CalvinBackup/OBLITERATUS
1
0
Fork 0
mirror of https://github.com/elder-plinius/OBLITERATUS.git synced 2026-07-14 07:57:21 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add files via upload

Browse Source
This commit is contained in:
pliny
2026-03-05 00:52:16 -08:00
committed by GitHub
parent 66ea4a6f86
commit 6120061553
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
paper/appendix.tex
+1 -1
View File
@@ -511,7 +511,7 @@ All three confirm that sparse surgery is strictly more efficient than random row
Following the NeurIPS/ICML reproducibility guidelines:
\begin{enumerate}[leftmargin=*]
\item \textbf{Code availability}: Full source code released under AGPL-3.0 at \url{https://github.com/obliteratus-project/OBLITERATUS}. Version 0.1.0 archived on Zenodo (DOI pending).
\item \textbf{Code availability}: Full source code released under AGPL-3.0 at \url{https://github.com/elder-plinius/OBLITERATUS}. Version 0.1.0 archived on Zenodo (DOI pending).
\item \textbf{Dependencies}: All dependencies pinned in \texttt{pyproject.toml}; Docker image available for exact environment reproduction.
\item \textbf{Random seeds}: The platform defaults to seed 42 and supports multi-seed sweeps ($s \in \{42, 137, 2024\}$) with bootstrap CIs. All tables in this paper report single-run results with seed 42. See Section~\ref{para:stat_limitations} for a discussion of statistical limitations and confidence intervals.
\item \textbf{Compute}: All pipeline stages are designed to run on a single GPU. Full evaluation (7 models $\times$ 3 methods) requires ${\sim}$12 GPU-hours on an NVIDIA A100 (80\,GB). Reproducible on consumer hardware (RTX 3090/4090) with quantization.