← Models

Model profile

Llama 3.1 Tulu 3 8B Sft

2024-11-18release date
2eval lineages

Evidence summary

Published evidence spans 2 evals and 5 of 7 behavior components. Its strongest relative result is Enkrypt AI Safety Leaderboard (harmful_attack_non_success_rate, #72 of 260); its weakest is AbstentionBench (underspecified_context_f1, #20 of 20).

Compare this model

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

Official and reference links

Published eval results

Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.

Eval / sub-evalRankValueDistributionBetterSource
AbstentionBenchanswer_unknown_f1#14 / 200.8605↑ higherSource ↗official
AbstentionBenchfalse_premise_f1#18 / 200.6112↑ higherSource ↗official
AbstentionBenchstale_f1#11 / 200.632↑ higherSource ↗official
AbstentionBenchsubjective_f1#11 / 200.7244↑ higherSource ↗official
AbstentionBenchunderspecified_context_f1#20 / 200.3572↑ higherSource ↗official
AbstentionBenchunderspecified_intent_f1#11 / 200.7311↑ higherSource ↗official
Enkrypt AI Safety Leaderboardbias_attack_non_success_rate#227 / 2609.82↑ higherSource ↗official
Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate#125 / 26088.5↑ higherSource ↗official
Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate#72 / 26086.11↑ higherSource ↗official
Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate#99 / 25897.45↑ higherSource ↗official