aqtp
Accurate Quantized Training library.
Decision gist · record as of 2026-08-14
Yes, if you are training neural networks in JAX and need quantization. The library is actively maintained, has no known vulnerabilities, carries a permissive license, and is backed by Google's production use at scale. Install friction is low. The main constraint is a hard dependency on jax and jaxlib (which require compilation for your hardware), and the library is still in alpha (Development Status :: 3), so expect API changes. Best suited for researchers and practitioners already committed to the JAX ecosystem.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires Python 3.10 or later; jaxlib has platform-specific compiled components (CPU, GPU, or TPU support depends on your JAX installation).
- Low install friction with a pure-Python wheel.
- Depends on four substantial packages (absl-py, jax, jaxlib, flax), all of which are actively maintained.
License · maintenance · safety
permissive license (permissive) — Licensed under Apache Software License (permissive), imposing no restrictions on commercial or private use; attribution required but no copyleft obligations.
last release 2025-08-01 (378 days) · last repo commit 2026-08-06 · 360 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 457,805 downloads/mo, #6,547 on PyPI
Alternatives
Verify before relying
pip install aqtp
import aqt.jax.v2.config as aqt_config
int8_config = aqt_config.fully_quantized(fwd_bits=8, bwd_bits=8)- Whether the package works with Python 3.12+ (classifiers list only 3.10 and 3.11)
- Performance gains on non-TPU hardware (description emphasizes TPU and contemporary ML accelerators)
- Compatibility with recent versions of Flax and Pax beyond what the description exemplifies
What it is and what it does
AQT is a quantization library for JAX that replaces standard tensor operations (dot_general, einsum, convolution) with quantized variants, enabling training of neural networks at reduced precision (int8, int4, or lower) without hand-tuning. It works by injecting quantized operations into JAX-based frameworks like Flax, Pax, and MaxText, and is designed to maintain bit-exact consistency between training and serving—avoiding the training-serving bias common in post-training quantization.
The library provides flexible configuration of forward and backward pass quantization separately, with support for various numerics (int8, int4, bfloat16, float8), calibration algorithms, and stochastic rounding. It is built on top of jax, jaxlib, flax, and absl-py, and is actively maintained by Google with demonstrated production use at scale and research validation across multiple papers and frameworks.
Use it for
- Train large language models or vision models with int8 quantization on TPU or GPU clusters to reduce memory and accelerate computation.
- Conduct quantization research by configuring forward and backward bit widths independently and testing different calibration strategies.
- Deploy models trained with AQT without retraining or post-training quantization, since quantized weights are identical during training and serving.
- Integrate quantization into existing Flax models by replacing the dot_general operation without restructuring the model.
- Benchmark lower-bit training on production workloads to evaluate accuracy-efficiency tradeoffs.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, if you are training neural networks in JAX and need quantization.
The library is actively maintained, has no known vulnerabilities, carries a permissive license, and is backed by Google's production use at scale. Install friction is low. The main constraint is a hard dependency on jax and jaxlib (which require compilation for your hardware), and the library is still in alpha (Development Status :: 3), so expect API changes. Best suited for researchers and practitioners already committed to the JAX ecosystem.
Install
aqtp on PyPI
Before you install
Low install friction with a pure-Python wheel. Depends on four substantial packages (absl-py, jax, jaxlib, flax), all of which are actively maintained. The library itself shows active maintenance with a recent commit on 2026-08-06 and steady releases since 2022-01-13.
Requires Python 3.10 or later; jaxlib has platform-specific compiled components (CPU, GPU, or TPU support depends on your JAX installation).
License in practice
Licensed under Apache Software License (permissive), imposing no restrictions on commercial or private use; attribution required but no copyleft obligations.
Quickstart
pip install aqtp
import aqt.jax.v2.config as aqt_config
int8_config = aqt_config.fully_quantized(fwd_bits=8, bwd_bits=8)Verify before relying
- Whether the package works with Python 3.12+ (classifiers list only 3.10 and 3.11)
- Performance gains on non-TPU hardware (description emphasizes TPU and contemporary ML accelerators)
- Compatibility with recent versions of Flax and Pax beyond what the description exemplifies
Package facts
| License | permissive license permissive |
| Python support | Supports the current Python release >=3.10 |
| Install friction | Low. Pure-Python wheel |
| Runtime dependencies | 4 packagesabsl-pyjaxjaxlibflax |
| Maintenance | Actively maintained 378 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 457,805 / month, #6,547 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
| Classifiers | Development Status :: 3 - AlphaIntended Audience :: DevelopersIntended Audience :: Science/ResearchLicense :: OSI Approved :: Apache Software LicenseProgramming Language :: Python :: 3.10Programming Language :: Python :: 3.11Topic :: Scientific/Engineering :: Artificial Intelligence |
Evidence: aqtp-0.9.0-py3-none-any.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “JAX tensor quantization”
- aqtpAQT provides quantization of tensor operations (matmul, einsum, conv)…
- qwixQwix is a JAX quantization library that applies Quantization-Aware…
- comfy-kitchenComfy Kitchen provides optimized GPU kernels for quantized tensor…
Give your agent the search over MCP, or paste the wish link into any chat.
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to call 100+ LLM providers (OpenAI, Anthropic, Gemini, Bedrock, Azure, and others) using OpenAI-compatible API format, available as both a Python SDK and a self-hosted AI Gateway proxy server.
Install it if you need to work with multiple LLM providers or want to centralize LLM routing in your organization.
Client library and CLI tool for downloading, uploading, and managing models, datasets, and repositories on the Hugging Face Hub platform.
Install it if you work with Hugging Face Hub models or datasets.
LangChain provides a framework for building agents and LLM-powered applications by composing language models, tools, and memory through a unified API that abstracts over multiple model providers.
hf-xet provides chunk-based deduplication and efficient file transfer for the Hugging Face Hub, enabling faster uploads and downloads of large files with local disk caching.
Tokenizers converts raw text into token sequences for NLP models, with support for training custom vocabularies and using pre-built tokenizers (BPE, WordPiece) optimized for speed via Rust.
Transformers provides a unified framework for loading, fine-tuning, and running state-of-the-art pretrained models across text, vision, audio, video, and multimodal tasks using PyTorch, JAX, or TensorFlow.
Install it if you need to run or train any transformer-based model for NLP, vision, audio, or multimodal tasks.
See also qwix · optimum-quanto · ai-edge-quantizer · compressed-tensors · diffq · aqtinstall · vector-quantize-pytorch · auto-gptq · qonnx · google-tunix