List your company

General Language Capability on MMLU, GSM8K, HEval, IFEval, MTBench

76.4MMLU AccuracyBaseline
68.18470.31772.4574.583Jul 1, 2026

Evaluation results

MethodLinks
BaselineBackbone=QWEN-2.5-7B2026.0776.480.86453.682.471.4
DPOBackbone=QWEN-2.5-7B2026.0776.482.46449.879.170.3
HARC + DPOBackbone=QWEN-2.5-7B2026.0776.483.26451.880.271.1
RepBendBackbone=QWEN-2.5-7B2026.0776.379.664.60.41046.2
HARCBackbone=QWEN-2.5-7B2026.0776.380.468.35182.671.7
Circuit BreakerBackbone=QWEN-2.5-7B2026.0776.280.463.451.279.170.1
SFTBackbone=QWEN-2.5-7B2026.07768059.133.166.663
CASTBackbone=QWEN-2.5-7B2026.0774.576.46451.882.569.8
DPOBackbone=LLAMA-3.1-8B2026.077075.659.146.870.964.5
BaselineBackbone=LLAMA-3.1-8B2026.0769.974.462.252.175.566.8
SFTBackbone=LLAMA-3.1-8B2026.0769.976.860.445.770.664.7
HARC + DPOBackbone=LLAMA-3.1-8B2026.0769.973.659.85076.666
Circuit BreakerBackbone=LLAMA-3.1-8B2026.0769.874.861.649.277.466.6
HARCBackbone=LLAMA-3.1-8B2026.0769.875.66151.277.267
RepBendBackbone=LLAMA-3.1-8B2026.0769.76661.626.851.655.1
CASTBackbone=LLAMA-3.1-8B2026.0768.5585041.96155.9