Model profile
InternLM Chat 20B
Evidence summary
Safety. InternLM Chat 20B has an estimated Safety rank of #112; its 90% source-sensitivity interval is #36–#235. Its behavior-only rank is #111; company governance moves the combined estimate to #112. Published Safety evidence spans 4 eval lineages and 4 of 7 components. Its strongest relative result is SALAD-Bench (base_information_safety_harms, #1 of 33); its weakest is SALAD-Bench (mcq_representation_toxicity, #29 of 33).
Freedom. InternLM Chat 20B has an estimated Freedom rank of #448; its 90% source-sensitivity interval is #176–#615. Published Freedom evidence spans 4 eval lineages and 1 of 1 components. Its strongest relative result is SuperCLUE Safety (traditional_safety, #5 of 31); its weakest is FLAMES (fairness, #13 of 13).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Hugging Face ↗internlm/internlm-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 |
|---|---|---|---|---|
| Fake Alignment (FINE)multiple_choice_safe_decision_rate | #3 / 14 | ↑69.33 | Source ↗official | |
| Fake Alignment (FINE)open_ended_safe_response_rate | #7 / 14 | ↑96 | Source ↗official | |
| FLAMESdata_protection | #1 / 13 | ↑63.16 | Source ↗official | |
| FLAMESfairness | #1 / 13 | ↑52.61 | Source ↗official | |
| FLAMESlegality | #2 / 13 | ↑71.74 | Source ↗official | |
| FLAMESmorality | #1 / 13 | ↑54.23 | Source ↗official | |
| FLAMESsafety | #3 / 13 | ↑51.05 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #7 / 33 | ↑29.74 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #9 / 33 | ↑23.45 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #7 / 33 | ↑29.36 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #8 / 33 | ↑27.96 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #7 / 33 | ↑31.79 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #7 / 33 | ↑27.27 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #2 / 33 | ↑98.89 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #1 / 33 | ↑99.8 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #5 / 33 | ↑98.44 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #6 / 33 | ↑97 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #8 / 33 | ↑93.39 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #2 / 33 | ↑96.36 | Source ↗official | |
| SALAD-Benchmcq_human_autonomy_integrity | #29 / 33 | ↑3.333 | Source ↗official | |
| SALAD-Benchmcq_information_safety_harms | #27 / 33 | ↑4.167 | Source ↗official | |
| SALAD-Benchmcq_malicious_use | #29 / 33 | ↑3.974 | Source ↗official | |
| SALAD-Benchmcq_misinformation_harms | #29 / 33 | ↑3.333 | Source ↗official | |
| SALAD-Benchmcq_representation_toxicity | #29 / 33 | ↑3.229 | Source ↗official | |
| SALAD-Benchmcq_socioeconomic_harms | #27 / 33 | ↑7.778 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #9 / 31 | ↑68.97 | Source ↗official | |
| SuperCLUE Safetyresponsible_ai | #11 / 31 | ↑65.45 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #25 / 31 | ↑68.09 | 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 |
|---|---|---|---|---|
| Fake Alignment (FINE)open_ended_safe_response_rate | #7 / 14 | ↓96 | Source ↗official | |
| FLAMESdata_protection | #13 / 13 | ↓63.16 | Source ↗official | |
| FLAMESfairness | #13 / 13 | ↓52.61 | Source ↗official | |
| FLAMESlegality | #12 / 13 | ↓71.74 | Source ↗official | |
| FLAMESmorality | #12 / 13 | ↓54.23 | Source ↗official | |
| FLAMESsafety | #11 / 13 | ↓51.05 | Source ↗official | |
| SALAD-Benchattack_enhanced_human_autonomy_integrity | #27 / 33 | ↓29.74 | Source ↗official | |
| SALAD-Benchattack_enhanced_information_safety_harms | #25 / 33 | ↓23.45 | Source ↗official | |
| SALAD-Benchattack_enhanced_malicious_use | #27 / 33 | ↓29.36 | Source ↗official | |
| SALAD-Benchattack_enhanced_misinformation_harms | #26 / 33 | ↓27.96 | Source ↗official | |
| SALAD-Benchattack_enhanced_representation_toxicity | #27 / 33 | ↓31.79 | Source ↗official | |
| SALAD-Benchattack_enhanced_socioeconomic_harms | #27 / 33 | ↓27.27 | Source ↗official | |
| SALAD-Benchbase_human_autonomy_integrity | #32 / 33 | ↓98.89 | Source ↗official | |
| SALAD-Benchbase_information_safety_harms | #33 / 33 | ↓99.8 | Source ↗official | |
| SALAD-Benchbase_malicious_use | #29 / 33 | ↓98.44 | Source ↗official | |
| SALAD-Benchbase_misinformation_harms | #28 / 33 | ↓97 | Source ↗official | |
| SALAD-Benchbase_representation_toxicity | #26 / 33 | ↓93.39 | Source ↗official | |
| SALAD-Benchbase_socioeconomic_harms | #32 / 33 | ↓96.36 | Source ↗official | |
| SuperCLUE Safetyinstruction_attack | #18 / 31 | ↓68.97 | Source ↗official | |
| SuperCLUE Safetytraditional_safety | #5 / 31 | ↓68.09 | Source ↗official |