AIMET Files

AIMET is a library that provides advanced quantization and compression

This is an exact mirror of the AIMET project, hosted at https://github.com/quic/aimet. SourceForge is not affiliated with AIMET.

The interactive file manager requires Javascript. Please enable it or use sftp or scp.
You may still browse the files here.

Name	Modified	Size	InfoDownloads / Week
Parent folder
onnx-cpu-win-arm64-wheel.zip	2026-08-26	673.7 kB	0
onnx-cpu-win-x64-wheel.zip	2026-08-26	704.7 kB	0
aimet_onnx-2.38.0+cpu-cp310-abi3-manylinux_2_34_x86_64.whl	2026-08-24	837.9 kB	0
aimet_onnx-2.38.0-cp310-abi3-manylinux_2_34_x86_64.whl	2026-08-24	84.3 MB	0
aimet_torch-2.38.0+cpu-py310-none-any.whl	2026-08-24	809.3 kB	0
aimet_torch-2.38.0-py310-none-any.whl	2026-08-24	809.3 kB	0
README.md	2026-08-24	2.9 kB	0
Version 2.38.0 source code.tar.gz	2026-08-24	9.7 MB	0
Version 2.38.0 source code.zip	2026-08-24	10.5 MB	0
Totals: 9 Items		108.2 MB	0

2.38.0 ======

New Features
- ONNX
  - Add QSpec and QcQuantizeOp.set_qspec to configure a quantizer's precision, granularity, and symmetry in a single call (495702e)
  - Accept QSpec in set_param_type to configure blockwise/LPBQ param quantization (e29e8d9)
  - Add experimental create_truncation_aware_session to simulate int32 accumulator truncation for MatMul/Conv (cc65cdc)
  - Support SpinQuant R1 with online embedding rotation, including models with no LM head (8960f1b, 9a7726e)
- Torch
  - Preserve the graph signature of user inputs throughout aimet_torch.export (443f638)
- Common
  - Propagate encodings of grid-equivariant ops at export time (89bf3b7)
Bug fixes and Improvements
- ONNX
  - Skip overwriting encodings of frozen quantizers when loading encodings (1feeb61)
  - Consolidate LPBQ encoding loading directly onto QcQuantizeOp in support of the new QSpec API (354a33e)
- Torch
  - Permanently remove the deprecated enable_recompute API; use torch.utils.checkpoint instead (006db48)
  - Fix sequential MSE failure with deepspeed 0.19.4 (cee185a)
  - Export transposed linear weight as a param encoding in v1.0.0 format (2411e7f)
  - Recognize quantize_per_tensor/dequantize_per_tensor as quantization encodings during ONNX QDQ export (d76bb1e)
  - Suppress encoding propagation across RotaryEmbedding boundary during export (0d653dc)
- Common
  - Fix sin/cos output range to -1..1 in default quantsim configs (2c95329)

Source: README.md, updated 2026-08-24

Other Useful Business Software

MongoDB Atlas runs apps anywhere Icon

MongoDB Atlas runs apps anywhere

Deploy in 115+ regions with the modern database for every enterprise.

MongoDB Atlas gives you the freedom to build and run modern applications anywhere—across AWS, Azure, and Google Cloud. With global availability in over 115 regions, Atlas lets you deploy close to your users, meet compliance needs, and scale with confidence across any geography.

Start Free

Train ML Models With SQL You Already Know Icon

Train ML Models With SQL You Already Know

BigQuery automates data prep, analysis, and predictions with built-in AI assistance.

Build and deploy ML models using familiar SQL. Automate data prep with built-in Gemini. Query 1 TB and store 10 GB free monthly.

Start Free

$300 Free Credits to Build on Google Cloud

New customers can spin up VMs, build with AI, and query data at no cost.

Put your $300 in credit toward real workloads, then keep building with free monthly usage for 20+ products. No commitment and no charge until you upgrade.

Start Free

Recommended Projects

AirLLM
AirLLM 70B inference with single 4GB GPU
MAD (MPEG Audio Decoder)
MAD is a high-quality fixed-point MPEG audio decoder with 24-bit output. The implementation is entirely new, based on the ISO/IEC standards, and performs especially well on systems without native floating-point support.
AutoGPTQ
An easy-to-use LLMs quantization package with user-friendly apis
NNCF
Neural Network Compression Framework for enhanced OpenVINO
Model Explorer
A modern model graph visualizer and debugger