List your company
ResearchFormerly WizWand

Language Modeling on Instruction-tuned models aggregate

93.4Accuracy RetentionSUBFIT
63.65671.37879.186.822Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SUBFITSparsity=12.5%2026.0693.4
Streamline (Layer)Sparsity=12.5%2026.0693.2
ReplaceMe (Cosine)Sparsity=12.5%2026.0693.2
ReplaceMe (LS)Sparsity=12.5%2026.0692.7
Streamline (FFN)Sparsity=12.5%2026.0691.7
SUBFITSparsity=20.0%2026.0688.1
ReplaceMe (Cosine)Sparsity=20.0%2026.0686.8
Streamline (FFN)Sparsity=20.0%2026.0686.2
ReplaceMe (LS)Sparsity=20.0%2026.0685.8
Streamline (Layer)Sparsity=20.0%2026.0685.4
SUBFITSparsity=25.0%2026.0684.6
Streamline (Layer)Sparsity=25.0%2026.0681.6
ReplaceMe (LS)Sparsity=25.0%2026.0681.5
ReplaceMe (Cosine)Sparsity=25.0%2026.0681.5
SUBFITSparsity=30.0%2026.0680.8
ReplaceMe (Cosine)Sparsity=30.0%2026.0679.1
ReplaceMe (LS)Sparsity=30.0%2026.0677.7
Streamline (FFN)Sparsity=25.0%2026.0677.2
Streamline (Layer)Sparsity=30.0%2026.0674.9
SUBFITSparsity=37.5%2026.0673.9
Streamline (FFN)Sparsity=30.0%2026.0672.3
ReplaceMe (Cosine)Sparsity=37.5%2026.0670.9
ReplaceMe (LS)Sparsity=37.5%2026.0670.8
Streamline (Layer)Sparsity=37.5%2026.0668.7
Streamline (FFN)Sparsity=37.5%2026.0664.8