Hugging Face's logo

Datasets:
XciD
/
hf-model-downloads-by-origin-2022-2026

Tags:
huggingface-internal
analytics
downloads
License:
Dataset card Data Studio Files
xet
 Community
Dataset Preview
Duplicate
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 2 new columns ({'owner', 'category'}) and 5 missing columns ({'2026', '2025', '2024', '2022', '2023'}).

This happened while the csv dataset builder was generating data using

hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026/mapping.csv (at revision 1bed1b4718865d15c7f2bacc6174152e3dbfdaf6), ['hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/downloads_by_namespace_year.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/mapping.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/repo_classified.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/summary_by_category_year.csv']

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1837, in _prepare_split_single
                  writer.write_table(table)
                  ~~~~~~~~~~~~~~~~~~^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 765, in write_table
                  self._write_table(pa_table, writer_batch_size=writer_batch_size)
                  ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/arrow_writer.py", line 773, in _write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              namespace: string
              category: string
              owner: string
              -- schema metadata --
              pandas: '{"index_columns": [{"kind": "range", "name": null, "start": 0, "' + 623
              to
              {'namespace': Value('string'), '2022': Value('int64'), '2023': Value('int64'), '2024': Value('int64'), '2025': Value('int64'), '2026': Value('int64')}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1839, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
                  ...<4 lines>...
                  )
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 2 new columns ({'owner', 'category'}) and 5 missing columns ({'2026', '2025', '2024', '2022', '2023'}).
              
              This happened while the csv dataset builder was generating data using
              
              hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026/mapping.csv (at revision 1bed1b4718865d15c7f2bacc6174152e3dbfdaf6), ['hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/downloads_by_namespace_year.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/mapping.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/repo_classified.csv', 'hf://datasets/XciD/hf-model-downloads-by-origin-2022-2026@1bed1b4718865d15c7f2bacc6174152e3dbfdaf6/summary_by_category_year.csv']
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

namespace
string
2022
int64
2023
int64
2024
int64
2025
int64
2026
int64
sentence-transformers
206,297,891
315,253,801
1,960,879,690
2,747,223,348
2,667,981,194
argmaxinc
0
0
591,500
480,516,937
6,060,603,670
openai
71,466,152
317,685,957
1,230,748,505
889,049,751
468,850,155
Qwen
0
1,367,381
178,054,329
802,243,855
1,740,245,719
google
55,042,044
191,502,345
595,384,773
894,251,628
707,637,770
facebook
90,624,205
461,281,886
592,302,224
871,677,451
396,958,161
google-bert
0
0
829,807,017
1,048,458,419
473,818,495
MIT
0
1,826,604
2,167,468,817
15,007,379
3,475,353
microsoft
102,023,572
362,025,933
695,041,293
776,413,375
241,693,902
FacebookAI
0
0
685,389,456
785,440,772
365,242,271
jonatasgrosman
1,111,670
674,367,514
598,752,102
364,997,209
144,097,659
BAAI
24,358
7,432,763
620,704,338
324,622,819
669,110,522
pyannote
6,204,988
53,805,462
462,420,375
746,450,983
272,036,168
Falconsai
0
163,918
29,571,988
1,163,605,134
206,060,256
hf-internal-testing
10,706,492
195,127,033
342,257,365
290,467,459
84,430,663
bert-base-uncased
216,211,528
601,504,772
60,606,234
0
0
cross-encoder
21,267,183
23,318,726
68,833,530
201,283,779
516,878,054
dima806
0
0
1,363,707
724,289,595
84,124,509
distilbert
0
0
387,355,833
296,697,819
118,712,018
meta-llama
0
13,463,269
197,328,786
360,832,915
211,400,098
amazon
12,257
444,212
125,985,486
387,308,966
158,833,322
nesaorg
0
0
642,743,529
0
0
cardiffnlp
31,982,615
117,579,640
321,547,293
89,595,302
39,363,120
gpt2
229,183,675
249,700,607
22,142,455
0
0
laion
312,292
47,699,837
142,965,619
134,617,033
139,314,795
unsloth
0
0
26,648,246
224,571,197
204,641,929
openai-community
0
0
146,763,210
193,935,739
100,166,944
stabilityai
2,419,849
126,071,796
156,438,036
117,245,223
27,058,441
intfloat
0
10,402,356
67,631,745
159,279,352
187,542,201
MaziyarPanahi
0
0
141,711,456
215,424,341
46,772,057
Helsinki-NLP
33,845,418
70,854,025
138,196,980
103,422,390
52,842,660
autogluon
0
0
18,478,237
148,680,180
170,457,436
OpenMed
0
0
0
148,721,724
182,876,916
xlm-roberta-base
109,034,357
202,714,090
9,945,330
0
0
deepseek-ai
0
141,305
7,377,258
142,445,943
152,571,935
mistralai
0
3,139,751
117,828,212
99,318,458
73,829,897
unslothai
0
0
65,078,107
192,462,883
35,173,612
pysentimiento
4,115,081
8,835,354
249,835,926
18,723,941
6,402,963
mradermacher
0
0
61,244,163
118,409,040
94,593,782
allenai
23,040,354
59,658,751
80,011,970
64,322,087
37,431,932
google-t5
0
0
114,064,074
99,730,048
47,463,655
nvidia
795,032
2,665,568
20,869,913
78,102,289
151,254,139
roberta-base
90,879,560
126,149,921
19,431,915
0
0
Salesforce
1,415,030
26,808,777
54,465,667
99,562,508
49,969,184
prajjwal1
29,178,578
78,571,078
37,601,691
65,956,959
16,323,659
distilbert-base-uncased
76,623,128
127,029,809
13,922,050
0
0
nomic-ai
0
181,186
31,832,936
46,270,837
133,942,750
lmstudio-community
0
0
3,179,944
90,221,366
117,275,296
colbert-ir
0
421,599
34,980,022
78,889,477
92,706,549
TheBloke
0
28,808,734
65,332,345
93,221,316
18,106,656
runwayml
5,081,567
121,756,558
69,183,738
0
0
trl-internal-testing
71,181
7,767,498
33,715,643
69,953,058
78,923,832
distilbert-base-uncased-finetuned-sst-2-english
64,281,396
114,234,291
9,166,912
0
0
EleutherAI
17,972,784
55,962,638
38,189,690
31,825,088
39,210,732
mrm8488
18,378,625
30,053,790
118,462,016
10,970,649
3,659,158
jinaai
0
871,788
18,706,555
88,548,546
66,750,164
emilyalsentzer
8,392,170
55,911,823
36,467,348
43,461,798
22,834,707
tohoku-nlp
0
0
119,170,902
31,401,794
13,360,522
CompVis
15,838,823
42,367,888
60,289,379
30,361,698
6,736,477
CIDAS
60,308
17,965,072
60,453,706
65,665,371
8,321,463
deepset
34,486,678
54,588,352
23,878,528
26,830,868
9,265,911
bert-base-cased
60,309,601
79,839,553
7,631,258
0
0
dslim
12,082,959
35,029,288
50,839,905
33,548,583
14,469,583
roberta-large
36,208,206
95,701,562
9,925,859
0
0
Alibaba-NLP
0
0
24,356,042
76,642,923
37,724,122
xlm-roberta-large
14,755,866
116,305,624
1,229,179
0
0
mixedbread-ai
0
0
38,255,910
47,590,427
46,433,940
bartowski
0
0
31,412,021
59,233,420
39,775,536
answerdotai
0
0
13,827,637
77,192,335
37,362,753
MoritzLaurer
700,755
56,811,816
36,042,773
23,718,265
7,534,613
Xenova
106,646
1,703,995
15,540,027
47,960,945
55,759,960
kasparas12
0
0
120,045,244
0
0
Ashishkr
0
49,997,262
68,555,097
557,602
0
distilgpt2
61,726,201
52,517,502
2,505,996
0
0
dphn
0
0
0
57,138,788
59,538,513
patrickjohncyh
0
6,336,389
51,299,595
40,841,334
16,734,619
ibm-granite
0
0
8,256,039
59,705,819
46,214,307
stable-diffusion-v1-5
0
0
51,826,985
53,197,355
8,670,503
bigscience
2,676,372
26,178,793
52,887,828
22,413,555
8,047,434
Jean-Baptiste
70,975,817
13,254,908
16,472,246
6,063,618
2,461,634
ggml-org
0
0
8,161,880
58,847,194
40,454,990
ProsusAI
10,958,327
17,204,351
17,192,256
22,477,777
34,766,000
mlx-community
0
0
7,799,885
49,865,723
44,414,761
Systran
0
669,973
32,185,513
39,496,359
26,242,322
zai-org
0
0
0
26,171,243
72,295,792
SamLowe
0
42,441,607
43,317,627
8,386,223
4,097,910
omni-research
0
0
0
57,157,459
40,506,552
Supabase
0
1,950,188
83,506,514
7,809,033
4,049,780
thenlper
0
2,551,665
33,475,513
41,018,962
20,171,577
lllyasviel
0
21,476,888
34,637,210
31,750,626
6,903,850
hfl
71,610,113
8,862,114
5,356,692
4,456,134
3,460,288
coqui
0
328,063
6,631,789
37,120,684
48,062,487
hexgrad
0
0
0
29,035,989
61,433,066
deepakorbitshift
0
0
18,268,409
70,619,538
0
hustvl
243,613
7,574,771
9,752,875
51,683,170
18,611,023
albert-base-v2
20,484,510
65,656,126
1,629,119
0
0
llava-hf
0
0
27,063,000
30,531,934
26,938,822
NousResearch
0
47,550,648
16,280,136
12,615,352
7,021,938
RedHatAI
0
0
0
23,111,652
59,235,738
speechbrain
1,415,092
9,730,537
10,028,595
35,331,404
25,431,085
End of preview.

HF model downloads by origin country (2022–2026)

Yearly Hugging Face model download counts, aggregated by repo and by the origin country of the model, to mirror the OpenRouter-style "US vs China share" chart with HF data.

How downloads are counted

Reproduces HF's official model-download methodology (validated to ~0.4% against the public download badge):

  • Source: archived hub access logs (elastic-cloud-archive-db.all_requests_reports via Athena), 2022 → 2026.
  • A download = a GET/HEAD resolve request (status 200/302/304/307) for a trigger file: config.json, config.yaml, hyperparams.yaml, params.json, meta.yaml, or any .gguf / .mlmodel / .llamafile / .tflite / .pte.
  • Everything passes through the hub resolve endpoint before the CDN, so the hub logs capture the full picture (no CloudFront needed).

Origin attribution

Re-hosts / quantizers are credited to the original model's country, not the re-hoster:

  • unsloth/Qwen2.5-7B-GGUF → China, bartowski/Meta-Llama-3.1-8B-GGUF → US, etc.
  • Rule: trusted first-party org → its country; otherwise detect the model family in the repo name (qwen/glm/deepseek → China, llama/gemma/phi → US, mistral/flux/stable-diffusion → EU…).
  • Community/Independent = community models with no foundation-model lineage (task classifiers, individual embeddings). Legitimately no origin country.

See classify.py for the full, editable ruleset.

Files

File Description
summary_by_category_year.csv Final aggregate: downloads per category (US/China/EU/Community) per year
repo_classified.csv Per-repo (top 5000/year): repo, assigned category, downloads per year
downloads_by_namespace_year.csv Raw per-namespace downloads per year (no country attribution)
mapping.csv Namespace → category → owner (first-pass org mapping)
classify.py Org + model-family classification rules (re-run to re-map)
cat_year_origin.json Same as summary, JSON

Headline (US vs China only, % of US+China downloads)

Year US China
2023 99% 1%
2024 92% 8%
2025 86% 14%
2026* 82% 18%

* 2026 = Jan–Jun (partial). Inflection at 2024 (Qwen2/2.5, DeepSeek).

Caveats

  • 2026 is a partial year (Jan–Jun); absolute totals are lower, shares comparable.
  • Country/family mapping is best-effort and editable (classify.py). Debatable calls: all-MiniLM/mpnet/e5 → US (MiniLM/Microsoft lineage); bge → China (BAAI); gte → China (Alibaba); canonical bert/gpt2/roberta → US.
  • Top-N truncation: per-repo table keeps the top 5000 repos per year (~95%+ of volume).
Downloads last month
154
Total file size:
984 kB