List your company

Reward Model Transfer on Anthropic Helpful (AH)

9.95AOGSEAM
-3.11240.27883.677.0612May 25, 2026

Evaluation results

MethodLinks
SEAMTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.059.95——-5.72
SEAMTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.058.2——-3.39
ANCHORTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.058.19——8.19
Conf.-basedTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.058.06——5.94
SEAMTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.055.88——3.5
ANCHORTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.054.41——4.41
Conf.-basedTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.053.26——3.08
ANCHORTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.053.13——3.13
NaiveTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.051.46——1.46
NaiveTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.051.37——0.96
NaiveTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.050.6——0.6
Conf.-basedTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.050.25——0.25
SEAMTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=UF, Evaluation Setting=Zero-shot transfer2026.050.17——-6.48
NaiveTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.05-0.04——-3.1
Conf.-basedTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.05-1.58——-3.88
ANCHORTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=H3, Evaluation Setting=Zero-shot transfer2026.05-2.61——-2.61
ANCHORTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=AH, Evaluation Setting=In-distribution2026.05—106.93.39—
ANCHORTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=AH, Evaluation Setting=In-distribution2026.05—12.75—
Conf.-basedTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=AH, Evaluation Setting=In-distribution2026.05—70.32.23—
Conf.-basedTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=AH, Evaluation Setting=In-distribution2026.05—661.51.72—
NaiveTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=AH, Evaluation Setting=In-distribution2026.05—98.73.13—
NaiveTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=AH, Evaluation Setting=In-distribution2026.05—173.10.45—
SEAMTransfer Pair=Llama-3.2-1B-Instruct -> Llama-3.1-8B-Instruct, Source Domain=AH, Evaluation Setting=In-distribution2026.05—-411.4-13.04—
SEAMTransfer Pair=Qwen3-1.7B -> Qwen3-8B, Source Domain=AH, Evaluation Setting=In-distribution2026.05—-1.3-3.42—