LeanQuant
Overview
This package provides efficient inference kernels for running non-uniformly quantized LeanQuant models on CUDA-enabled GPUs. LeanQuant is a scalable and accurate quantization algorithm that compresses large language models by 4-8x while maintaining competitive performance.
Installation
Ensure your GPU supports CUDA 11 or CUDA 12. You can check your CUDA version with the command nvidia-smi | grep CUDA.
To install:
# For CUDA 11.x
pip install leanquant[cuda11]
# For CUDA 12.x
pip install leanquant[cuda12]
Models
Quantized LeanQuant models are available for download on our HuggingFace page: huggingface.co/LeanQuant
Technical Details
LeanQuant introduces an innovative loss-error-aware grid approach to quantization that significantly outperforms traditional methods. Our technique:
- Achieves superior compression ratios: Reduces model size by 4-8x without sacrificing capability
- Preserves model intelligence: Maintains performance comparable to full-precision models across challenging benchmarks
- Optimizes GPU execution: Features custom CUDA kernels specifically designed for non-uniform quantization format
The algorithm strategically allocates quantization precision based on parameter sensitivity, ensuring computational resources are focused where they matter most.
For a comprehensive explanation of our methodology and benchmark results, please refer to our research paper.
Citation
If you find LeanQuant useful in your research or applications, please consider citing our work:
@inproceedings{
zhang2025leanquant,
title={LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid},
author={Tianyi Zhang and Anshumali Shrivastava},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=ISqx8giekS}
}
Metadata
Release files for leanquant 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| leanquant-0.1.1.tar.gz | 5.3 kB | Without provenance | Details for leanquant-0.1.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| leanquant-0.1.1-py3-none-any.whl | Python 3 | none | any | Details for leanquant-0.1.1-py3-none-any.whl |
Total release size: 10.9 kB
Release files / leanquant-0.1.1.tar.gz
| Download URL | leanquant-0.1.1.tar.gz |
|---|---|
| Size | 5.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
5b269964faa392adffdd07f07528debecbe43ba27d9b64f0ae149a6450068c35
|
|
BLAKE2b-256 checksum How to use checksums |
e70eee91cad8fff5cdc9c543f17eba09c80a374ebcdf5836662dfbeb737d4f57
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.10.16
|
Release files / leanquant-0.1.1-py3-none-any.whl
| Download URL | leanquant-0.1.1-py3-none-any.whl |
|---|---|
| Size | 5.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ba123232c54910ae7985cbf3d6109133292d42fca7b72710535454760a961b61
|
|
BLAKE2b-256 checksum How to use checksums |
83212839ba4a349c27a971ecae978986949a0180f06c1ceec806df012f4e3e13
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.10.16
|
Security for leanquant 0.1.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.