75 lines
2.0 KiB
JSON
75 lines
2.0 KiB
JSON
|
{
|
||
|
|
"models": {
|
||
|
|
"student": "unsloth/Qwen2.5-3B-Instruct-bnb-4bit",
|
||
|
|
"teacher": "Qwen/Qwen2.5-14B-Instruct"
|
||
|
|
},
|
||
|
|
"n_train_examples": 1471,
|
||
|
|
"n_val_examples": 127,
|
||
|
|
"held_out_validity": 1.0,
|
||
|
|
"held_out_avg_claims": 2.45,
|
||
|
|
"base": {
|
||
|
|
"n": 15,
|
||
|
|
"validity": 1.0,
|
||
|
|
"avg_claims": 2.067,
|
||
|
|
"avg_verification_questions": 3.71,
|
||
|
|
"pct_claims_with_3plus_questions": 1,
|
||
|
|
"contradiction_recall": 1.0,
|
||
|
|
"date_extraction_rate": 1.0,
|
||
|
|
"evidence_verbatim_rate": 1.0,
|
||
|
|
"avg_claim_words": 5.548,
|
||
|
|
"sarcasm_handling": 0.75,
|
||
|
|
"avg_latency_s": 30.214
|
||
|
|
},
|
||
|
|
"teacher": {
|
||
|
|
"n": 15,
|
||
|
|
"validity": 0.933,
|
||
|
|
"avg_claims": 2.357,
|
||
|
|
"avg_verification_questions": 3.242,
|
||
|
|
"pct_claims_with_3plus_questions": 1,
|
||
|
|
"contradiction_recall": 1.0,
|
||
|
|
"date_extraction_rate": 1.0,
|
||
|
|
"evidence_verbatim_rate": 1.0,
|
||
|
|
"avg_claim_words": 7.333,
|
||
|
|
"sarcasm_handling": 1.0,
|
||
|
|
"avg_latency_s": 60.618
|
||
|
|
},
|
||
|
|
"finetuned": {
|
||
|
|
"n": 15,
|
||
|
|
"validity": 1.0,
|
||
|
|
"avg_claims": 2.067,
|
||
|
|
"avg_verification_questions": 3,
|
||
|
|
"pct_claims_with_3plus_questions": 1,
|
||
|
|
"contradiction_recall": 0.75,
|
||
|
|
"date_extraction_rate": 1.0,
|
||
|
|
"evidence_verbatim_rate": 1.0,
|
||
|
|
"avg_claim_words": 7.806,
|
||
|
|
"sarcasm_handling": 1.0,
|
||
|
|
"avg_latency_s": 25.992
|
||
|
|
},
|
||
|
|
"improvement_vs_base": {
|
||
|
|
"validity": 0.0,
|
||
|
|
"avg_claims": 0.0,
|
||
|
|
"avg_verification_questions": -0.71,
|
||
|
|
"pct_claims_with_3plus_questions": 0,
|
||
|
|
"contradiction_recall": -0.25,
|
||
|
|
"date_extraction_rate": 0.0,
|
||
|
|
"evidence_verbatim_rate": 0.0,
|
||
|
|
"sarcasm_handling": 0.25,
|
||
|
|
"avg_claim_words": 2.258,
|
||
|
|
"avg_latency_s": -4.222
|
||
|
|
},
|
||
|
|
"finetuned_vs_teacher": {
|
||
|
|
"validity": 0.067,
|
||
|
|
"avg_claims": -0.29,
|
||
|
|
"avg_verification_questions": -0.242,
|
||
|
|
"pct_claims_with_3plus_questions": 0,
|
||
|
|
"contradiction_recall": -0.25,
|
||
|
|
"date_extraction_rate": 0.0,
|
||
|
|
"evidence_verbatim_rate": 0.0,
|
||
|
|
"sarcasm_handling": 0.0,
|
||
|
|
"avg_claim_words": 0.473,
|
||
|
|
"avg_latency_s": -34.626
|
||
|
|
},
|
||
|
|
"n_benchmark_cases": 15,
|
||
|
|
"timestamp": "2026-06-07 19:56:00"
|
||
|
|
}
|