README.md

metadata

license: mit

Hub Stats (Lance format)

This dataset contains Hugging Face Hub statistics in Lance format, converted from the original cfahlgren1/hub-stats dataset.

Files

• models.lance - Statistics for all models on the Hub (~2.5M rows)
• datasets.lance - Statistics for all datasets on the Hub
• spaces.lance - Statistics for all spaces on the Hub

Usage

import lance

# Load a dataset remotely
ds = lance.dataset("hf://datasets/julien-c/hub-stats-lance/datasets.lance")

# Convert to pandas
df = ds.to_table().to_pandas()

# Or query with SQL-like filters
table = ds.to_table(filter="downloads > 1000")

Example: Query datasets by author

import lance

ds = lance.dataset("hf://datasets/julien-c/hub-stats-lance/datasets.lance")
results = ds.to_table(filter="author = 'microsoft'").to_pandas()

# Sort by downloads
top = results.sort_values("downloads", ascending=False).head(10)
print(top[["id", "likes", "downloads"]])

Output:

                                       id  likes  downloads
                     microsoft/ms_marco    221      11120
  microsoft/orca-math-word-problems-200k    468       6499
    microsoft/bing_coronavirus_query_set      0       6002
                       microsoft/wiki_qa     69       5737
                  microsoft/rStar-Coder    225       3492
                  microsoft/Updesh_beta      8       3223
                       microsoft/Dayhoff      7       2922
                      microsoft/meta_woz      6       2801
                  microsoft/cats_vs_dogs     61       1883
          microsoft/IMAGE_UNDERSTANDING      6       1833

Example: Vector similarity search

import lance
import numpy as np

ds = lance.dataset("hf://datasets/julien-c/hub-stats-lance/datasets.lance")

# Get an embedding to use as query (e.g., from microsoft/ms_marco)
query_row = ds.to_table(filter="id = 'microsoft/ms_marco'").to_pandas()
query_embedding = np.array(query_row["embedding"].iloc[0])

# Find 10 nearest neighbors
results = ds.to_table(
    nearest={"column": "embedding", "q": query_embedding, "k": 10}
).to_pandas()

print(results[["id", "likes", "downloads", "_distance"]])

Output:

                                id  likes  downloads  _distance
                microsoft/ms_marco    221      11120       2.23
               jiwonii97/atalk_as3      0          0      10.61
                 AI-Art-Collab/ae5      0          1      10.85
                 wgwgwgwgw/dbbdbbd      0          9      10.90
                      1FDSFS/56803      0          8      10.94

Why Lance?

Lance is a modern columnar data format optimized for ML workflows:

• Fast random access and filtering
• Efficient for large datasets
• Native support for vector search
• Zero-copy integration with PyArrow/Pandas