license: other
license_name: tensorust-sample-license
license_link: https://tensorust.com/datasets/hf-model-adoption/
pretty_name: Open Model Adoption Tracker (free sample)
language:
- en
tags:
- tensorust
- sample
- agent-ai-economy
- government-data
- finance
size_categories:
- n<1K
configs:
- config_name: sample
data_files:
- split: train
path: sample.csv
- config_name: summary
data_files:
- split: train
path: summary_by_task.csv
Open Model Adoption Tracker: free sample (Tensorust LLC)
The 50,000 most-pulled open models on Hugging Face with downloads, likes and trend, rolled up by family and org: BERT family, Qwen and E5/BGE/GTE embeddings are the most-pulled named families.
Headline: A lab's own download count understates its reach: most of an open model family's pulls go to other people's fine-tunes, quantizations and merges, and nobody reports that number.
This repository is a free teaser, not the dataset. It holds 25 example rows and a summary table with no names in it. The full file has 50,000 rows and is sold by Tensorust LLC: https://tensorust.com/datasets/hf-model-adoption/.
Files here
| file | what it is |
|---|---|
sample.csv |
25 rows of the full file (the same sample shown on tensorust.com) |
summary_by_task.csv |
row counts and totals by task across the full file, no entity names (groups with fewer than 3 rows are folded into "Other") |
Summary by task (largest groups)
| Task | rows | Downloads, last 30 days |
|---|---|---|
| (blank) | 19,132 | 313,304,181 |
| text-generation | 13,895 | 430,487,850 |
| image-text-to-text | 4,137 | 315,766,154 |
| text-to-image | 1,698 | 63,387,815 |
| image-classification | 1,080 | 76,427,990 |
| automatic-speech-recognition | 1,023 | 207,391,156 |
| token-classification | 955 | 53,001,835 |
| text-classification | 913 | 64,256,024 |
| feature-extraction | 738 | 147,524,861 |
| text-to-speech | 642 | 54,666,142 |
| sentence-similarity | 617 | 362,820,092 |
| fill-mask | 553 | 107,061,547 |
| TOTAL | 50,000 | 2,636,516,536 |
Full table: summary_by_task.csv.
What the full file has
- 50,000 rows
- Key fields: model_id, org, family, base_model, derivative, task, library, downloads_30d, downloads_all_time, likes
- Parquet and CSV, a README with sources, method and limits, and the build script
- Free 25-row sample (no personal contact data)
Coverage: Top 50,000 models by 30-day downloads (Hugging Face hosts over two million); 9,182 orgs. Hugging Face API pull of Nov 6, 2026 (change columns compare with the Nov 5, 2026 pull of the top 200)
Columns in the full file
| column | meaning |
|---|---|
model_id |
Model repo |
org |
Publishing org or user |
family |
Model family |
base_model |
Declared base model |
derivative |
original, fine-tune, quantized, adapter or merge |
task |
Pipeline task |
library |
Library |
downloads_30d |
Downloads in the last 30 days |
downloads_all_time |
All-time downloads |
likes |
Likes |
trending_score |
Hugging Face trending score |
gated |
Gated access |
license |
License tag |
gguf |
Has GGUF files |
created_at |
Created |
last_modified |
Last modified |
url |
Model page |
downloads_30d_nov5 |
30-day downloads in the Nov 5 pull (top 200 only) |
likes_nov5 |
Likes on Nov 5 (top 200 only) |
downloads_30d_change_1d |
Change since Nov 5 |
likes_change_1d |
Like change since Nov 5 |
rank_downloads_30d |
Rank by 30-day downloads |
Get the full file
| tier | price | includes |
|---|---|---|
| Snapshot file | $900 | All 50,000 models and both rollups. |
| Weekly tracker | $600 a month | A fresh pull every week with week-over-week change per model, family and org. |
| Enterprise license | $12,000 a year | Daily pulls, full history, custom family definitions and use across your company. |
Email mattbusel@tensorust.com with the tier and a billing contact, and we will send a proper invoice (ACH or card). The file is delivered once the invoice is paid. Details and the live sample: https://tensorust.com/datasets/hf-model-adoption/. All datasets: https://tensorust.com/datasets/
Method and limits
Page through /api/models sorted by downloads, 1,000 at a time, with downloads, all-time downloads, likes, trending score, tags, task and library. Family comes from the base_model tag where present, else the model id, matched against named family patterns (build.py). Derivative type comes from the base_model relation tags. Rollups sum by org and by family. Change columns exist for the 200 models in the Nov 5 pull; each new pull extends the change history.
Not usage or inference volume: a download is a file pull, and CI systems and mirrors inflate some counts. Not the whole Hub (top 50,000 by downloads). Family tagging is pattern-based and can misfile a model. Not affiliated with Hugging Face.
Sources
- Hugging Face Hub API, /api/models sorted by downloads, https://huggingface.co/api/models (pulls of Nov 5 and Nov 6 2026)
License
The sample and summary here are free to view, analyze and cite with attribution to "Tensorust LLC, https://tensorust.com/datasets/hf-model-adoption/". The full dataset is licensed separately per tier. No resale or redistribution of Tensorust data without a written license.
Tensorust LLC, Tuckahoe NY. mattbusel@tensorust.com | https://tensorust.com