monobert_large = ModelMeta(
239
239
 
240
240
  # languages unclear: https://huggingface.co/jinaai/jina-reranker-v2-base-multilingual/discussions/28
241
241
  jina_reranker_multilingual = ModelMeta(
242
- loader=JinaReranker, # type: ignore
242
+ loader=JinaReranker,
243
243
  loader_kwargs=dict(
244
244
  fp_options="float16",
245
245
  ),
@@ -263,7 +263,7 @@ jina_reranker_multilingual = ModelMeta(
263
263
  )
264
264
 
265
265
  bge_reranker_v2_m3 = ModelMeta(
266
- loader=BGEReranker, # type: ignore
266
+ loader=BGEReranker,
267
267
  loader_kwargs=dict(
268
268
  fp_options="float16",
269
269
  ),
@@ -343,7 +343,7 @@ monot5_small = ModelMeta(
343
343
  )
344
344
 
345
345
  monot5_base = ModelMeta(
346
- loader=MonoT5Reranker, # type: ignore
346
+ loader=MonoT5Reranker,
347
347
  loader_kwargs=dict(
348
348
  fp_options="float16",
349
349
  ),
@@ -442,7 +442,7 @@ monot5_3b = ModelMeta(
442
442
  )
443
443
 
444
444
  flant5_base = ModelMeta(
445
- loader=FLANT5Reranker, # type: ignore
445
+ loader=FLANT5Reranker,
446
446
  loader_kwargs=dict(
447
447
  fp_options="float16",
448
448
  ),
@@ -902,7 +902,7 @@ mt5_base_mmarco_v2 = ModelMeta(
902
902
  )
903
903
 
904
904
  mt5_13b_mmarco_100k = ModelMeta(
905
- loader=MonoT5Reranker, # type: ignore
905
+ loader=MonoT5Reranker,
906
906
  loader_kwargs=dict(
907
907
  fp_options="float16",
908
908
  ),
@@ -123,7 +123,7 @@ siglip_training_datasets = set(
123
123
  )
124
124
 
125
125
  siglip_so400m_patch14_224 = ModelMeta(
126
- loader=SiglipModelWrapper, # type: ignore
126
+ loader=SiglipModelWrapper,
127
127
  name="google/siglip-so400m-patch14-224",
128
128
  model_type=["dense"],
129
129
  languages=["eng-Latn"],
@@ -147,7 +147,7 @@ siglip_so400m_patch14_224 = ModelMeta(
147
147
  )
148
148
 
149
149
  siglip_so400m_patch14_384 = ModelMeta(
150
- loader=SiglipModelWrapper, # type: ignore
150
+ loader=SiglipModelWrapper,
151
151
  name="google/siglip-so400m-patch14-384",
152
152
  model_type=["dense"],
153
153
  languages=["eng-Latn"],
@@ -171,7 +171,7 @@ siglip_so400m_patch14_384 = ModelMeta(
171
171
  )
172
172
 
173
173
  siglip_so400m_patch16_256_i18n = ModelMeta(
174
- loader=SiglipModelWrapper, # type: ignore
174
+ loader=SiglipModelWrapper,
175
175
  name="google/siglip-so400m-patch16-256-i18n",
176
176
  model_type=["dense"],
177
177
  languages=["eng-Latn"],
@@ -195,7 +195,7 @@ siglip_so400m_patch16_256_i18n = ModelMeta(
195
195
  )
196
196
 
197
197
  siglip_base_patch16_256_multilingual = ModelMeta(
198
- loader=SiglipModelWrapper, # type: ignore
198
+ loader=SiglipModelWrapper,
199
199
  name="google/siglip-base-patch16-256-multilingual",
200
200
  model_type=["dense"],
201
201
  languages=["eng-Latn"],
@@ -219,7 +219,7 @@ siglip_base_patch16_256_multilingual = ModelMeta(
219
219
  )
220
220
 
221
221
  siglip_base_patch16_256 = ModelMeta(
222
- loader=SiglipModelWrapper, # type: ignore
222
+ loader=SiglipModelWrapper,
223
223
  name="google/siglip-base-patch16-256",
224
224
  model_type=["dense"],
225
225
  languages=["eng-Latn"],
@@ -243,7 +243,7 @@ siglip_base_patch16_256 = ModelMeta(
243
243
  )
244
244
 
245
245
  siglip_base_patch16_512 = ModelMeta(
246
- loader=SiglipModelWrapper, # type: ignore
246
+ loader=SiglipModelWrapper,
247
247
  name="google/siglip-base-patch16-512",
248
248
  model_type=["dense"],
249
249
  languages=["eng-Latn"],
@@ -267,7 +267,7 @@ siglip_base_patch16_512 = ModelMeta(
267
267
  )
268
268
 
269
269
  siglip_base_patch16_384 = ModelMeta(
270
- loader=SiglipModelWrapper, # type: ignore
270
+ loader=SiglipModelWrapper,
271
271
  name="google/siglip-base-patch16-384",
272
272
  model_type=["dense"],
273
273
  languages=["eng-Latn"],
@@ -291,7 +291,7 @@ siglip_base_patch16_384 = ModelMeta(
291
291
  )
292
292
 
293
293
  siglip_base_patch16_224 = ModelMeta(
294
- loader=SiglipModelWrapper, # type: ignore
294
+ loader=SiglipModelWrapper,
295
295
  name="google/siglip-base-patch16-224",
296
296
  model_type=["dense"],
297
297
  languages=["eng-Latn"],
@@ -315,7 +315,7 @@ siglip_base_patch16_224 = ModelMeta(
315
315
  )
316
316
 
317
317
  siglip_large_patch16_256 = ModelMeta(
318
- loader=SiglipModelWrapper, # type: ignore
318
+ loader=SiglipModelWrapper,
319
319
  name="google/siglip-large-patch16-256",
320
320
  model_type=["dense"],
321
321
  languages=["eng-Latn"],
@@ -339,7 +339,7 @@ siglip_large_patch16_256 = ModelMeta(
339
339
  )
340
340
 
341
341
  siglip_large_patch16_384 = ModelMeta(
342
- loader=SiglipModelWrapper, # type: ignore
342
+ loader=SiglipModelWrapper,
343
343
  name="google/siglip-large-patch16-384",
344
344
  model_type=["dense"],
345
345
  languages=["eng-Latn"],
@@ -41,7 +41,7 @@ class VLM2VecWrapper(AbsEncoder):
41
41
  model_name,
42
42
  "pip install flash-attn --no-build-isolation",
43
43
  ):
44
- import flash_attn # noqa
44
+ pass
45
45
 
46
46
  requires_package(self, "peft", model_name, "pip install 'mteb[peft]'")
47
47
  from peft import LoraConfig, PeftModel