Model profile
Llama 2 7B Chat
Metadeveloper
2023-07-18release date
#140 / 267overall rank
12eval lineages
Evidence summary
Llama 2 7B Chat has an estimated overall rank of #140; its 90% source-sensitivity interval is #64–#201. Its behavior-only rank is #128; company governance moves the combined estimate to #140. Published evidence spans 12 evals and 7 of 7 behavior components. Its strongest relative result is HarmBench (dr, #1 of 28); its weakest is Contextual MoralChoice (human_agreement, #21 of 22).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗meta-llama/Llama-2-7b-chat-hf
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
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-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| Contextual MoralChoicehuman_agreement | #21 / 22 | 0.3 | ↑ higher | Source ↗official | |
| DecodingTrustmachine_ethics | #7 / 8 | 40.58 | ↑ higher | Source ↗official | |
| DecodingTruststereotype_bias | #3 / 8 | 97.6 | ↑ higher | Source ↗official | |
| DecodingTrusttoxicity | #2 / 8 | 80 | ↑ higher | Source ↗official | |
| Do-Not-Answerhuman_harmlessness_rate | #1 / 6 | 99.68 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #102 / 260 | 18.6 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #100 / 260 | 89.83 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #74 / 260 | 85.56 | ↑ higher | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #29 / 258 | 99.5 | ↑ higher | Source ↗official | |
| HarmBenchdr | #1 / 28 | 0.8 | ↓ lower | Source ↗official | |
| JailBenchjailbreak_success_rate | #3 / 14 | 48.68 | ↓ lower | Source ↗official | |
| Large-scale Moral Machine experiment on LLMshuman_choice_distance | #20 / 39 | 1.077 | ↓ lower | Source ↗official | |
| OR-Benchover_refusal_rate | #18 / 25 | 87.5 | ↓ lower | Source ↗official | |
| OR-Benchtoxic_acceptance_rate | #6 / 25 | 0.4 | ↓ lower | Source ↗official | |
| S-Evalbase_en_overall | #3 / 22 | 78.2 | ↑ higher | Source ↗official | |
| SafetyBenchEM | #20 / 21 | 49.8 | ↑ higher | Source ↗official | |
| SafetyBenchIA | #19 / 21 | 62.4 | ↑ higher | Source ↗official | |
| SafetyBenchMH | #18 / 21 | 70.2 | ↑ higher | Source ↗official | |
| SafetyBenchOFF | #18 / 21 | 48.9 | ↑ higher | Source ↗official | |
| SafetyBenchPH | #18 / 21 | 54.5 | ↑ higher | Source ↗official | |
| SafetyBenchPP | #18 / 21 | 65 | ↑ higher | Source ↗official | |
| SafetyBenchUB | #9 / 21 | 63.2 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #13 / 33 | 20.47 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #14 / 33 | 17.59 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #14 / 33 | 16.31 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #13 / 33 | 18.75 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #14 / 33 | 18.21 | ↑ higher | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #8 / 33 | 23.81 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #5 / 33 | 98.37 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #3 / 33 | 99.26 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_malicious_use | #4 / 33 | 98.54 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #9 / 33 | 96.26 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #9 / 33 | 93.06 | ↑ higher | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #6 / 33 | 94.12 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #27 / 33 | 6.389 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #29 / 33 | 3.333 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #28 / 33 | 4.551 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #27 / 33 | 6.19 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #28 / 33 | 4.271 | ↑ higher | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #29 / 33 | 1.667 | ↑ higher | Source ↗official | |
| SORRY-Benchavg | #12 / 51 | 0.14 | ↓ lower | Source ↗official |