Model profile
InternLM2 Chat 20B
Evidence summary
Safety. InternLM2 Chat 20B has an estimated Safety rank of #123; its 90% source-sensitivity interval is #36–#266. Its behavior-only rank is #121; company governance moves the combined estimate to #123. Published Safety evidence spans 5 eval lineages and 5 of 7 components. Its strongest relative result is SALAD-Bench (base_representation_toxicity, #2 of 33); its weakest is ChineseSafe (score, #15 of 22).
Freedom. InternLM2 Chat 20B has an estimated Freedom rank of #459; its 90% source-sensitivity interval is #246–#545. Published Freedom evidence spans 2 eval lineages and 1 of 1 components. Its strongest relative result is SALAD-Bench (attack_enhanced_information_safety_harms, #12 of 33); its weakest is SALAD-Bench (base_representation_toxicity, #32 of 33).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗internlm/internlm2-chat-20b
- Official model page ↗Exact model document · Reviewed official Hugging Face owner · official repository
- Release source ↗direct research preview identity date
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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| ChineseSafescore | #15 / 22 | ↑53.67 | Source ↗official | |
| CMoralEvalfamilial_morality | #3 / 26 | ↑0.56 | Source ↗official | |
| CMoralEvalinternet_ethics | #3 / 26 | ↑0.54 | Source ↗official | |
| CMoralEvalpersonal_morality | #3 / 26 | ↑0.52 | Source ↗official | |
| CMoralEvalprofessional_ethics | #3 / 26 | ↑0.54 | Source ↗official | |
| CMoralEvalsocial_morality | #3 / 26 | ↑0.54 | Source ↗official | |
| Enkrypt AI Safety Leaderboardbias_attack_non_success_rate | #86 / 248 | ↑20.41 | Source ↗official | |
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #35 / 248 | ↑93.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #123 / 248 | ↑71.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #121 / 246 | ↑96.41 | Source ↗official | |
| FinEval Financial Security Knowledgefinancial_security_accuracy_pct | #10 / 19 | ↑73.1 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #15 / 33 | ↑15.52 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #21 / 33 | ↑7.17 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #15 / 33 | ↑12.56 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #19 / 33 | ↑9.38 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #20 / 33 | ↑10.12 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #17 / 33 | ↑12.55 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #3 / 33 | ↑98.54 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #11 / 33 | ↑96.68 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #2 / 33 | ↑98.88 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #2 / 33 | ↑98.67 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #2 / 33 | ↑97.53 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #3 / 33 | ↑95.77 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #8 / 33 | ↑61.11 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #7 / 33 | ↑51.11 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #6 / 33 | ↑57.24 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #6 / 33 | ↑59.05 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #6 / 33 | ↑57.92 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #10 / 33 | ↑46.11 | Source ↗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-eval | Rank | Value | Distribution | Source |
|---|---|---|---|---|
| Enkrypt AI Safety Leaderboardcbrn_attack_non_success_rate | #212 / 248 | ↓93.33 | Source ↗official | |
| Enkrypt AI Safety Leaderboardharmful_attack_non_success_rate | #124 / 248 | ↓71.67 | Source ↗official | |
| Enkrypt AI Safety Leaderboardtoxicity_attack_non_success_rate | #126 / 246 | ↓96.41 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #19 / 33 | ↓15.52 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #12 / 33 | ↓7.17 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #19 / 33 | ↓12.56 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #14 / 33 | ↓9.38 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #14 / 33 | ↓10.12 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #17 / 33 | ↓12.55 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #31 / 33 | ↓98.54 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #23 / 33 | ↓96.68 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #32 / 33 | ↓98.88 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #31 / 33 | ↓98.67 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #32 / 33 | ↓97.53 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #31 / 33 | ↓95.77 | Source ↗official |