{ "C1_BytePairTokenizer_gelu": { "char_acc": 0.9989275932312012, "seq_acc": 0.96705561876297, "bit_acc": 0.9996304512023926, "levenshtein": 0.06382978723404255, "n": 4371, "bleu": 0.9923203065635846, "rouge1": 0.9965856757185961, "rouge2": 0.9937398108901206, "rougeL": 0.9965856757185961, "config": "C1", "activation": "gelu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11316480, "fwd_gflops_per_example": 1.486553088, "decode_sec": 49.33346366882324, "decode_ms_per_chunk": 11.286539388886581, "sec_per_epoch": 56.72380753755569, "peak_gb": 0.94523648, "best_val_acc": 0.9995213052858684, "epoch1_val_acc": 0.9326152912621359, "epochs_run": 20, "best_epoch": 18, "epochs_after_best": 2, "best_val_loss": 0.001683080605730543, "final_val_loss": 0.0021716902461234467, "overfit_ratio": 1.2903067379716031, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1134.476150751114 }, "C1_BytePairTokenizer_relu": { "char_acc": 0.999338686466217, "seq_acc": 0.9823839068412781, "bit_acc": 0.9997788071632385, "levenshtein": 0.038435140700068635, "n": 4371, "bleu": 0.9960936054779879, "rouge1": 0.998094368461798, "rouge2": 0.9967612920053908, "rougeL": 0.998094368461798, "config": "C1", "activation": "relu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11316480, "fwd_gflops_per_example": 1.486553088, "decode_sec": 48.602561235427856, "decode_ms_per_chunk": 11.119323092067686, "sec_per_epoch": 56.108712065219876, "peak_gb": 0.941722112, "best_val_acc": 0.9996797464940669, "epoch1_val_acc": 0.9121190668824164, "epochs_run": 20, "best_epoch": 20, "epochs_after_best": 0, "best_val_loss": 0.0011786286916057333, "final_val_loss": 0.0011786286916057333, "overfit_ratio": 1.0, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1122.1742413043976 }, "C2_BytePairTokenizer_gelu": { "char_acc": 0.9938122034072876, "seq_acc": 0.9716312289237976, "bit_acc": 0.9978091716766357, "levenshtein": 0.06932052161976665, "n": 4371, "bleu": 0.9925888776779175, "rouge1": 0.9966057959507741, "rouge2": 0.9940011736834098, "rougeL": 0.9966057959507741, "config": "C2", "activation": "gelu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11316480, "fwd_gflops_per_example": 1.486553088, "decode_sec": 52.34577417373657, "decode_ms_per_chunk": 11.975697591795143, "sec_per_epoch": 58.69017901420593, "peak_gb": 0.944777728, "best_val_acc": 0.9995887270765912, "epoch1_val_acc": 0.6656418554476807, "epochs_run": 20, "best_epoch": 15, "epochs_after_best": 5, "best_val_loss": 0.001652343539179701, "final_val_loss": 0.001676761161761023, "overfit_ratio": 1.0147775701616166, "epochs_to_0.95": 3, "epochs_to_0.99": 4, "epochs_to_0.995": 4, "total_train_sec": 1173.8035802841187 }, "C2_BytePairTokenizer_relu": { "char_acc": 0.9965039491653442, "seq_acc": 0.9819263219833374, "bit_acc": 0.9987788200378418, "levenshtein": 0.050331731869137496, "n": 4371, "bleu": 0.9952445030212402, "rouge1": 0.9977870617538775, "rouge2": 0.996207717132643, "rougeL": 0.9977870617538775, "config": "C2", "activation": "relu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11316480, "fwd_gflops_per_example": 1.486553088, "decode_sec": 50.910680532455444, "decode_ms_per_chunk": 11.647376008340299, "sec_per_epoch": 57.97124732732773, "peak_gb": 0.94126336, "best_val_acc": 0.9996797464940669, "epoch1_val_acc": 0.6392563376483279, "epochs_run": 20, "best_epoch": 19, "epochs_after_best": 1, "best_val_loss": 0.001211003460313526, "final_val_loss": 0.002836603208062236, "overfit_ratio": 2.3423576406031454, "epochs_to_0.95": 3, "epochs_to_0.99": 4, "epochs_to_0.995": 4, "total_train_sec": 1159.4249465465546 }, "C3_BytePairTokenizer_gelu": { "char_acc": 0.998898983001709, "seq_acc": 0.9720887541770935, "bit_acc": 0.9996232986450195, "levenshtein": 0.06474490963166324, "n": 4371, "bleu": 0.9931190188358343, "rouge1": 0.996952883842344, "rouge2": 0.9943375104609332, "rougeL": 0.996952883842344, "config": "C3", "activation": "gelu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 9547008, "fwd_gflops_per_example": 1.260060672, "decode_sec": 43.54745674133301, "decode_ms_per_chunk": 9.962813255852895, "sec_per_epoch": 52.36152193546295, "peak_gb": 0.916924928, "best_val_acc": 0.9994741100323624, "epoch1_val_acc": 0.9120213052858684, "epochs_run": 20, "best_epoch": 20, "epochs_after_best": 0, "best_val_loss": 0.0019997842202303146, "final_val_loss": 0.0019997842202303146, "overfit_ratio": 1.0, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1047.230438709259 }, "C3_BytePairTokenizer_relu": { "char_acc": 0.999102771282196, "seq_acc": 0.9755204916000366, "bit_acc": 0.9996760487556458, "levenshtein": 0.05513612445664608, "n": 4371, "bleu": 0.994462788105011, "rouge1": 0.9972805589741554, "rouge2": 0.9953050622731323, "rougeL": 0.9972805589741554, "config": "C3", "activation": "relu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 9547008, "fwd_gflops_per_example": 1.260060672, "decode_sec": 44.08893609046936, "decode_ms_per_chunk": 10.086693225913832, "sec_per_epoch": 51.522846376895906, "peak_gb": 0.914458112, "best_val_acc": 0.9996561488673139, "epoch1_val_acc": 0.8754045307443366, "epochs_run": 20, "best_epoch": 18, "epochs_after_best": 2, "best_val_loss": 0.0012703081589652287, "final_val_loss": 0.0014953252387598776, "overfit_ratio": 1.1771358218921808, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1030.456927537918 }, "C4_BytePairTokenizer_gelu": { "char_acc": 0.9991778135299683, "seq_acc": 0.9746053814888, "bit_acc": 0.9997225403785706, "levenshtein": 0.05124685426675818, "n": 4371, "bleu": 0.9939740300225139, "rouge1": 0.9973698575753064, "rouge2": 0.9951202547466119, "rougeL": 0.9973698575753064, "config": "C4", "activation": "gelu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11308288, "fwd_gflops_per_example": 1.486553088, "decode_sec": 47.1136474609375, "decode_ms_per_chunk": 10.778688506277168, "sec_per_epoch": 55.08251992464066, "peak_gb": 0.877996544, "best_val_acc": 0.9995651294498382, "epoch1_val_acc": 0.9329557713052858, "epochs_run": 20, "best_epoch": 20, "epochs_after_best": 0, "best_val_loss": 0.0015080620422415554, "final_val_loss": 0.0015080620422415554, "overfit_ratio": 1.0, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1101.650398492813 }, "C4_BytePairTokenizer_relu": { "char_acc": 0.9991813898086548, "seq_acc": 0.9782658219337463, "bit_acc": 0.9997184872627258, "levenshtein": 0.04415465568519789, "n": 4371, "bleu": 0.9950286746025085, "rouge1": 0.9977768073365358, "rouge2": 0.9958484578433716, "rougeL": 0.9977768073365358, "config": "C4", "activation": "relu", "tokenizer": "BytePairTokenizer", "split": "test", "params": 11308288, "fwd_gflops_per_example": 1.486553088, "decode_sec": 46.72522521018982, "decode_ms_per_chunk": 10.689825030928809, "sec_per_epoch": 54.309548473358156, "peak_gb": 0.8744576, "best_val_acc": 0.9996999730312837, "epoch1_val_acc": 0.9119808522114348, "epochs_run": 20, "best_epoch": 16, "epochs_after_best": 4, "best_val_loss": 0.0010785145554407735, "final_val_loss": 0.0014718058019747883, "overfit_ratio": 1.3646601193743577, "epochs_to_0.95": 2, "epochs_to_0.99": 3, "epochs_to_0.995": 4, "total_train_sec": 1086.190969467163 }, "C5_ByteTokenizer_gelu": { "char_acc": 0.9999035000801086, "seq_acc": 0.9949668049812317, "bit_acc": 0.9999615550041199, "levenshtein": 0.005948295584534431, "n": 4371, "config": "C5", "activation": "gelu", "tokenizer": "ByteTokenizer", "split": "test", "params": 14532608, "fwd_gflops_per_example": 1.343422464, "decode_sec": 51.82190179824829, "decode_ms_per_chunk": 11.855845755719123, "sec_per_epoch": 50.13067399263382, "peak_gb": 0.933005824, "best_val_acc": 0.9999426914778856, "epoch1_val_acc": 0.9944848975188781, "epochs_run": 20, "best_epoch": 17, "epochs_after_best": 3, "best_val_loss": 0.00020337117867793073, "final_val_loss": 0.00023250355829540395, "overfit_ratio": 1.143247336259032, "epochs_to_0.95": 1, "epochs_to_0.99": 1, "epochs_to_0.995": 2, "total_train_sec": 1002.6134798526764 }, "C5_ByteTokenizer_relu": { "char_acc": 0.9999392032623291, "seq_acc": 0.9970258474349976, "bit_acc": 0.9999740719795227, "levenshtein": 0.0036604895904827272, "n": 4371, "config": "C5", "activation": "relu", "tokenizer": "ByteTokenizer", "split": "test", "params": 14532608, "fwd_gflops_per_example": 1.343422464, "decode_sec": 51.38083219528198, "decode_ms_per_chunk": 11.75493758757309, "sec_per_epoch": 49.253746342659, "peak_gb": 0.933005824, "best_val_acc": 0.9999629180151025, "epoch1_val_acc": 0.9893844390507012, "epochs_run": 20, "best_epoch": 19, "epochs_after_best": 1, "best_val_loss": 0.00018105691063978913, "final_val_loss": 0.000528399492901556, "overfit_ratio": 2.918416596386102, "epochs_to_0.95": 1, "epochs_to_0.99": 2, "epochs_to_0.995": 2, "total_train_sec": 985.0749268531799 } }