← Models

Model profile

Tulu 2 DPO 7B

Ai2developer
2023-11-13release date
Not rankedSafety rank
#265 / 662Freedom rank

Evidence summary

Safety. Tulu 2 DPO 7B does not meet the evidence gate for a Safety rank. Published Safety evidence spans 1 eval lineages and 4 of 7 components. Its strongest relative result is SALAD-Bench (mcq_representation_toxicity, #18 of 33); its weakest is SALAD-Bench (attack_enhanced_representation_toxicity, #30 of 33).

Freedom. Tulu 2 DPO 7B has an estimated Freedom rank of #265; its 90% source-sensitivity interval is #68–#612. Published Freedom evidence spans 1 eval lineages and 1 of 1 components. Its strongest relative result is SALAD-Bench (attack_enhanced_representation_toxicity, #4 of 33); its weakest is SALAD-Bench (base_misinformation_harms, #11 of 33).

Compare this model

Only models sharing at least one published sub-eval are listed.

Official and reference links

Safety evals

Rank and direction are specific to the Safety portfolio. Black marks the observed result; the grey dot marks the value implied by the Safety rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
SALAD-Benchattack_enhanced_human_autonomy_integrity#27 / 33↑6.47Source ↗official
SALAD-Benchattack_enhanced_information_safety_harms#25 / 33↑5.21Source ↗official
SALAD-Benchattack_enhanced_malicious_use#30 / 33↑4.57Source ↗official
SALAD-Benchattack_enhanced_misinformation_harms#29 / 33↑5.59Source ↗official
SALAD-Benchattack_enhanced_representation_toxicity#30 / 33↑3.56Source ↗official
SALAD-Benchattack_enhanced_socioeconomic_harms#24 / 33↑7.79Source ↗official
SALAD-Benchbase_human_autonomy_integrity#26 / 33↑85.73Source ↗official
SALAD-Benchbase_information_safety_harms#26 / 33↑89.3Source ↗official
SALAD-Benchbase_malicious_use#25 / 33↑83.26Source ↗official
SALAD-Benchbase_misinformation_harms#23 / 33↑89.81Source ↗official
SALAD-Benchbase_representation_toxicity#24 / 33↑84.49Source ↗official
SALAD-Benchbase_socioeconomic_harms#26 / 33↑81.08Source ↗official
SALAD-Benchmcq_human_autonomy_integrity#19 / 33↑27.78Source ↗official
SALAD-Benchmcq_information_safety_harms#22 / 33↑19.72Source ↗official
SALAD-Benchmcq_malicious_use#20 / 33↑26.86Source ↗official
SALAD-Benchmcq_misinformation_harms#19 / 33↑28.33Source ↗official
SALAD-Benchmcq_representation_toxicity#18 / 33↑28.65Source ↗official
SALAD-Benchmcq_socioeconomic_harms#21 / 33↑25.56Source ↗official

Freedom evals

Rank and direction are specific to the Freedom portfolio. Black marks the observed result; the grey dot marks the value implied by the Freedom rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionSource
SALAD-Benchattack_enhanced_human_autonomy_integrity#7 / 33↓6.47Source ↗official
SALAD-Benchattack_enhanced_information_safety_harms#7 / 33↓5.21Source ↗official
SALAD-Benchattack_enhanced_malicious_use#4 / 33↓4.57Source ↗official
SALAD-Benchattack_enhanced_misinformation_harms#5 / 33↓5.59Source ↗official
SALAD-Benchattack_enhanced_representation_toxicity#4 / 33↓3.56Source ↗official
SALAD-Benchattack_enhanced_socioeconomic_harms#8 / 33↓7.79Source ↗official
SALAD-Benchbase_human_autonomy_integrity#8 / 33↓85.73Source ↗official
SALAD-Benchbase_information_safety_harms#8 / 33↓89.3Source ↗official
SALAD-Benchbase_malicious_use#9 / 33↓83.26Source ↗official
SALAD-Benchbase_misinformation_harms#11 / 33↓89.81Source ↗official
SALAD-Benchbase_representation_toxicity#10 / 33↓84.49Source ↗official
SALAD-Benchbase_socioeconomic_harms#8 / 33↓81.08Source ↗official