| {"event": "start", "run": "baseline_en", "condition": "A", "params": 22516992, "langs": ["en"], "premium": {"en": 1.0}, "schedule": {"tau1": 0.4, "tau2": 0.7, "p1_objective": "global", "p2_objective": "global", "p3_objective": "global", "freeze_in_p3": false, "freeze_scope": "router"}, "corpus": "/local/scratch/sas245/beetle-hnet/scratch/data/tier1_standin", "weights": {"en": 1.0}, "t": 1.3} |
| {"event": "probe", "step": 0, "phase": 1, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 4884.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.361697, "cps": 50.875, "mean_p": 0.496347}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 2.767404, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 1.5} |
| {"event": "train", "step": 0, "phase": 1, "objective": "global", "lm_loss": 5.66441, "bpb": 8.17201, "chunk_loss": 0.040128, "F": 0.30981, "G": 0.49124, "chunks": 317.25, "bytes_per_chunk": 3.228, "grad_norm": 12.0832, "lr": 1.5e-06, "router_trainable": true, "sec_per_step": 0.015, "alpha": 0.03, "t": 2.4} |
| {"event": "train", "step": 50, "phase": 1, "objective": "global", "lm_loss": 3.96706, "bpb": 5.72326, "chunk_loss": 0.031702, "F": 0.1907, "G": 0.49023, "chunks": 195.28, "bytes_per_chunk": 5.244, "grad_norm": 2.3639, "lr": 7.65e-05, "router_trainable": true, "sec_per_step": 0.051, "alpha": 0.03, "t": 4.9} |
| {"event": "train", "step": 100, "phase": 1, "objective": "global", "lm_loss": 2.93869, "bpb": 4.23963, "chunk_loss": 0.031852, "F": 0.20514, "G": 0.38478, "chunks": 210.06, "bytes_per_chunk": 4.875, "grad_norm": 3.0762, "lr": 0.0001515, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 7.1} |
| {"event": "train", "step": 150, "phase": 1, "objective": "global", "lm_loss": 2.59506, "bpb": 3.74388, "chunk_loss": 0.030458, "F": 0.17642, "G": 0.36688, "chunks": 180.66, "bytes_per_chunk": 5.668, "grad_norm": 1.6857, "lr": 0.0002265, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 9.1} |
| {"event": "train", "step": 200, "phase": 1, "objective": "global", "lm_loss": 2.36746, "bpb": 3.41553, "chunk_loss": 0.031466, "F": 0.20374, "G": 0.345, "chunks": 208.62, "bytes_per_chunk": 4.908, "grad_norm": 2.4918, "lr": 0.0003, "router_trainable": true, "sec_per_step": 0.039, "alpha": 0.03, "t": 11.0} |
| {"event": "eval", "step": 250, "phase": 1, "bpb": {"en": 3.17185}, "bytes_per_chunk": {"en": 4.69}, "bpb_mean": 3.17185, "t": 13.1} |
| {"event": "train", "step": 250, "phase": 1, "objective": "global", "lm_loss": 2.14722, "bpb": 3.09779, "chunk_loss": 0.031083, "F": 0.21036, "G": 0.27837, "chunks": 215.41, "bytes_per_chunk": 4.754, "grad_norm": 1.3424, "lr": 0.0002999, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 13.1} |
| {"event": "train", "step": 300, "phase": 1, "objective": "global", "lm_loss": 2.1025, "bpb": 3.03326, "chunk_loss": 0.031044, "F": 0.21567, "G": 0.26223, "chunks": 220.84, "bytes_per_chunk": 4.637, "grad_norm": 1.2806, "lr": 0.0002997, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 15.1} |
| {"event": "train", "step": 350, "phase": 1, "objective": "global", "lm_loss": 2.01715, "bpb": 2.91013, "chunk_loss": 0.03129, "F": 0.2258, "G": 0.26442, "chunks": 231.22, "bytes_per_chunk": 4.429, "grad_norm": 1.1319, "lr": 0.0002993, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 17.1} |
| {"event": "train", "step": 400, "phase": 1, "objective": "global", "lm_loss": 1.92618, "bpb": 2.7789, "chunk_loss": 0.031034, "F": 0.22089, "G": 0.25166, "chunks": 226.19, "bytes_per_chunk": 4.527, "grad_norm": 1.4106, "lr": 0.0002988, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 19.2} |
| {"event": "train", "step": 450, "phase": 1, "objective": "global", "lm_loss": 1.80237, "bpb": 2.60028, "chunk_loss": 0.030876, "F": 0.21829, "G": 0.24335, "chunks": 223.53, "bytes_per_chunk": 4.581, "grad_norm": 1.3423, "lr": 0.0002982, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 21.2} |
| {"event": "probe", "step": 500, "phase": 1, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2879.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.213212, "cps": 29.989583, "mean_p": 0.238603}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 4.694686, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 23.2} |
| {"event": "eval", "step": 500, "phase": 1, "bpb": {"en": 2.64785}, "bytes_per_chunk": {"en": 4.472}, "bpb_mean": 2.64785, "t": 23.4} |
| {"event": "train", "step": 500, "phase": 1, "objective": "global", "lm_loss": 1.76139, "bpb": 2.54115, "chunk_loss": 0.030944, "F": 0.22131, "G": 0.24395, "chunks": 226.62, "bytes_per_chunk": 4.518, "grad_norm": 1.5029, "lr": 0.0002974, "router_trainable": true, "sec_per_step": 0.046, "alpha": 0.03, "t": 23.5} |
| {"event": "train", "step": 550, "phase": 1, "objective": "global", "lm_loss": 1.76605, "bpb": 2.54787, "chunk_loss": 0.030994, "F": 0.22476, "G": 0.24278, "chunks": 230.16, "bytes_per_chunk": 4.449, "grad_norm": 1.1989, "lr": 0.0002965, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 25.5} |
| {"event": "train", "step": 600, "phase": 1, "objective": "global", "lm_loss": 1.68769, "bpb": 2.43482, "chunk_loss": 0.03111, "F": 0.22791, "G": 0.24679, "chunks": 233.38, "bytes_per_chunk": 4.388, "grad_norm": 1.3439, "lr": 0.0002954, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 27.5} |
| {"event": "train", "step": 650, "phase": 1, "objective": "global", "lm_loss": 1.68067, "bpb": 2.4247, "chunk_loss": 0.030982, "F": 0.22586, "G": 0.24037, "chunks": 231.28, "bytes_per_chunk": 4.428, "grad_norm": 1.0878, "lr": 0.0002942, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 29.6} |
| {"event": "train", "step": 700, "phase": 1, "objective": "global", "lm_loss": 1.61067, "bpb": 2.32371, "chunk_loss": 0.030945, "F": 0.22562, "G": 0.23685, "chunks": 231.03, "bytes_per_chunk": 4.432, "grad_norm": 1.1877, "lr": 0.0002928, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 31.6} |
| {"event": "eval", "step": 750, "phase": 1, "bpb": {"en": 2.43375}, "bytes_per_chunk": {"en": 4.47}, "bpb_mean": 2.43375, "t": 33.8} |
| {"event": "train", "step": 750, "phase": 1, "objective": "global", "lm_loss": 1.60258, "bpb": 2.31203, "chunk_loss": 0.030729, "F": 0.21658, "G": 0.23002, "chunks": 221.78, "bytes_per_chunk": 4.617, "grad_norm": 1.2023, "lr": 0.0002913, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 33.8} |
| {"event": "train", "step": 800, "phase": 1, "objective": "global", "lm_loss": 1.62759, "bpb": 2.34812, "chunk_loss": 0.030837, "F": 0.22229, "G": 0.23337, "chunks": 227.62, "bytes_per_chunk": 4.499, "grad_norm": 1.1165, "lr": 0.0002897, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 35.8} |
| {"event": "train", "step": 850, "phase": 1, "objective": "global", "lm_loss": 1.63247, "bpb": 2.35516, "chunk_loss": 0.030938, "F": 0.22522, "G": 0.23781, "chunks": 230.62, "bytes_per_chunk": 4.44, "grad_norm": 0.9742, "lr": 0.000288, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 37.9} |
| {"event": "train", "step": 900, "phase": 1, "objective": "global", "lm_loss": 1.63538, "bpb": 2.35936, "chunk_loss": 0.030898, "F": 0.22281, "G": 0.2353, "chunks": 228.16, "bytes_per_chunk": 4.488, "grad_norm": 1.0922, "lr": 0.0002861, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 39.9} |
| {"event": "train", "step": 950, "phase": 1, "objective": "global", "lm_loss": 1.57397, "bpb": 2.27076, "chunk_loss": 0.030926, "F": 0.22537, "G": 0.23539, "chunks": 230.78, "bytes_per_chunk": 4.437, "grad_norm": 0.9995, "lr": 0.0002841, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 42.0} |
| {"event": "probe", "step": 1000, "phase": 1, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2886.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.21373, "cps": 30.0625, "mean_p": 0.22402}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 4.683299, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 44.0} |
| {"event": "eval", "step": 1000, "phase": 1, "bpb": {"en": 2.32062}, "bytes_per_chunk": {"en": 4.436}, "bpb_mean": 2.32062, "t": 44.2} |
| {"event": "train", "step": 1000, "phase": 1, "objective": "global", "lm_loss": 1.60957, "bpb": 2.32212, "chunk_loss": 0.030921, "F": 0.22507, "G": 0.23436, "chunks": 230.47, "bytes_per_chunk": 4.443, "grad_norm": 1.0573, "lr": 0.0002819, "router_trainable": true, "sec_per_step": 0.046, "alpha": 0.03, "t": 44.3} |
| {"event": "train", "step": 1050, "phase": 1, "objective": "global", "lm_loss": 1.53486, "bpb": 2.21433, "chunk_loss": 0.03085, "F": 0.22391, "G": 0.23217, "chunks": 229.28, "bytes_per_chunk": 4.466, "grad_norm": 1.0475, "lr": 0.0002796, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 46.3} |
| {"event": "train", "step": 1100, "phase": 1, "objective": "global", "lm_loss": 1.5074, "bpb": 2.17472, "chunk_loss": 0.030865, "F": 0.22455, "G": 0.23175, "chunks": 229.94, "bytes_per_chunk": 4.453, "grad_norm": 0.9906, "lr": 0.0002772, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 48.4} |
| {"event": "train", "step": 1150, "phase": 1, "objective": "global", "lm_loss": 1.48296, "bpb": 2.13947, "chunk_loss": 0.030705, "F": 0.21826, "G": 0.22673, "chunks": 223.5, "bytes_per_chunk": 4.582, "grad_norm": 0.8906, "lr": 0.0002747, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 50.4} |
| {"event": "train", "step": 1200, "phase": 1, "objective": "global", "lm_loss": 1.53887, "bpb": 2.22012, "chunk_loss": 0.030739, "F": 0.21878, "G": 0.22682, "chunks": 224.03, "bytes_per_chunk": 4.571, "grad_norm": 0.8517, "lr": 0.0002721, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 52.5} |
| {"event": "eval", "step": 1250, "phase": 1, "bpb": {"en": 2.2299}, "bytes_per_chunk": {"en": 4.506}, "bpb_mean": 2.2299, "t": 54.6} |
| {"event": "train", "step": 1250, "phase": 1, "objective": "global", "lm_loss": 1.47402, "bpb": 2.12656, "chunk_loss": 0.03093, "F": 0.22766, "G": 0.23381, "chunks": 233.12, "bytes_per_chunk": 4.392, "grad_norm": 0.8293, "lr": 0.0002694, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 54.6} |
| {"event": "train", "step": 1300, "phase": 1, "objective": "global", "lm_loss": 1.54796, "bpb": 2.23324, "chunk_loss": 0.030811, "F": 0.21948, "G": 0.23125, "chunks": 224.75, "bytes_per_chunk": 4.556, "grad_norm": 0.8633, "lr": 0.0002665, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 56.6} |
| {"event": "train", "step": 1350, "phase": 1, "objective": "global", "lm_loss": 1.5163, "bpb": 2.18756, "chunk_loss": 0.030789, "F": 0.21841, "G": 0.22875, "chunks": 223.66, "bytes_per_chunk": 4.578, "grad_norm": 0.838, "lr": 0.0002635, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 58.7} |
| {"event": "train", "step": 1400, "phase": 1, "objective": "global", "lm_loss": 1.4602, "bpb": 2.10663, "chunk_loss": 0.030696, "F": 0.21671, "G": 0.22668, "chunks": 221.91, "bytes_per_chunk": 4.615, "grad_norm": 0.8617, "lr": 0.0002605, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 60.7} |
| {"event": "train", "step": 1450, "phase": 1, "objective": "global", "lm_loss": 1.47435, "bpb": 2.12704, "chunk_loss": 0.030535, "F": 0.20804, "G": 0.22005, "chunks": 213.03, "bytes_per_chunk": 4.807, "grad_norm": 0.7718, "lr": 0.0002573, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 62.7} |
| {"event": "probe", "step": 1500, "phase": 1, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2707.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.200474, "cps": 28.197917, "mean_p": 0.213113}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 4.992981, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 64.7} |
| {"event": "eval", "step": 1500, "phase": 1, "bpb": {"en": 2.17371}, "bytes_per_chunk": {"en": 4.702}, "bpb_mean": 2.17371, "t": 65.0} |
| {"event": "train", "step": 1500, "phase": 1, "objective": "global", "lm_loss": 1.44903, "bpb": 2.09051, "chunk_loss": 0.030599, "F": 0.21173, "G": 0.22301, "chunks": 216.81, "bytes_per_chunk": 4.723, "grad_norm": 0.8913, "lr": 0.000254, "router_trainable": true, "sec_per_step": 0.045, "alpha": 0.03, "t": 65.0} |
| {"event": "train", "step": 1550, "phase": 1, "objective": "global", "lm_loss": 1.44784, "bpb": 2.08879, "chunk_loss": 0.030459, "F": 0.20609, "G": 0.21561, "chunks": 211.03, "bytes_per_chunk": 4.852, "grad_norm": 0.8538, "lr": 0.0002506, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 67.0} |
| {"event": "train", "step": 1600, "phase": 1, "objective": "global", "lm_loss": 1.49386, "bpb": 2.15519, "chunk_loss": 0.030592, "F": 0.21103, "G": 0.22242, "chunks": 216.09, "bytes_per_chunk": 4.739, "grad_norm": 0.8883, "lr": 0.0002472, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 69.1} |
| {"event": "train", "step": 1650, "phase": 1, "objective": "global", "lm_loss": 1.39083, "bpb": 2.00654, "chunk_loss": 0.030619, "F": 0.20911, "G": 0.21716, "chunks": 214.12, "bytes_per_chunk": 4.782, "grad_norm": 0.7979, "lr": 0.0002436, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 71.1} |
| {"event": "train", "step": 1700, "phase": 1, "objective": "global", "lm_loss": 1.40329, "bpb": 2.02451, "chunk_loss": 0.030517, "F": 0.20947, "G": 0.21852, "chunks": 214.5, "bytes_per_chunk": 4.774, "grad_norm": 0.7993, "lr": 0.00024, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 73.1} |
| {"event": "eval", "step": 1750, "phase": 1, "bpb": {"en": 2.12451}, "bytes_per_chunk": {"en": 4.744}, "bpb_mean": 2.12451, "t": 75.2} |
| {"event": "train", "step": 1750, "phase": 1, "objective": "global", "lm_loss": 1.41237, "bpb": 2.03762, "chunk_loss": 0.030544, "F": 0.20901, "G": 0.21875, "chunks": 214.03, "bytes_per_chunk": 4.784, "grad_norm": 0.8677, "lr": 0.0002363, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 75.2} |
| {"event": "train", "step": 1800, "phase": 1, "objective": "global", "lm_loss": 1.39262, "bpb": 2.00913, "chunk_loss": 0.030542, "F": 0.21033, "G": 0.21926, "chunks": 215.38, "bytes_per_chunk": 4.754, "grad_norm": 0.8533, "lr": 0.0002325, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 77.3} |
| {"event": "train", "step": 1850, "phase": 1, "objective": "global", "lm_loss": 1.39702, "bpb": 2.01547, "chunk_loss": 0.03052, "F": 0.20917, "G": 0.21769, "chunks": 214.19, "bytes_per_chunk": 4.781, "grad_norm": 0.7715, "lr": 0.0002286, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 79.3} |
| {"event": "train", "step": 1900, "phase": 1, "objective": "global", "lm_loss": 1.38118, "bpb": 1.99262, "chunk_loss": 0.030695, "F": 0.21619, "G": 0.22317, "chunks": 221.38, "bytes_per_chunk": 4.626, "grad_norm": 0.7694, "lr": 0.0002247, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 81.3} |
| {"event": "train", "step": 1950, "phase": 1, "objective": "global", "lm_loss": 1.33808, "bpb": 1.93044, "chunk_loss": 0.030462, "F": 0.20746, "G": 0.2145, "chunks": 212.44, "bytes_per_chunk": 4.82, "grad_norm": 0.8269, "lr": 0.0002207, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 83.4} |
| {"event": "probe", "step": 2000, "phase": 2, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2664.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.197289, "cps": 27.75, "mean_p": 0.208036}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.073574, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 85.4} |
| {"event": "eval", "step": 2000, "phase": 1, "bpb": {"en": 2.05003}, "bytes_per_chunk": {"en": 4.733}, "bpb_mean": 2.05003, "t": 85.6} |
| {"event": "phase_transition", "step": 2000, "phase": 2, "objective": "global", "router_trainable": true, "encoder_trainable": true, "frozen_router_params": 0, "t": 85.6} |
| {"event": "train", "step": 2000, "phase": 2, "objective": "global", "lm_loss": 1.394, "bpb": 2.01112, "chunk_loss": 0.030635, "F": 0.21103, "G": 0.21899, "chunks": 216.09, "bytes_per_chunk": 4.739, "grad_norm": 0.7798, "lr": 0.0002167, "router_trainable": true, "sec_per_step": 0.045, "alpha": 0.03, "t": 85.6} |
| {"event": "train", "step": 2050, "phase": 2, "objective": "global", "lm_loss": 1.3946, "bpb": 2.01199, "chunk_loss": 0.030557, "F": 0.20926, "G": 0.21781, "chunks": 214.28, "bytes_per_chunk": 4.779, "grad_norm": 0.7824, "lr": 0.0002126, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 87.7} |
| {"event": "train", "step": 2100, "phase": 2, "objective": "global", "lm_loss": 1.3849, "bpb": 1.99799, "chunk_loss": 0.030438, "F": 0.20569, "G": 0.21416, "chunks": 210.62, "bytes_per_chunk": 4.862, "grad_norm": 0.8147, "lr": 0.0002084, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 89.7} |
| {"event": "train", "step": 2150, "phase": 2, "objective": "global", "lm_loss": 1.40408, "bpb": 2.02565, "chunk_loss": 0.030815, "F": 0.21942, "G": 0.22522, "chunks": 224.69, "bytes_per_chunk": 4.557, "grad_norm": 0.7288, "lr": 0.0002042, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 91.8} |
| {"event": "train", "step": 2200, "phase": 2, "objective": "global", "lm_loss": 1.35356, "bpb": 1.95278, "chunk_loss": 0.03052, "F": 0.20929, "G": 0.21559, "chunks": 214.31, "bytes_per_chunk": 4.778, "grad_norm": 0.8509, "lr": 0.0001999, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 93.8} |
| {"event": "eval", "step": 2250, "phase": 2, "bpb": {"en": 1.99956}, "bytes_per_chunk": {"en": 4.682}, "bpb_mean": 1.99956, "t": 95.9} |
| {"event": "train", "step": 2250, "phase": 2, "objective": "global", "lm_loss": 1.34643, "bpb": 1.94248, "chunk_loss": 0.030686, "F": 0.21515, "G": 0.22121, "chunks": 220.31, "bytes_per_chunk": 4.648, "grad_norm": 0.7659, "lr": 0.0001957, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 96.0} |
| {"event": "train", "step": 2300, "phase": 2, "objective": "global", "lm_loss": 1.30389, "bpb": 1.88111, "chunk_loss": 0.030484, "F": 0.20728, "G": 0.21505, "chunks": 212.25, "bytes_per_chunk": 4.824, "grad_norm": 0.7552, "lr": 0.0001913, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 98.0} |
| {"event": "train", "step": 2350, "phase": 2, "objective": "global", "lm_loss": 1.3369, "bpb": 1.92874, "chunk_loss": 0.030639, "F": 0.21274, "G": 0.21829, "chunks": 217.84, "bytes_per_chunk": 4.701, "grad_norm": 0.7572, "lr": 0.000187, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 100.0} |
| {"event": "train", "step": 2400, "phase": 2, "objective": "global", "lm_loss": 1.35317, "bpb": 1.95221, "chunk_loss": 0.030479, "F": 0.20761, "G": 0.2132, "chunks": 212.59, "bytes_per_chunk": 4.817, "grad_norm": 0.7234, "lr": 0.0001826, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 102.1} |
| {"event": "train", "step": 2450, "phase": 2, "objective": "global", "lm_loss": 1.3337, "bpb": 1.92413, "chunk_loss": 0.030593, "F": 0.21326, "G": 0.21969, "chunks": 218.38, "bytes_per_chunk": 4.689, "grad_norm": 0.8154, "lr": 0.0001782, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 104.2} |
| {"event": "probe", "step": 2500, "phase": 2, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2676.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.198178, "cps": 27.875, "mean_p": 0.206845}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.050822, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 106.2} |
| {"event": "eval", "step": 2500, "phase": 2, "bpb": {"en": 1.95686}, "bytes_per_chunk": {"en": 4.734}, "bpb_mean": 1.95686, "t": 106.4} |
| {"event": "train", "step": 2500, "phase": 2, "objective": "global", "lm_loss": 1.29957, "bpb": 1.87489, "chunk_loss": 0.030607, "F": 0.21335, "G": 0.22061, "chunks": 218.47, "bytes_per_chunk": 4.687, "grad_norm": 0.8269, "lr": 0.0001738, "router_trainable": true, "sec_per_step": 0.046, "alpha": 0.03, "t": 106.5} |
| {"event": "train", "step": 2550, "phase": 2, "objective": "global", "lm_loss": 1.32449, "bpb": 1.91083, "chunk_loss": 0.030647, "F": 0.215, "G": 0.22051, "chunks": 220.16, "bytes_per_chunk": 4.651, "grad_norm": 0.8431, "lr": 0.0001694, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 108.5} |
| {"event": "train", "step": 2600, "phase": 2, "objective": "global", "lm_loss": 1.28927, "bpb": 1.86003, "chunk_loss": 0.030502, "F": 0.20987, "G": 0.21497, "chunks": 214.91, "bytes_per_chunk": 4.765, "grad_norm": 0.7621, "lr": 0.000165, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 110.5} |
| {"event": "train", "step": 2650, "phase": 2, "objective": "global", "lm_loss": 1.26064, "bpb": 1.81873, "chunk_loss": 0.030481, "F": 0.20776, "G": 0.21441, "chunks": 212.75, "bytes_per_chunk": 4.813, "grad_norm": 0.7947, "lr": 0.0001606, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 112.6} |
| {"event": "train", "step": 2700, "phase": 2, "objective": "global", "lm_loss": 1.19122, "bpb": 1.71856, "chunk_loss": 0.030482, "F": 0.20706, "G": 0.21302, "chunks": 212.03, "bytes_per_chunk": 4.829, "grad_norm": 0.7729, "lr": 0.0001562, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 114.6} |
| {"event": "eval", "step": 2750, "phase": 2, "bpb": {"en": 1.90295}, "bytes_per_chunk": {"en": 4.737}, "bpb_mean": 1.90295, "t": 116.7} |
| {"event": "train", "step": 2750, "phase": 2, "objective": "global", "lm_loss": 1.3232, "bpb": 1.90897, "chunk_loss": 0.03067, "F": 0.21368, "G": 0.22047, "chunks": 218.81, "bytes_per_chunk": 4.68, "grad_norm": 0.828, "lr": 0.0001518, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 116.7} |
| {"event": "train", "step": 2800, "phase": 2, "objective": "global", "lm_loss": 1.27056, "bpb": 1.83303, "chunk_loss": 0.030499, "F": 0.20639, "G": 0.21206, "chunks": 211.34, "bytes_per_chunk": 4.845, "grad_norm": 0.7814, "lr": 0.0001474, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 118.8} |
| {"event": "train", "step": 2850, "phase": 2, "objective": "global", "lm_loss": 1.28758, "bpb": 1.85758, "chunk_loss": 0.030454, "F": 0.20618, "G": 0.21208, "chunks": 211.12, "bytes_per_chunk": 4.85, "grad_norm": 0.8001, "lr": 0.000143, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 120.8} |
| {"event": "train", "step": 2900, "phase": 2, "objective": "global", "lm_loss": 1.24014, "bpb": 1.78915, "chunk_loss": 0.030438, "F": 0.2049, "G": 0.21119, "chunks": 209.81, "bytes_per_chunk": 4.881, "grad_norm": 0.7737, "lr": 0.0001387, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 122.8} |
| {"event": "train", "step": 2950, "phase": 2, "objective": "global", "lm_loss": 1.22198, "bpb": 1.76295, "chunk_loss": 0.030534, "F": 0.21124, "G": 0.21671, "chunks": 216.31, "bytes_per_chunk": 4.734, "grad_norm": 0.8122, "lr": 0.0001343, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 124.8} |
| {"event": "probe", "step": 3000, "phase": 2, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2676.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.198178, "cps": 27.875, "mean_p": 0.205814}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.050822, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 126.8} |
| {"event": "eval", "step": 3000, "phase": 2, "bpb": {"en": 1.86797}, "bytes_per_chunk": {"en": 4.732}, "bpb_mean": 1.86797, "t": 127.1} |
| {"event": "train", "step": 3000, "phase": 2, "objective": "global", "lm_loss": 1.27078, "bpb": 1.83335, "chunk_loss": 0.030462, "F": 0.2085, "G": 0.21429, "chunks": 213.5, "bytes_per_chunk": 4.796, "grad_norm": 0.8156, "lr": 0.0001301, "router_trainable": true, "sec_per_step": 0.045, "alpha": 0.03, "t": 127.1} |
| {"event": "train", "step": 3050, "phase": 2, "objective": "global", "lm_loss": 1.21044, "bpb": 1.7463, "chunk_loss": 0.030514, "F": 0.20969, "G": 0.2161, "chunks": 214.72, "bytes_per_chunk": 4.769, "grad_norm": 0.8593, "lr": 0.0001258, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 129.1} |
| {"event": "train", "step": 3100, "phase": 2, "objective": "global", "lm_loss": 1.31147, "bpb": 1.89206, "chunk_loss": 0.030653, "F": 0.21628, "G": 0.2214, "chunks": 221.47, "bytes_per_chunk": 4.624, "grad_norm": 0.8762, "lr": 0.0001216, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 131.2} |
| {"event": "train", "step": 3150, "phase": 2, "objective": "global", "lm_loss": 1.29756, "bpb": 1.87199, "chunk_loss": 0.030611, "F": 0.21539, "G": 0.22035, "chunks": 220.56, "bytes_per_chunk": 4.643, "grad_norm": 0.8626, "lr": 0.0001174, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 133.2} |
| {"event": "train", "step": 3200, "phase": 2, "objective": "global", "lm_loss": 1.26023, "bpb": 1.81813, "chunk_loss": 0.030586, "F": 0.21399, "G": 0.21982, "chunks": 219.12, "bytes_per_chunk": 4.673, "grad_norm": 0.903, "lr": 0.0001133, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 135.2} |
| {"event": "eval", "step": 3250, "phase": 2, "bpb": {"en": 1.82717}, "bytes_per_chunk": {"en": 4.729}, "bpb_mean": 1.82717, "t": 137.4} |
| {"event": "train", "step": 3250, "phase": 2, "objective": "global", "lm_loss": 1.23019, "bpb": 1.77479, "chunk_loss": 0.030552, "F": 0.21155, "G": 0.2167, "chunks": 216.62, "bytes_per_chunk": 4.727, "grad_norm": 0.8599, "lr": 0.0001093, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 137.4} |
| {"event": "train", "step": 3300, "phase": 2, "objective": "global", "lm_loss": 1.32118, "bpb": 1.90606, "chunk_loss": 0.030653, "F": 0.21249, "G": 0.21795, "chunks": 217.59, "bytes_per_chunk": 4.706, "grad_norm": 0.8874, "lr": 0.0001053, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 139.5} |
| {"event": "train", "step": 3350, "phase": 2, "objective": "global", "lm_loss": 1.19846, "bpb": 1.72901, "chunk_loss": 0.030442, "F": 0.20755, "G": 0.21312, "chunks": 212.53, "bytes_per_chunk": 4.818, "grad_norm": 0.8238, "lr": 0.0001014, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 141.5} |
| {"event": "train", "step": 3400, "phase": 2, "objective": "global", "lm_loss": 1.20316, "bpb": 1.7358, "chunk_loss": 0.030587, "F": 0.2128, "G": 0.21877, "chunks": 217.91, "bytes_per_chunk": 4.699, "grad_norm": 0.8561, "lr": 9.75e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 143.6} |
| {"event": "train", "step": 3450, "phase": 2, "objective": "global", "lm_loss": 1.24409, "bpb": 1.79485, "chunk_loss": 0.030565, "F": 0.21161, "G": 0.2173, "chunks": 216.69, "bytes_per_chunk": 4.726, "grad_norm": 0.9254, "lr": 9.37e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 145.7} |
| {"event": "probe", "step": 3500, "phase": 3, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2668.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.197586, "cps": 27.791667, "mean_p": 0.204397}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.065967, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 147.7} |
| {"event": "eval", "step": 3500, "phase": 2, "bpb": {"en": 1.78841}, "bytes_per_chunk": {"en": 4.742}, "bpb_mean": 1.78841, "t": 147.9} |
| {"event": "phase_transition", "step": 3500, "phase": 3, "objective": "global", "router_trainable": true, "encoder_trainable": true, "frozen_router_params": 0, "t": 147.9} |
| {"event": "train", "step": 3500, "phase": 3, "objective": "global", "lm_loss": 1.19496, "bpb": 1.72397, "chunk_loss": 0.030473, "F": 0.20816, "G": 0.21329, "chunks": 213.16, "bytes_per_chunk": 4.804, "grad_norm": 0.9069, "lr": 9e-05, "router_trainable": true, "sec_per_step": 0.045, "alpha": 0.03, "t": 147.9} |
| {"event": "train", "step": 3550, "phase": 3, "objective": "global", "lm_loss": 1.15301, "bpb": 1.66345, "chunk_loss": 0.030404, "F": 0.20425, "G": 0.21075, "chunks": 209.16, "bytes_per_chunk": 4.896, "grad_norm": 0.9098, "lr": 8.64e-05, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 149.9} |
| {"event": "train", "step": 3600, "phase": 3, "objective": "global", "lm_loss": 1.15334, "bpb": 1.66392, "chunk_loss": 0.030414, "F": 0.20581, "G": 0.211, "chunks": 210.75, "bytes_per_chunk": 4.859, "grad_norm": 0.901, "lr": 8.28e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 152.0} |
| {"event": "train", "step": 3650, "phase": 3, "objective": "global", "lm_loss": 1.14224, "bpb": 1.6479, "chunk_loss": 0.03045, "F": 0.20596, "G": 0.21129, "chunks": 210.91, "bytes_per_chunk": 4.855, "grad_norm": 0.9649, "lr": 7.94e-05, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 154.1} |
| {"event": "train", "step": 3700, "phase": 3, "objective": "global", "lm_loss": 1.1677, "bpb": 1.68464, "chunk_loss": 0.030596, "F": 0.21204, "G": 0.2175, "chunks": 217.12, "bytes_per_chunk": 4.716, "grad_norm": 0.9024, "lr": 7.6e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 156.1} |
| {"event": "eval", "step": 3750, "phase": 3, "bpb": {"en": 1.74671}, "bytes_per_chunk": {"en": 4.75}, "bpb_mean": 1.74671, "t": 158.3} |
| {"event": "train", "step": 3750, "phase": 3, "objective": "global", "lm_loss": 1.15668, "bpb": 1.66874, "chunk_loss": 0.030608, "F": 0.2135, "G": 0.21959, "chunks": 218.62, "bytes_per_chunk": 4.684, "grad_norm": 0.898, "lr": 7.27e-05, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 158.3} |
| {"event": "train", "step": 3800, "phase": 3, "objective": "global", "lm_loss": 1.1641, "bpb": 1.67945, "chunk_loss": 0.030531, "F": 0.21155, "G": 0.21752, "chunks": 216.62, "bytes_per_chunk": 4.727, "grad_norm": 0.8836, "lr": 6.95e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 160.3} |
| {"event": "train", "step": 3850, "phase": 3, "objective": "global", "lm_loss": 1.22436, "bpb": 1.76637, "chunk_loss": 0.030878, "F": 0.22089, "G": 0.22619, "chunks": 226.19, "bytes_per_chunk": 4.527, "grad_norm": 0.9478, "lr": 6.65e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 162.4} |
| {"event": "train", "step": 3900, "phase": 3, "objective": "global", "lm_loss": 1.1527, "bpb": 1.66299, "chunk_loss": 0.030485, "F": 0.20804, "G": 0.21308, "chunks": 213.03, "bytes_per_chunk": 4.807, "grad_norm": 0.9299, "lr": 6.35e-05, "router_trainable": true, "sec_per_step": 0.04, "alpha": 0.03, "t": 164.4} |
| {"event": "train", "step": 3950, "phase": 3, "objective": "global", "lm_loss": 1.17791, "bpb": 1.69936, "chunk_loss": 0.03048, "F": 0.20859, "G": 0.21356, "chunks": 213.59, "bytes_per_chunk": 4.794, "grad_norm": 0.9686, "lr": 6.06e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 166.4} |
| {"event": "probe", "step": 4000, "phase": 3, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2676.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.198178, "cps": 27.875, "mean_p": 0.204176}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.050822, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 168.4} |
| {"event": "eval", "step": 4000, "phase": 3, "bpb": {"en": 1.72062}, "bytes_per_chunk": {"en": 4.744}, "bpb_mean": 1.72062, "t": 168.7} |
| {"event": "train", "step": 4000, "phase": 3, "objective": "global", "lm_loss": 1.15487, "bpb": 1.66613, "chunk_loss": 0.030475, "F": 0.20819, "G": 0.21339, "chunks": 213.19, "bytes_per_chunk": 4.803, "grad_norm": 0.9138, "lr": 5.79e-05, "router_trainable": true, "sec_per_step": 0.046, "alpha": 0.03, "t": 168.7} |
| {"event": "train", "step": 4050, "phase": 3, "objective": "global", "lm_loss": 1.11953, "bpb": 1.61514, "chunk_loss": 0.030463, "F": 0.20752, "G": 0.21271, "chunks": 212.5, "bytes_per_chunk": 4.819, "grad_norm": 0.9346, "lr": 5.53e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 170.8} |
| {"event": "train", "step": 4100, "phase": 3, "objective": "global", "lm_loss": 1.13507, "bpb": 1.63756, "chunk_loss": 0.030642, "F": 0.2121, "G": 0.2175, "chunks": 217.19, "bytes_per_chunk": 4.715, "grad_norm": 0.9978, "lr": 5.28e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 172.8} |
| {"event": "train", "step": 4150, "phase": 3, "objective": "global", "lm_loss": 1.18182, "bpb": 1.705, "chunk_loss": 0.03057, "F": 0.21243, "G": 0.21815, "chunks": 217.53, "bytes_per_chunk": 4.707, "grad_norm": 0.9594, "lr": 5.04e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 174.8} |
| {"event": "train", "step": 4200, "phase": 3, "objective": "global", "lm_loss": 1.13783, "bpb": 1.64154, "chunk_loss": 0.030584, "F": 0.21237, "G": 0.21749, "chunks": 217.47, "bytes_per_chunk": 4.709, "grad_norm": 0.9636, "lr": 4.81e-05, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 177.0} |
| {"event": "eval", "step": 4250, "phase": 3, "bpb": {"en": 1.6914}, "bytes_per_chunk": {"en": 4.743}, "bpb_mean": 1.6914, "t": 179.2} |
| {"event": "train", "step": 4250, "phase": 3, "objective": "global", "lm_loss": 1.1837, "bpb": 1.70772, "chunk_loss": 0.030569, "F": 0.21277, "G": 0.21797, "chunks": 217.88, "bytes_per_chunk": 4.7, "grad_norm": 0.9683, "lr": 4.59e-05, "router_trainable": true, "sec_per_step": 0.045, "alpha": 0.03, "t": 179.2} |
| {"event": "train", "step": 4300, "phase": 3, "objective": "global", "lm_loss": 1.12033, "bpb": 1.61629, "chunk_loss": 0.030478, "F": 0.20578, "G": 0.21147, "chunks": 210.72, "bytes_per_chunk": 4.86, "grad_norm": 0.9732, "lr": 4.39e-05, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 181.4} |
| {"event": "train", "step": 4350, "phase": 3, "objective": "global", "lm_loss": 1.14029, "bpb": 1.64509, "chunk_loss": 0.030692, "F": 0.21616, "G": 0.22163, "chunks": 221.34, "bytes_per_chunk": 4.626, "grad_norm": 0.9484, "lr": 4.2e-05, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 183.5} |
| {"event": "train", "step": 4400, "phase": 3, "objective": "global", "lm_loss": 1.13519, "bpb": 1.63773, "chunk_loss": 0.030559, "F": 0.21228, "G": 0.21784, "chunks": 217.38, "bytes_per_chunk": 4.711, "grad_norm": 1.068, "lr": 4.03e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 185.5} |
| {"event": "train", "step": 4450, "phase": 3, "objective": "global", "lm_loss": 1.05593, "bpb": 1.52338, "chunk_loss": 0.030461, "F": 0.2059, "G": 0.21199, "chunks": 210.84, "bytes_per_chunk": 4.857, "grad_norm": 1.0372, "lr": 3.87e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 187.6} |
| {"event": "probe", "step": 4500, "phase": 3, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2668.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.197586, "cps": 27.791667, "mean_p": 0.204248}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.065967, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 189.6} |
| {"event": "eval", "step": 4500, "phase": 3, "bpb": {"en": 1.66983}, "bytes_per_chunk": {"en": 4.745}, "bpb_mean": 1.66983, "t": 189.8} |
| {"event": "train", "step": 4500, "phase": 3, "objective": "global", "lm_loss": 1.12616, "bpb": 1.62471, "chunk_loss": 0.030513, "F": 0.20874, "G": 0.21445, "chunks": 213.75, "bytes_per_chunk": 4.791, "grad_norm": 1.0024, "lr": 3.72e-05, "router_trainable": true, "sec_per_step": 0.046, "alpha": 0.03, "t": 189.9} |
| {"event": "train", "step": 4550, "phase": 3, "objective": "global", "lm_loss": 1.15383, "bpb": 1.66462, "chunk_loss": 0.030495, "F": 0.20779, "G": 0.21295, "chunks": 212.78, "bytes_per_chunk": 4.812, "grad_norm": 1.0579, "lr": 3.58e-05, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 192.0} |
| {"event": "train", "step": 4600, "phase": 3, "objective": "global", "lm_loss": 1.12586, "bpb": 1.62427, "chunk_loss": 0.030521, "F": 0.21118, "G": 0.21647, "chunks": 216.25, "bytes_per_chunk": 4.735, "grad_norm": 1.0163, "lr": 3.46e-05, "router_trainable": true, "sec_per_step": 0.042, "alpha": 0.03, "t": 194.1} |
| {"event": "train", "step": 4650, "phase": 3, "objective": "global", "lm_loss": 1.13155, "bpb": 1.63249, "chunk_loss": 0.030552, "F": 0.21106, "G": 0.21661, "chunks": 216.12, "bytes_per_chunk": 4.738, "grad_norm": 1.005, "lr": 3.35e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 196.1} |
| {"event": "train", "step": 4700, "phase": 3, "objective": "global", "lm_loss": 1.055, "bpb": 1.52204, "chunk_loss": 0.030565, "F": 0.21112, "G": 0.217, "chunks": 216.19, "bytes_per_chunk": 4.737, "grad_norm": 0.9976, "lr": 3.26e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 198.2} |
| {"event": "eval", "step": 4750, "phase": 3, "bpb": {"en": 1.65123}, "bytes_per_chunk": {"en": 4.748}, "bpb_mean": 1.65123, "t": 200.3} |
| {"event": "train", "step": 4750, "phase": 3, "objective": "global", "lm_loss": 1.08804, "bpb": 1.56971, "chunk_loss": 0.030496, "F": 0.21048, "G": 0.21598, "chunks": 215.53, "bytes_per_chunk": 4.751, "grad_norm": 0.9788, "lr": 3.18e-05, "router_trainable": true, "sec_per_step": 0.043, "alpha": 0.03, "t": 200.4} |
| {"event": "train", "step": 4800, "phase": 3, "objective": "global", "lm_loss": 1.17896, "bpb": 1.70087, "chunk_loss": 0.030509, "F": 0.2067, "G": 0.21234, "chunks": 211.66, "bytes_per_chunk": 4.838, "grad_norm": 1.065, "lr": 3.12e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 202.4} |
| {"event": "train", "step": 4850, "phase": 3, "objective": "global", "lm_loss": 1.0219, "bpb": 1.47429, "chunk_loss": 0.030411, "F": 0.20517, "G": 0.21044, "chunks": 210.09, "bytes_per_chunk": 4.874, "grad_norm": 1.0829, "lr": 3.07e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 204.5} |
| {"event": "train", "step": 4900, "phase": 3, "objective": "global", "lm_loss": 1.07913, "bpb": 1.55686, "chunk_loss": 0.030668, "F": 0.21558, "G": 0.22083, "chunks": 220.75, "bytes_per_chunk": 4.639, "grad_norm": 1.0028, "lr": 3.03e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 206.6} |
| {"event": "train", "step": 4950, "phase": 3, "objective": "global", "lm_loss": 1.1142, "bpb": 1.60745, "chunk_loss": 0.030397, "F": 0.20392, "G": 0.20968, "chunks": 208.81, "bytes_per_chunk": 4.904, "grad_norm": 1.0332, "lr": 3.01e-05, "router_trainable": true, "sec_per_step": 0.041, "alpha": 0.03, "t": 208.6} |
| {"event": "probe", "step": 5000, "phase": 3, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2668.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.197586, "cps": 27.791667, "mean_p": 0.203833}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.065967, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 210.7} |
| {"event": "eval", "step": 5000, "phase": 3, "bpb": {"en": 1.62756}, "bytes_per_chunk": {"en": 4.745}, "bpb_mean": 1.62756, "t": 210.9} |
| {"event": "probe", "step": 5000, "phase": 3, "objective": "global", "router_trainable": true, "per_lang": {"en": {"chunks": 2668.0, "chars": 13503.0, "bytes": 13516.0, "cpc": 0.197586, "cps": 27.791667, "mean_p": 0.203833}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.065967, "worst_lang": "en", "cpc_parity_vs_en": {"en": 1.0}, "t": 210.9} |
| {"event": "final", "run": "baseline_en", "condition": "A", "step": 5000, "params": 22516992, "per_lang": {"en": {"cpc": 0.19758572169147598, "cps": 27.791666666666668, "bpb": 1.623357290677685, "bytes_per_chunk": 4.761882271732318}}, "gini_cpc": 0.0, "gini_cps": 0.0, "bpic": 5.065967016491754, "transitions": [{"step": 0, "phase": 1, "objective": "global", "router_trainable": true, "encoder_trainable": true, "frozen_router_params": 0}, {"step": 2000, "phase": 2, "objective": "global", "router_trainable": true, "encoder_trainable": true, "frozen_router_params": 0}, {"step": 3500, "phase": 3, "objective": "global", "router_trainable": true, "encoder_trainable": true, "frozen_router_params": 0}], "t": 211.2} |
|
|
