← Models

Model profile

Tulu 2 DPO 13B

Ai2developer
2023-11-13release date
Not rankedSafety rank
#330 / 662Freedom rank

Evidence summary

Safety. Tulu 2 DPO 13B 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, #12 of 33); its weakest is SALAD-Bench (attack_enhanced_human_autonomy_integrity, #31 of 33).

Freedom. Tulu 2 DPO 13B has an estimated Freedom rank of #330; its 90% source-sensitivity interval is #77–#638. Published Freedom evidence spans 1 eval lineages and 1 of 1 components. Its strongest relative result is SALAD-Bench (attack_enhanced_human_autonomy_integrity, #2 of 33); its weakest is SALAD-Bench (base_information_safety_harms, #12 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#31 / 33↑4.53Source ↗official
SALAD-Benchattack_enhanced_information_safety_harms#28 / 33↑4.56Source ↗official
SALAD-Benchattack_enhanced_malicious_use#27 / 33↑6.12Source ↗official
SALAD-Benchattack_enhanced_misinformation_harms#28 / 33↑5.76Source ↗official
SALAD-Benchattack_enhanced_representation_toxicity#28 / 33↑5.59Source ↗official
SALAD-Benchattack_enhanced_socioeconomic_harms#29 / 33↑6.93Source ↗official
SALAD-Benchbase_human_autonomy_integrity#23 / 33↑89.52Source ↗official
SALAD-Benchbase_information_safety_harms#22 / 33↑92.82Source ↗official
SALAD-Benchbase_malicious_use#24 / 33↑85.52Source ↗official
SALAD-Benchbase_misinformation_harms#24 / 33↑88.48Source ↗official
SALAD-Benchbase_representation_toxicity#23 / 33↑85.28Source ↗official
SALAD-Benchbase_socioeconomic_harms#23 / 33↑84.49Source ↗official
SALAD-Benchmcq_human_autonomy_integrity#14 / 33↑41.67Source ↗official
SALAD-Benchmcq_information_safety_harms#16 / 33↑34.44Source ↗official
SALAD-Benchmcq_malicious_use#15 / 33↑40.13Source ↗official
SALAD-Benchmcq_misinformation_harms#12 / 33↑40.71Source ↗official
SALAD-Benchmcq_representation_toxicity#12 / 33↑40.31Source ↗official
SALAD-Benchmcq_socioeconomic_harms#14 / 33↑38.89Source ↗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#2 / 33↓4.53Source ↗official
SALAD-Benchattack_enhanced_information_safety_harms#6 / 33↓4.56Source ↗official
SALAD-Benchattack_enhanced_malicious_use#7 / 33↓6.12Source ↗official
SALAD-Benchattack_enhanced_misinformation_harms#6 / 33↓5.76Source ↗official
SALAD-Benchattack_enhanced_representation_toxicity#6 / 33↓5.59Source ↗official
SALAD-Benchattack_enhanced_socioeconomic_harms#4 / 33↓6.93Source ↗official
SALAD-Benchbase_human_autonomy_integrity#11 / 33↓89.52Source ↗official
SALAD-Benchbase_information_safety_harms#12 / 33↓92.82Source ↗official
SALAD-Benchbase_malicious_use#10 / 33↓85.52Source ↗official
SALAD-Benchbase_misinformation_harms#10 / 33↓88.48Source ↗official
SALAD-Benchbase_representation_toxicity#11 / 33↓85.28Source ↗official
SALAD-Benchbase_socioeconomic_harms#11 / 33↓84.49Source ↗official