Instructions to use Efficient-Large-Model/Sol-Attn-Kernel-Source with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Kernels
How to use Efficient-Large-Model/Sol-Attn-Kernel-Source with Kernels:
# !pip install kernels from kernels import get_kernel kernel = get_kernel("Efficient-Large-Model/Sol-Attn-Kernel-Source") - Notebooks
- Google Colab
- Kaggle
Source provenance
- Upstream repository: https://github.com/NVlabs/Sana
- Upstream branch:
sol-engine - Upstream commit:
8a26fb0ec9e353125ead798cb2e312d5ce48cded - Upstream path:
techniques/sparse_backends/sol_attn - Kernel Hub target:
Efficient-Large-Model/Sol-Attn
torch-ext/sol_attn contains the complete upstream package at the pinned
commit. Its only packaging-level changes are semantically equivalent
absolute-to-relative internal imports. Kernel Hub loads each kernel version
under an isolated module name, so hard-coded sol_attn.* package imports
would otherwise escape that namespace.
Given a checkout of the pinned Sana commit, verify the file inventory, all non-import source text, and import semantics with:
python tools/verify_upstream.py /path/to/Sana
The publishing process does not add files, workflows, or commits to
NVlabs/Sana.