Dear ICCAD 2026 FloorSet organizers,
We would like to clarify the allowed runtime environment for the final evaluation.
For reproducibility, we would like to ensure that our local validation environment can be matched as closely as possible during official evaluation. Our
validated environment uses:
Python 3.10 / 3.13
numba
scipy
shapely
numpy
We would like to ask:
- Are participants allowed to include a requirements.txt that pins the PyTorch CUDA wheel, for example:
--extra-index-url https://download.pytorch.org/whl/cu128
torch==2.10.0+cu128
numba>=0.58.0
scipy>=1.10.0
-
During official evaluation, will the grader install each submission’s requirements.txt, or will all submissions be evaluated only under the organizer-
provided default environment?
-
If external package indexes are not allowed, is there a recommended way to request a specific PyTorch/CUDA version?
-
Are Docker or container-based submissions allowed, so that participants can provide a fully reproducible runtime environment?
Our goal is to avoid dependency-version drift between local validation and official evaluation, and to make the submitted results as reproducible as possible.
Thank you for your clarification.
Dear ICCAD 2026 FloorSet organizers,
We would like to clarify the allowed runtime environment for the final evaluation.
For reproducibility, we would like to ensure that our local validation environment can be matched as closely as possible during official evaluation. Our
validated environment uses:
Python 3.10 / 3.13
numba
scipy
shapely
numpy
We would like to ask:
--extra-index-url https://download.pytorch.org/whl/cu128
torch==2.10.0+cu128
numba>=0.58.0
scipy>=1.10.0
During official evaluation, will the grader install each submission’s requirements.txt, or will all submissions be evaluated only under the organizer-
provided default environment?
If external package indexes are not allowed, is there a recommended way to request a specific PyTorch/CUDA version?
Are Docker or container-based submissions allowed, so that participants can provide a fully reproducible runtime environment?
Our goal is to avoid dependency-version drift between local validation and official evaluation, and to make the submitted results as reproducible as possible.
Thank you for your clarification.