{"library":"aqtp","type":"library","category":null,"description":"AQT (Accurate Quantized Training) is a Python software library designed for easy tensor operation quantization in JAX, providing excellent quantized int8 model quality without extensive manual tuning. It enables significant training speedup on modern ML accelerators and offers simple, flexible APIs suitable for both production and research. AQT focuses on quantizing tensor operations like matmul, einsum, and conv, without making assumptions about their use in neural networks, making it injectable into any JAX computation. It has been extensively tested with frameworks such as Flax, Pax, and MaxText at Google. The current version is 0.9.0, with a rapid release cadence for minor versions (monthly/bi-monthly).","language":"python","status":"active","version":"0.9.0","tags":["quantization","jax","machine learning","deep learning","ai","neural networks","tensor operations","flax"],"last_verified":"Sat May 23","install":[{"cmd":"pip install aqtp","imports":["import aqt.jax.v2 as aqt"]}],"homepage":null,"github":"https://github.com/google/aqt","docs":null,"changelog":null,"pypi":"https://pypi.org/project/aqtp/","npm":null,"openapi_spec":null,"status_page":null,"smithery":null,"compatibility":{"summary":{"python_range":"3.10–3.9","success_rate":100,"avg_install_s":17.1,"avg_import_s":0,"wheel_type":"wheel"},"url":"https://checklist.day/v1/registry/aqtp/compatibility"}}