PyPI Stats
microsoft

minference

[NeurIPS'24 Spotlight, ICLR'25, ICML'25] To speed up Long-context LLMs' inference, approximate and dynamic sparse calculate the attention, which reduces inference latency by up to 10x for pre-filling on an A100 while maintaining accuracy.
PyPI GitHub
Downloads
Day
41
Week
99
Month
442
Stars
1.2K
Forks
78
Issues
83
Version
0.1.6.0

Package Insights

What's this?
Score breakdown - each component scores 0–100, weighted by percentage
Maintenance
60
Popularity
62
Adoption
26
Community
65
How the maintenance status was determined
Last commit
59 days ago
Monthly downloads
442
Active
Activity within 30 days
• Stable
Activity within 31-180 days
Minimal
Activity within 181-365 days, or no activity data
At Risk
No activity for > 365 days, but > 10,000 monthly downloads
Archived
Repository flagged as archived on GitHub
How the popularity category was determined
Monthly downloads
442
Stars
1,211
Stars per 1k downloads
2739.82
Industry Standard
> 10,000,000 monthly downloads
Hidden Gem
< 0.5 stars per 1k downloads and > 100,000 monthly downloads
Community Favorite
> 5 stars per 1k downloads
Rising Star
Balanced ratio of stars to downloads
• Unknown
Fewer than 1,000 monthly downloads, or no star data
Needs at least 1,000 monthly downloads to classify popularity.
How download momentum is calculated
Week daily avg
14
Month daily avg
15
Momentum
—
((week_daily_avg - month_daily_avg) / month_daily_avg) * 100
Momentum not reported for packages with < 5,000 monthly downloads — percentages are noisy at low volume.
Python version support tiers
requires_python
>=3.8.0
Cutting Edge
Minimum supported version is 3.12 or newer
Modern
Minimum supported version is 3.9, 3.10, or 3.11
• Legacy
Supports 3.8 or older (or Python 2)
Unknown
No requires_python specified or unparseable
License categories
License
MIT License
• Permissive
MIT, Apache, BSD, ISC, Unlicense, CC0, etc.
Copyleft
GPL, LGPL, AGPL, MPL, EPL, etc.
Proprietary
Commercial, all rights reserved, custom EULAs
Unknown
No license specified, or none of the above match
How the release cadence was determined
Last release
323 days ago
Total releases
8
Releases/year
4.4
Rapid
Last release within 365 days and > 12 releases/year
• Regular
Last release within 365 days
Slow
Last release 366-730 days ago
Stalled
Last release > 730 days ago, or no releases
Unknown
No usable release date available
How contributor risk was determined
Contributors
10,000
Monthly downloads
442
High Load
≤ 2 contributors and ≥ 50,000 monthly downloads
Moderate Load
< 10 contributors and ≥ 500,000 monthly downloads
Low Load
< 25 contributors and ≥ 5,000,000 monthly downloads, or sole maintainer with ≥ 10,000 monthly downloads
• Balanced
Enough contributors for the package's popularity
Unknown
No contributor data available
Engagement
65
 
Issue Burden
68.5
per 1k stars
Discussions
2.5
per 1k stars
Fork Ratio
12820.5%
contrib / forks
Stale Release
264d
 
PR Velocity
Quiet
 

Weekly Downloads

GitHub Stars

Not all downloads include detailed metadata. Breakdowns reflect the subset that does.

Downloads by OS

Linux 8

Python Versions

3.10
75.0% 6
3.11
25.0% 2

Top Countries

1. US US
8

Dependencies

4
16 optional dependencies

Used By

1

Package Information

License
MIT License
First Release
July 5, 2024
Last Release
10 months ago
Last Commit
1 month ago
Latest Version
0.1.6.0
Total Releases
8
Contributors
10,000
Open PRs
7
Archived
No
Author
The MInference team
Python Versions
3
Keywords
LLMs Inference Long-Context LLMs Dynamic Sparse Attention Efficient Inference
Topics
-
Tracking Since
January 14, 2026
Data Updated
May 3, 2026
Data Freshness
Fresh

Links