{ "steps": 40, "unique_training_texts": 372, "losses": [ 14.162500381469727, 12.354166984558105, 12.982142448425293, 12.465909004211426, 12.401785850524902, 11.859375, 12.511363983154297, 12.65625, 10.3125, 12.0, 11.401785850524902, 11.10546875, 13.330357551574707, 10.870535850524902, 10.474431991577148, 11.971590995788574, 8.4921875, 11.274999618530273, 10.290178298950195, 8.956250190734863, 10.924107551574707, 10.133928298950195, 10.276785850524902, 9.416666984558105, 11.010416984558105, 10.364583015441895, 8.723958015441895, 9.328125, 10.334821701049805, 9.627604484558105, 9.754687309265137, 5.798295497894287, 10.287946701049805, 11.689583778381348, 10.24609375, 9.551562309265137, 7.107954502105713, 9.4765625, 7.7414774894714355, 12.395833015441895 ], "encoder_parameter_delta_l2": 0.06532496213912964, "seconds": 5.581527638000011, "newly_initialized_mlm_head": true, "limitation": "Small synthetic corpus and new MLM head; not large-scale continued pretraining" }