airllm
AirLLM runs 70B large language models on a single 4GB GPU without quantization, distillation or pruning. Kimi K3 2.8T on under 4GB, Qwen3.8-Flash-Next 125B on 6GB, DeepSeek-V3 671B on ~12GB. Train Qwen3.8-Flash-Next under 6GB.
airllm has been downloaded 314,242 times in total on PyPI, including 15,153 in the last 30 days. The latest version is 4.0.0, released Sep 5, 2026.
Version4.0.0
Downloads
314.24k
License—
AuthorGavin Li
UpdatedSep 5, 2026
Downloads
Weekly, last 90d.
Includes CI traffic.
VersionsTotal4.*3.*
Range
View
Granularity
Group by
CI traffic
Stack: OffCI: Included3 / 40 series
Selected total66.84k
314.2kAll-time
15.2kLast 30 days
441Last 24 h
<0.01/sPer second
Sponsored
Sponsorships keep pepy free to read
Version distribution
Share of downloads by released version. Computed over the last quarter.
- 0157.2%
4.0.0
20.5k downloadsDownloads20.5k57.2% - 0212.0%
3.2.0
4.3k downloadsDownloads4.3k12.0% - 037.7%
3.3.0
2.8k downloadsDownloads2.8k7.7% - 045.0%
3.1.0
1.8k downloadsDownloads1.8k5.0% - 054.1%
2.11.0
1.5k downloadsDownloads1.5k4.1% - 061.0%
3.0.1
346 downloadsDownloads3461.0% - 070.9%
3.0.0
313 downloadsDownloads3130.9% - 080.5%
2.8.3
191 downloadsDownloads1910.5% - 0911.7%
Other
4.2k downloadsDownloads4.2k11.7%
Guess the next day
Thirteen recent days of airllm downloads. Drag the green handle on the right to guess where day fourteen lands.