{"library":"jmp","type":"library","category":null,"description":"JMP is a Python library from DeepMind that provides abstractions for mixed precision training in JAX. It enables the use of full and half-precision floating-point numbers during model training to reduce memory bandwidth and improve computational efficiency. It is currently at version 0.0.4 and sees active development with new releases addressing JAX compatibility and feature enhancements.","language":"python","status":"active","version":"0.0.4","tags":["jax","mixed precision","deep learning","deepmind","machine learning"],"last_verified":"Sat May 23","install":[{"cmd":"pip install jmp","imports":["import jmp","from jmp import Policy","from jmp import get_policy","from jmp import DynamicLossScale","from jmp import all_finite"]},{"cmd":"# First, install JAX with accelerator support (e.g., CPU, CUDA, TPU) following JAX's official instructions.\n# Example for CPU:\npip install jax[cpu]\n\npip install jmp","imports":[]}],"homepage":null,"github":"https://github.com/deepmind/jmp","docs":null,"changelog":null,"pypi":"https://pypi.org/project/jmp/","npm":null,"openapi_spec":null,"status_page":null,"smithery":null,"compatibility":{"summary":{"python_range":"3.10–3.9","success_rate":100,"avg_install_s":3.6,"avg_import_s":null,"wheel_type":"wheel"},"url":"https://checklist.day/v1/registry/jmp/compatibility"}}