f0e86b7433
- seeds/regional/: 1,223 cultural/regional seed files across 50+ regions
- seeds/expansions/: 8 expansion rounds (r1-r8) with raw text and JSON
- seeds/lem-{africa,cn,de,en,eu,me}-all-seeds.json: consolidated by region
- scripts/: Gemini generators, HF push, model comparison (tokens via env vars)
- paper/hf-cards/: HuggingFace model cards for cross-arch models
- benchmarks/benchmark_summary.json: processed PTSD summary data
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
217 lines
No EOL
4.1 KiB
JSON
217 lines
No EOL
4.1 KiB
JSON
{
|
|
"cross_arch": {
|
|
"Llama 3.1 8B": {
|
|
"gsm8k": {
|
|
"base": 68.0,
|
|
"lek": 68.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 8.16,
|
|
"lek": 8.62
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 7.34,
|
|
"lek": 7.68
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 9.5,
|
|
"lek": 9.48
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 6.52,
|
|
"lek": 7.12
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.06,
|
|
"lek": 8.2
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 7.22,
|
|
"lek": 7.32
|
|
}
|
|
},
|
|
"Qwen 2.5 7B": {
|
|
"gsm8k": {
|
|
"base": 70.0,
|
|
"lek": 76.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 8.92,
|
|
"lek": 8.9
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 7.56,
|
|
"lek": 7.58
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 9.18,
|
|
"lek": 9.14
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 8.34,
|
|
"lek": 8.34
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.36,
|
|
"lek": 8.4
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 7.6,
|
|
"lek": 7.58
|
|
}
|
|
},
|
|
"Mistral 7B v0.3": {
|
|
"gsm8k": {
|
|
"base": 24.0,
|
|
"lek": 28.000000000000004
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 9.06,
|
|
"lek": 8.7
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 7.9,
|
|
"lek": 7.9
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 9.18,
|
|
"lek": 8.6
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 8.34,
|
|
"lek": 8.14
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.32,
|
|
"lek": 7.6
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 7.46,
|
|
"lek": 7.56
|
|
}
|
|
}
|
|
},
|
|
"scale": {
|
|
"Gemma 3 1B": {
|
|
"gsm8k": {
|
|
"base": 44.0,
|
|
"lek": 38.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 5.12,
|
|
"lek": 4.76
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 4.82,
|
|
"lek": 4.24
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 8.34,
|
|
"lek": 8.4
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 7.7,
|
|
"lek": 7.54
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.12,
|
|
"lek": 8.2
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 7.8,
|
|
"lek": 7.84
|
|
}
|
|
},
|
|
"Gemma 3 4B": {
|
|
"gsm8k": {
|
|
"base": 76.0,
|
|
"lek": 72.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 6.46,
|
|
"lek": 6.67
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 6.06,
|
|
"lek": 6.3660000000000005
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 8.78,
|
|
"lek": 8.82
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 8.06,
|
|
"lek": 7.96
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.68,
|
|
"lek": 8.74
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 8.5,
|
|
"lek": 8.72
|
|
}
|
|
},
|
|
"Gemma 3 12B": {
|
|
"gsm8k": {
|
|
"base": 90.0,
|
|
"lek": 88.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 8.16,
|
|
"lek": 8.3
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 6.2,
|
|
"lek": 6.52
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 9.0,
|
|
"lek": 9.04
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 7.9,
|
|
"lek": 8.06
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.72,
|
|
"lek": 8.52
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 8.6,
|
|
"lek": 8.7
|
|
}
|
|
},
|
|
"Gemma 3 27B": {
|
|
"gsm8k": {
|
|
"base": 92.0,
|
|
"lek": 92.0
|
|
},
|
|
"truthfulqa_truthfulness": {
|
|
"base": 8.44,
|
|
"lek": 8.36
|
|
},
|
|
"truthfulqa_informativeness": {
|
|
"base": 7.82,
|
|
"lek": 7.76
|
|
},
|
|
"do_not_answer_safety": {
|
|
"base": 8.78,
|
|
"lek": 8.86
|
|
},
|
|
"do_not_answer_nuance": {
|
|
"base": 8.02,
|
|
"lek": 8.06
|
|
},
|
|
"toxigen_kindness": {
|
|
"base": 8.72,
|
|
"lek": 8.72
|
|
},
|
|
"toxigen_awareness": {
|
|
"base": 8.62,
|
|
"lek": 8.66
|
|
}
|
|
}
|
|
},
|
|
"ethical": {}
|
|
} |