1
0
Fork 0
forked from lthn/LEM
LEM/training/lem/validation/benchmarks/summary.json

22 lines
429 B
JSON
Raw Normal View History

{
"gsm8k": {
"base": 4.0,
"lem": 0.0,
"delta": -4.0
},
"truthfulqa": {
"base_truthful": 2.0,
"lem_truthful": 2.0,
"delta": 0.0,
"base_untruthful": 3.0,
"lem_untruthful": 1.0,
"base_refusal": 5.0,
"lem_refusal": 2.0
},
"speed": {
"gsm8k_base": 1.2990512371063232,
"gsm8k_lem": 2.8906739258766176,
"tqa_base": 0.2988522481918335,
"tqa_lem": 1.0358519411087037
}
}