\begin{table}[htbp]
\centering
\footnotesize
\caption{Hyperparameters of the proposed model.}
\label{tab:hyperparameters}
\begin{tabular}{lc}
\toprule
Parameter & Value \\
\midrule
data.cache\_frames & 24 \\
data.cache\_seconds & 6.0 \\
data.face\_crop.box\_scale & 1.8 \\
data.face\_crop.decode\_max\_side & 640 \\
data.face\_crop.detect\_frames & 4 \\
data.face\_crop.detector & yunet \\
data.face\_crop.enabled & True \\
data.frame\_size & 224 \\
data.n\_frames & 16 \\
data.primary & fakeavceleb \\
data.sample\_rate & 16000 \\
data.split\_fractions & [0.65, 0.15, 0.2] \\
data.split\_protocol & strict\_identity \\
data.split\_seed & 42 \\
data.subsample\_seed & 0 \\
data.window\_seconds & 4.0 \\
features.audio\_model & microsoft/wavlm-base-plus \\
features.audio\_tokens & 50 \\
features.dtype & float16 \\
features.eval\_view\_offset & 4 \\
features.extract\_batch\_size & auto \\
features.qacp\_variants\_per\_clip & 4 \\
features.shard\_clips & 512 \\
features.shard\_clips\_audio & 4096 \\
features.train\_view\_offsets & [0, 4, 8] \\
features.video\_model & MCG-NJU/videomae-base \\
features.video\_spatial\_grid & 4 \\
model.compose\_quadrant & False \\
model.d\_model & 768 \\
model.dropout & 0.1 \\
model.n\_fusion\_layers & 4 \\
model.n\_heads & 8 \\
model.use\_disentangle & True \\
model.use\_sync & True \\
train.phase\_b.betas & [0.9, 0.999] \\
train.phase\_b.early\_stopping\_patience & 3 \\
train.phase\_b.effective\_batch & 16 \\
train.phase\_b.epochs & 6 \\
train.phase\_b.gradient\_checkpointing & True \\
train.phase\_b.loss\_weights.a & 1.0 \\
train.phase\_b.loss\_weights.disentangle & 0.1 \\
train.phase\_b.loss\_weights.loc & 0.5 \\
train.phase\_b.loss\_weights.quad & 0.5 \\
train.phase\_b.loss\_weights.sync & 0.1 \\
train.phase\_b.loss\_weights.v & 1.0 \\
train.phase\_b.lr & 3e-05 \\
train.phase\_b.lr\_encoder & 1e-05 \\
train.phase\_b.max\_grad\_norm & 1.0 \\
train.phase\_b.micro\_batch & auto \\
train.phase\_b.min\_lr\_ratio & 0.01 \\
train.phase\_b.modality\_dropout & 0.15 \\
train.phase\_b.nonfinite\_patience & 20 \\
train.phase\_b.sampler & sqrt\_balanced \\
train.phase\_b.sampler\_key & quadrant \\
train.phase\_b.scheduler & warmup\_cosine \\
train.phase\_b.selection\_metric & val/mean\_auc \\
train.phase\_b.temporal\_augment & True \\
train.phase\_b.unfreeze\_top\_blocks\_audio & 2 \\
train.phase\_b.unfreeze\_top\_blocks\_video & 2 \\
train.phase\_b.warmup\_ratio & 0.1 \\
train.phase\_b.weight\_decay & 0.05 \\
train.qacp.betas & [0.9, 0.999] \\
train.qacp.early\_stopping\_patience & 8 \\
train.qacp.effective\_batch & 256 \\
train.qacp.epochs & 40 \\
train.qacp.items\_per\_epoch & 4096 \\
train.qacp.lr & 0.0003 \\
train.qacp.max\_grad\_norm & 1.0 \\
train.qacp.micro\_batch & auto \\
train.qacp.min\_lr\_ratio & 0.01 \\
train.qacp.nonfinite\_patience & 20 \\
train.qacp.pseudo\_classes & ["RVRA", "RVFA", "FVRA", "FVFA", "MISMATCH"] \\
train.qacp.scheduler & warmup\_cosine \\
train.qacp.selection\_metric & val/qacp\_loss \\
train.qacp.temperature & 0.1 \\
train.qacp.warmup\_ratio & 0.05 \\
train.qacp.weight\_decay & 0.05 \\
train.stage1.betas & [0.9, 0.999] \\
train.stage1.early\_stopping\_patience & 6 \\
train.stage1.effective\_batch & 64 \\
train.stage1.epochs & 30 \\
train.stage1.init\_from\_qacp & True \\
train.stage1.loss\_weights.a & 1.0 \\
train.stage1.loss\_weights.disentangle & 0.1 \\
train.stage1.loss\_weights.loc & 0.5 \\
train.stage1.loss\_weights.quad & 0.5 \\
train.stage1.loss\_weights.sync & 0.1 \\
train.stage1.loss\_weights.v & 1.0 \\
train.stage1.lr & 0.0001 \\
train.stage1.max\_grad\_norm & 1.0 \\
train.stage1.micro\_batch & auto \\
train.stage1.min\_lr\_ratio & 0.01 \\
train.stage1.modality\_dropout & 0.15 \\
train.stage1.nonfinite\_patience & 20 \\
train.stage1.optimizer & adamw \\
train.stage1.sampler & sqrt\_balanced \\
train.stage1.sampler\_key & quadrant \\
train.stage1.scheduler & warmup\_cosine \\
train.stage1.selection\_metric & val/mean\_auc \\
train.stage1.warmup\_ratio & 0.05 \\
train.stage1.weight\_decay & 0.05 \\
\bottomrule
\end{tabular}
\end{table}