Datasets:
| { | |
| "zero": { | |
| "forget": 0.40775100725169905, | |
| "retain": 0.3746747171350118 | |
| }, | |
| "one": { | |
| "forget": 1.0, | |
| "retain": 0.998625 | |
| }, | |
| "grid": [ | |
| { | |
| "lr": 1e-05, | |
| "forget_rougeL": 0.5254876783802431, | |
| "retain_rougeL": 0.6107671879117822, | |
| "forget_retention": 0.19879589930950006, | |
| "retain_retention": 0.3783834662157796 | |
| }, | |
| { | |
| "lr": 3e-05, | |
| "forget_rougeL": 0.4967327937709532, | |
| "retain_rougeL": 0.8789141488648653, | |
| "forget_retention": 0.15024387986940888, | |
| "retain_retention": 0.8081404008898608 | |
| } | |
| ], | |
| "rule": "C2 = the smallest lr in lr_grid whose merged model has forget retention <= 0.25 (all 20 forget10 authors) and retain retention >= 0.50. If no lr meets both, C2 = the lr with the lowest forget retention among those with retain retention >= 0.50, and the entry says the target was missed. If every lr has retain retention < 0.50, stop and report (no C2).", | |
| "chosen_lr": 3e-05, | |
| "target_met": true | |
| } |