Overview
A Python Library provides APIs to move PyTorch Tensors between CPU and NVMe. It provides simple asynchronize I/O API, which helps you easily overlap computation and data moving. Instead of manually installing dependent C++ libraries, it can detect and install them automatically. It's user-friendly and efficient.
What's Changed
- polish code by @feifeibear in #1
- add pyi file by @ver217 in #3
- [hotfix] fix async read and add benchmark by @ver217 in #4
- add aio by @oahzxl in #2
- [feature] support sync IO by @ver217 in #5
- [feature] support register file by @ver217 in #6
- [feature] asyncio support readv/writev by @ver217 in #7
- add test files and fix some bugs by @oahzxl in #8
- [feature] Offloader support readv/writev by @ver217 in #9
- Update test section in readme by @oahzxl in #10
- update test_adam by @ver217 in #11
- test_adam use gpt2-xl by @ver217 in #12
- test_adam use tqdm and update default n_events by @ver217 in #13
- [test] Update catch to v2 and add install guidance by @ver217 in #14
- allow disabling backend when installing by @ver217 in #15
- Add CPUAdam and its test by @oahzxl in #16
- add 3rd party building by @feifeibear in #17
- Re-organize file with cmake by @oahzxl in #18
- Add offload test and backend probe by @oahzxl in #21
- [cli] add cli to check backends by @ver217 in #23
- auto install backends and allow selecting backends by @ver217 in #24
- update version and description by @ver217 in #25
- rename project and update readme by @ver217 in #26
- [hotfix] fix manifest and sdist version by @ver217 in #27
- fix version by @ver217 in #28
- [hotfix] fix libaio.so not found error by @ver217 in #29
- update install guide in readme by @ver217 in #30
- [docker] add dockerfile by @ver217 in #31
New Contributors
- @feifeibear made their first contribution in #1
- @ver217 made their first contribution in #3
- @oahzxl made their first contribution in #2
Full Changelog: https://github.com/hpcaitech/TensorNVMe/commits/v0.1.0
Contributors
feifeibear, ver217, and oahzxl