List your company

Causal Reasoning on CRASS OOD (held-out)

95Performance P0Qwen2.5-32B
70.0476.528389.48Jun 14, 2026

Evaluation results

MethodLinks
Qwen2.5-32BAdapter type=Vernier, QLoRA=true2026.069572.522.5-18
VernierBackbone=Qwen2.5-32B, Quantization=QLoRA2026.069572.522.5-18
Qwen2.5-14BAdapter type=Vernier, QLoRA=true2026.0691.872.219.7-27
VernierBackbone=Qwen2.5-14B2026.0691.872.219.7-27
Qwen2.5-32BAdapter type=Base, QLoRA=true2026.069163.527.5—
Base (CLadder adapter)Backbone=Qwen2.5-32B, Quantization=QLoRA2026.069163.527.5—
Qwen2.5-14BAdapter type=Base, QLoRA=true2026.0689.562.527—
Base (CLadder adapter)Backbone=Qwen2.5-14B2026.0689.562.527—
Llama-3.1-8BAdapter type=Vernier2026.0687.255.831.3-23
VernierBackbone=Llama-3.1-8B2026.0687.255.831.3-23
Qwen2.5-7BAdapter type=Vernier2026.068762.224.8-27
VernierBackbone=Qwen2.5-7B2026.068762.224.8-27
Llama-3.1-8BAdapter type=Base2026.068544.540.5—
Base (CLadder adapter)Backbone=Llama-3.1-8B2026.068544.540.5—
Phi-3-mini (3.8B)Adapter type=Base2026.0684.554.530—
Qwen2.5-7BAdapter type=Base2026.0684.550.534—
Base (CLadder adapter)Backbone=Phi-3-mini (3.8B)2026.0684.554.530—
Base (CLadder adapter)Backbone=Qwen2.5-7B2026.0684.550.534—
Phi-3-mini (3.8B)Adapter type=Vernier2026.06845529-3
VernierBackbone=Phi-3-mini (3.8B)2026.06845529-3
Mistral-7BAdapter type=Base2026.0679.541.538—
Base (CLadder adapter)Backbone=Mistral-7B, Recipe=Default, lr=1e-42026.0679.541.538—
Base (CLadder adapter)Backbone=Mistral-7B, Recipe=Rescued, lr=5e-6, step=1502026.0679.541.538—
Mistral-7BAdapter type=Vernier, recipe=rescued, lr=5e-6, early stopping step=1502026.0673.536.537-3
VernierBackbone=Mistral-7B, Recipe=Rescued, lr=5e-6, step=1502026.0673.536.537-3
Mistral-7BAdapter type=Vernier, recipe=default, lr=1e-42026.06715021-45
VernierBackbone=Mistral-7B, Recipe=Default, lr=1e-42026.06715021-45