Add files using upload-large-folder tool

training/q3_1p7b_sft_shuffled/metrics/train_metrics.jsonl

@@ -0,0 +1,102 @@
+{"global_step": 1, "epoch": 0, "batch_id": 63, "num_supervised_tokens": 35595, "learning_rate": 1e-05, "batch_loss": 0.6115079851172449, "token_entropy": 0.5522983591749789, "gradient_norm_pre_clip": 11.750782043291009, "gradient_norm_post_clip": 11.750782043291009, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 38.62890625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 38.62890625, "max_sequence_length": 774.0}
+{"global_step": 2, "epoch": 0, "batch_id": 127, "num_supervised_tokens": 38458, "learning_rate": 1e-05, "batch_loss": 0.5554303640308981, "token_entropy": 0.6265359287657736, "gradient_norm_pre_clip": 5.908129601727622, "gradient_norm_post_clip": 5.908129601727622, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 39.5078125, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 39.5078125, "max_sequence_length": 1320.0}
+{"global_step": 3, "epoch": 0, "batch_id": 191, "num_supervised_tokens": 38079, "learning_rate": 1e-05, "batch_loss": 0.5245464682925752, "token_entropy": 0.6038362795122187, "gradient_norm_pre_clip": 4.204065467710992, "gradient_norm_post_clip": 4.204065467710992, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 39.822265625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 39.822265625, "max_sequence_length": 1492.0}
+{"global_step": 4, "epoch": 0, "batch_id": 255, "num_supervised_tokens": 38749, "learning_rate": 1e-05, "batch_loss": 0.5122589599112823, "token_entropy": 0.554884494623094, "gradient_norm_pre_clip": 2.8432035146354644, "gradient_norm_post_clip": 2.8432035146354644, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 39.822265625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 39.822265625, "max_sequence_length": 1103.0}
+{"global_step": 5, "epoch": 0, "batch_id": 319, "num_supervised_tokens": 38197, "learning_rate": 1e-05, "batch_loss": 0.43897279049475274, "token_entropy": 0.3563731217236407, "gradient_norm_pre_clip": 2.9185904018118345, "gradient_norm_post_clip": 2.9185904018118345, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 39.822265625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 39.822265625, "max_sequence_length": 897.0}
+{"global_step": 6, "epoch": 0, "batch_id": 383, "num_supervised_tokens": 35968, "learning_rate": 1e-05, "batch_loss": 0.3761693582931381, "token_entropy": 0.31643749240560465, "gradient_norm_pre_clip": 2.273513271894914, "gradient_norm_post_clip": 2.273513271894914, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 39.822265625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 39.822265625, "max_sequence_length": 764.0}
+{"global_step": 7, "epoch": 0, "batch_id": 447, "num_supervised_tokens": 40919, "learning_rate": 1e-05, "batch_loss": 0.8958215145995757, "token_entropy": 0.8189288171995519, "gradient_norm_pre_clip": 2.164089535473581, "gradient_norm_post_clip": 2.164089535473581, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.74642324447632, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3279.0}
+{"global_step": 8, "epoch": 0, "batch_id": 511, "num_supervised_tokens": 39919, "learning_rate": 1e-05, "batch_loss": 0.9529291996077472, "token_entropy": 0.9220841613542479, "gradient_norm_pre_clip": 1.7415795805261365, "gradient_norm_post_clip": 1.7415795805261365, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.602320194244385, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3192.0}
+{"global_step": 9, "epoch": 0, "batch_id": 575, "num_supervised_tokens": 37415, "learning_rate": 1e-05, "batch_loss": 0.44158571376862904, "token_entropy": 0.3907286083859173, "gradient_norm_pre_clip": 1.956837277451057, "gradient_norm_post_clip": 1.956837277451057, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 736.0}
+{"global_step": 10, "epoch": 0, "batch_id": 639, "num_supervised_tokens": 38899, "learning_rate": 1e-05, "batch_loss": 0.43926952919210516, "token_entropy": 0.4070600166508108, "gradient_norm_pre_clip": 1.689292725881464, "gradient_norm_post_clip": 1.689292725881464, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.7323203086853, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 2063.0, "parameter_l2_norm": 2564.8865227618226, "parameter_l2_from_init": 1.7989767383518753, "relative_parameter_l2_from_init": 0.0007013850924057833, "update_norm": 1.7989767383518753, "relative_update_norm": 0.0007013864833344636}
+{"global_step": 11, "epoch": 0, "batch_id": 703, "num_supervised_tokens": 38072, "learning_rate": 1e-05, "batch_loss": 0.42055766502609365, "token_entropy": 0.4051801346009157, "gradient_norm_pre_clip": 1.4741618271214918, "gradient_norm_post_clip": 1.4741618271214918, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1055.0}
+{"global_step": 12, "epoch": 0, "batch_id": 767, "num_supervised_tokens": 38064, "learning_rate": 1e-05, "batch_loss": 0.4427435262281974, "token_entropy": 0.4401419875801186, "gradient_norm_pre_clip": 1.4852969922885346, "gradient_norm_post_clip": 1.4852969922885346, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1083.0}
+{"global_step": 13, "epoch": 0, "batch_id": 831, "num_supervised_tokens": 39473, "learning_rate": 1e-05, "batch_loss": 0.5211687956717022, "token_entropy": 0.5191764226330757, "gradient_norm_pre_clip": 1.3259705061231144, "gradient_norm_post_clip": 1.3259705061231144, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1593.0}
+{"global_step": 14, "epoch": 0, "batch_id": 895, "num_supervised_tokens": 38530, "learning_rate": 1e-05, "batch_loss": 0.432923480924127, "token_entropy": 0.43425283578040536, "gradient_norm_pre_clip": 1.3022622868216214, "gradient_norm_post_clip": 1.3022622868216214, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 818.0}
+{"global_step": 15, "epoch": 0, "batch_id": 959, "num_supervised_tokens": 43468, "learning_rate": 1e-05, "batch_loss": 1.0000247014947856, "token_entropy": 0.9678406387394745, "gradient_norm_pre_clip": 1.3556812490154921, "gradient_norm_post_clip": 1.3556812490154921, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.64373064041138, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3217.0}
+{"global_step": 16, "epoch": 0, "batch_id": 1023, "num_supervised_tokens": 38795, "learning_rate": 1e-05, "batch_loss": 0.504660376119566, "token_entropy": 0.5002966089613706, "gradient_norm_pre_clip": 1.2535649806630977, "gradient_norm_post_clip": 1.2535649806630977, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1056.0}
+{"global_step": 17, "epoch": 0, "batch_id": 1087, "num_supervised_tokens": 39931, "learning_rate": 1e-05, "batch_loss": 0.5422547872776429, "token_entropy": 0.5280099497966084, "gradient_norm_pre_clip": 1.3104690814561608, "gradient_norm_post_clip": 1.3104690814561608, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1549.0}
+{"global_step": 18, "epoch": 0, "batch_id": 1151, "num_supervised_tokens": 42151, "learning_rate": 1e-05, "batch_loss": 0.4949117014593129, "token_entropy": 0.4807967120157448, "gradient_norm_pre_clip": 1.2097122730612455, "gradient_norm_post_clip": 1.2097122730612455, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 39.418041706085205, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 2477.0}
+{"global_step": 19, "epoch": 0, "batch_id": 1215, "num_supervised_tokens": 36531, "learning_rate": 1e-05, "batch_loss": 0.5733890745634115, "token_entropy": 0.5659576365735678, "gradient_norm_pre_clip": 1.282091509620998, "gradient_norm_post_clip": 1.282091509620998, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1428.0}
+{"global_step": 20, "epoch": 0, "batch_id": 1279, "num_supervised_tokens": 40288, "learning_rate": 1e-05, "batch_loss": 0.41561071466058314, "token_entropy": 0.4048386906895778, "gradient_norm_pre_clip": 1.2773711593421102, "gradient_norm_post_clip": 1.2773711593421102, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1035.0, "parameter_l2_norm": 2564.88332677438, "parameter_l2_from_init": 2.5555543228394066, "relative_parameter_l2_from_init": 0.0009963595785651132, "update_norm": 1.0305875518913967, "relative_update_norm": 0.000401806796096052}
+{"global_step": 21, "epoch": 0, "batch_id": 1343, "num_supervised_tokens": 38170, "learning_rate": 1e-05, "batch_loss": 0.41761356979083464, "token_entropy": 0.4030980381754515, "gradient_norm_pre_clip": 1.2762568380942494, "gradient_norm_post_clip": 1.2762568380942494, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1183.0}
+{"global_step": 22, "epoch": 0, "batch_id": 1407, "num_supervised_tokens": 36739, "learning_rate": 1e-05, "batch_loss": 0.44722668783394304, "token_entropy": 0.4446814653678044, "gradient_norm_pre_clip": 1.301584362139485, "gradient_norm_post_clip": 1.301584362139485, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1183.0}
+{"global_step": 23, "epoch": 0, "batch_id": 1471, "num_supervised_tokens": 38244, "learning_rate": 1e-05, "batch_loss": 0.438172042152232, "token_entropy": 0.441769512611607, "gradient_norm_pre_clip": 1.2115678564472445, "gradient_norm_post_clip": 1.2115678564472445, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 974.0}
+{"global_step": 24, "epoch": 0, "batch_id": 1535, "num_supervised_tokens": 39408, "learning_rate": 1e-05, "batch_loss": 0.39379193175970145, "token_entropy": 0.40187964580236196, "gradient_norm_pre_clip": 1.1682806666444265, "gradient_norm_post_clip": 1.1682806666444265, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 924.0}
+{"global_step": 25, "epoch": 0, "batch_id": 1599, "num_supervised_tokens": 36372, "learning_rate": 1e-05, "batch_loss": 0.42351364390079593, "token_entropy": 0.4311174127668303, "gradient_norm_pre_clip": 1.2175395490922194, "gradient_norm_post_clip": 1.2175395490922194, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1423.0}
+{"global_step": 26, "epoch": 0, "batch_id": 1663, "num_supervised_tokens": 37702, "learning_rate": 1e-05, "batch_loss": 0.4125767993771319, "token_entropy": 0.4270622494884602, "gradient_norm_pre_clip": 1.2772858153846423, "gradient_norm_post_clip": 1.2772858153846423, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 878.0}
+{"global_step": 27, "epoch": 0, "batch_id": 1727, "num_supervised_tokens": 40490, "learning_rate": 1e-05, "batch_loss": 0.43035515632216326, "token_entropy": 0.43507689734864274, "gradient_norm_pre_clip": 1.1776633546014645, "gradient_norm_post_clip": 1.1776633546014645, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.713294982910156, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3259.0}
+{"global_step": 28, "epoch": 0, "batch_id": 1791, "num_supervised_tokens": 39980, "learning_rate": 1e-05, "batch_loss": 0.7371608465206457, "token_entropy": 0.7403877516398732, "gradient_norm_pre_clip": 1.2472833486777928, "gradient_norm_post_clip": 1.2472833486777928, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.635448932647705, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3212.0}
+{"global_step": 29, "epoch": 0, "batch_id": 1855, "num_supervised_tokens": 42085, "learning_rate": 1e-05, "batch_loss": 0.5536885042491053, "token_entropy": 0.5565725971234161, "gradient_norm_pre_clip": 1.1651220904674928, "gradient_norm_post_clip": 1.1651220904674928, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1826.0}
+{"global_step": 30, "epoch": 0, "batch_id": 1919, "num_supervised_tokens": 39896, "learning_rate": 1e-05, "batch_loss": 0.4733060929591461, "token_entropy": 0.4737771046718615, "gradient_norm_pre_clip": 1.177064205436616, "gradient_norm_post_clip": 1.177064205436616, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1704.0, "parameter_l2_norm": 2564.8801262902507, "parameter_l2_from_init": 3.059634497360145, "relative_parameter_l2_from_init": 0.001192890368679753, "update_norm": 0.8160430241245262, "relative_update_norm": 0.0003181602975359403}
+{"global_step": 31, "epoch": 0, "batch_id": 1983, "num_supervised_tokens": 38810, "learning_rate": 1e-05, "batch_loss": 0.5767777736991052, "token_entropy": 0.5802622256712704, "gradient_norm_pre_clip": 1.204872637477699, "gradient_norm_post_clip": 1.204872637477699, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1779.0}
+{"global_step": 32, "epoch": 0, "batch_id": 2047, "num_supervised_tokens": 38369, "learning_rate": 1e-05, "batch_loss": 0.5623053391071685, "token_entropy": 0.564259632163985, "gradient_norm_pre_clip": 1.237818523830255, "gradient_norm_post_clip": 1.237818523830255, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1625.0}
+{"global_step": 33, "epoch": 0, "batch_id": 2111, "num_supervised_tokens": 37995, "learning_rate": 1e-05, "batch_loss": 0.4162265927477022, "token_entropy": 0.4118520450303391, "gradient_norm_pre_clip": 1.2767576814048438, "gradient_norm_post_clip": 1.2767576814048438, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 767.0}
+{"global_step": 34, "epoch": 0, "batch_id": 2175, "num_supervised_tokens": 40364, "learning_rate": 1e-05, "batch_loss": 0.43085935420318205, "token_entropy": 0.4330923018825502, "gradient_norm_pre_clip": 1.2023724277455845, "gradient_norm_post_clip": 1.2023724277455845, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 854.0}
+{"global_step": 35, "epoch": 0, "batch_id": 2239, "num_supervised_tokens": 37573, "learning_rate": 1e-05, "batch_loss": 0.5045610012406416, "token_entropy": 0.5153670109822523, "gradient_norm_pre_clip": 1.217861587240192, "gradient_norm_post_clip": 1.217861587240192, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1441.0}
+{"global_step": 36, "epoch": 0, "batch_id": 2303, "num_supervised_tokens": 37059, "learning_rate": 1e-05, "batch_loss": 0.32607971006006836, "token_entropy": 0.34078944131171873, "gradient_norm_pre_clip": 1.1991254644064167, "gradient_norm_post_clip": 1.1991254644064167, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 745.0}
+{"global_step": 37, "epoch": 0, "batch_id": 2367, "num_supervised_tokens": 37358, "learning_rate": 1e-05, "batch_loss": 0.342957629232209, "token_entropy": 0.35673879220275206, "gradient_norm_pre_clip": 1.19456616799883, "gradient_norm_post_clip": 1.19456616799883, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 943.0}
+{"global_step": 38, "epoch": 0, "batch_id": 2431, "num_supervised_tokens": 36234, "learning_rate": 1e-05, "batch_loss": 0.37653978863037973, "token_entropy": 0.3994591684736755, "gradient_norm_pre_clip": 1.2332911458696274, "gradient_norm_post_clip": 1.2332911458696274, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 868.0}
+{"global_step": 39, "epoch": 0, "batch_id": 2495, "num_supervised_tokens": 40328, "learning_rate": 1e-05, "batch_loss": 0.38619560618903154, "token_entropy": 0.38873427502103, "gradient_norm_pre_clip": 1.1777427634040272, "gradient_norm_post_clip": 1.1777427634040272, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1058.0}
+{"global_step": 40, "epoch": 0, "batch_id": 2559, "num_supervised_tokens": 37200, "learning_rate": 1e-05, "batch_loss": 0.5031413834068363, "token_entropy": 0.5185064822114924, "gradient_norm_pre_clip": 1.240205655129607, "gradient_norm_post_clip": 1.240205655129607, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1184.0, "parameter_l2_norm": 2564.877541703326, "parameter_l2_from_init": 3.4883586672180336, "relative_parameter_l2_from_init": 0.001360041357951495, "update_norm": 0.766428833091556, "relative_update_norm": 0.0002988169300989603}
+{"global_step": 41, "epoch": 0, "batch_id": 2623, "num_supervised_tokens": 38747, "learning_rate": 1e-05, "batch_loss": 0.3846446139913475, "token_entropy": 0.3932966566345481, "gradient_norm_pre_clip": 1.1857453735701038, "gradient_norm_post_clip": 1.1857453735701038, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1169.0}
+{"global_step": 42, "epoch": 0, "batch_id": 2687, "num_supervised_tokens": 42200, "learning_rate": 1e-05, "batch_loss": 0.5628710920376023, "token_entropy": 0.5627155288135836, "gradient_norm_pre_clip": 1.1745494692233858, "gradient_norm_post_clip": 1.1745494692233858, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 39.56214427947998, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 2564.0}
+{"global_step": 43, "epoch": 0, "batch_id": 2751, "num_supervised_tokens": 35239, "learning_rate": 1e-05, "batch_loss": 0.31295179143180546, "token_entropy": 0.3314907350883591, "gradient_norm_pre_clip": 1.2502804246850447, "gradient_norm_post_clip": 1.2502804246850447, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 789.0}
+{"global_step": 44, "epoch": 0, "batch_id": 2815, "num_supervised_tokens": 42046, "learning_rate": 1e-05, "batch_loss": 0.5928385910338718, "token_entropy": 0.6035941001073009, "gradient_norm_pre_clip": 1.1374794457052393, "gradient_norm_post_clip": 1.1374794457052393, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1218.0}
+{"global_step": 45, "epoch": 0, "batch_id": 2879, "num_supervised_tokens": 45848, "learning_rate": 1e-05, "batch_loss": 0.44319771831989085, "token_entropy": 0.44399270810000957, "gradient_norm_pre_clip": 1.200124483205029, "gradient_norm_post_clip": 1.200124483205029, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1134.0}
+{"global_step": 46, "epoch": 0, "batch_id": 2943, "num_supervised_tokens": 38744, "learning_rate": 1e-05, "batch_loss": 0.8785585609946193, "token_entropy": 0.9030287023814483, "gradient_norm_pre_clip": 1.2492340857454842, "gradient_norm_post_clip": 1.2492340857454842, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.39030933380127, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3064.0}
+{"global_step": 47, "epoch": 0, "batch_id": 3007, "num_supervised_tokens": 41873, "learning_rate": 1e-05, "batch_loss": 0.494750945416429, "token_entropy": 0.5037620540698429, "gradient_norm_pre_clip": 1.1820907330121793, "gradient_norm_post_clip": 1.1820907330121793, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 39.464420318603516, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 2505.0}
+{"global_step": 48, "epoch": 0, "batch_id": 3071, "num_supervised_tokens": 40374, "learning_rate": 1e-05, "batch_loss": 0.41811611204755367, "token_entropy": 0.43513855480856045, "gradient_norm_pre_clip": 1.191288129221852, "gradient_norm_post_clip": 1.191288129221852, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1326.0}
+{"global_step": 49, "epoch": 0, "batch_id": 3135, "num_supervised_tokens": 39395, "learning_rate": 1e-05, "batch_loss": 0.722566135214536, "token_entropy": 0.7432812312968962, "gradient_norm_pre_clip": 1.258989549167586, "gradient_norm_post_clip": 1.258989549167586, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 39.752620697021484, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 2679.0}
+{"global_step": 50, "epoch": 0, "batch_id": 3199, "num_supervised_tokens": 38033, "learning_rate": 1e-05, "batch_loss": 0.41951992628404444, "token_entropy": 0.43724067072048023, "gradient_norm_pre_clip": 1.254145857281094, "gradient_norm_post_clip": 1.254145857281094, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 38.52034950256348, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 1115.0, "parameter_l2_norm": 2564.874815726263, "parameter_l2_from_init": 3.8863062059669624, "relative_parameter_l2_from_init": 0.0015151931535738114, "update_norm": 0.7439927575923858, "relative_update_norm": 0.0002900698127763085}
+{"global_step": 51, "epoch": 0, "batch_id": 3263, "num_supervised_tokens": 42229, "learning_rate": 1e-05, "batch_loss": 0.9724291207361417, "token_entropy": 0.9776299726915336, "gradient_norm_pre_clip": 1.1828823007393239, "gradient_norm_post_clip": 1.1828823007393239, "cuda_current_allocated_gib": 25.701066970825195, "cuda_current_reserved_gib": 42.791015625, "cuda_peak_allocated_gib": 40.579132080078125, "cuda_peak_reserved_gib": 42.791015625, "max_sequence_length": 3178.0}
+{"global_step": 52, "epoch": 0, "batch_id": 3327, "num_supervised_tokens": 35733, "learning_rate": 1e-05, "batch_loss": 0.4264874977063285, "token_entropy": 0.452328435046825, "gradient_norm_pre_clip": 1.2448511991520748, "gradient_norm_post_clip": 1.2448511991520748, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 38.630859375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 38.630859375, "max_sequence_length": 926.0}
+{"global_step": 53, "epoch": 0, "batch_id": 3391, "num_supervised_tokens": 39918, "learning_rate": 1e-05, "batch_loss": 0.4074231522593889, "token_entropy": 0.4310880101057372, "gradient_norm_pre_clip": 1.1481139746964604, "gradient_norm_post_clip": 1.1481139746964604, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 38.650390625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 38.650390625, "max_sequence_length": 957.0}
+{"global_step": 54, "epoch": 0, "batch_id": 3455, "num_supervised_tokens": 38852, "learning_rate": 1e-05, "batch_loss": 0.2907225647773914, "token_entropy": 0.314105983326858, "gradient_norm_pre_clip": 1.14349062741223, "gradient_norm_post_clip": 1.14349062741223, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 40.64553451538086, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 3218.0}
+{"global_step": 55, "epoch": 0, "batch_id": 3519, "num_supervised_tokens": 39919, "learning_rate": 1e-05, "batch_loss": 0.5470495853846422, "token_entropy": 0.5722634434631327, "gradient_norm_pre_clip": 1.190520586462106, "gradient_norm_post_clip": 1.190520586462106, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 40.62565898895264, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 3206.0}
+{"global_step": 56, "epoch": 0, "batch_id": 3583, "num_supervised_tokens": 38729, "learning_rate": 1e-05, "batch_loss": 0.36827851605970663, "token_entropy": 0.39008631830959845, "gradient_norm_pre_clip": 1.1916799953151915, "gradient_norm_post_clip": 1.1916799953151915, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1260.0}
+{"global_step": 57, "epoch": 0, "batch_id": 3647, "num_supervised_tokens": 36849, "learning_rate": 1e-05, "batch_loss": 0.3551905220680028, "token_entropy": 0.38072224331567234, "gradient_norm_pre_clip": 1.249509083923842, "gradient_norm_post_clip": 1.249509083923842, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1327.0}
+{"global_step": 58, "epoch": 0, "batch_id": 3711, "num_supervised_tokens": 40791, "learning_rate": 1e-05, "batch_loss": 0.41962754659052215, "token_entropy": 0.4370021586273082, "gradient_norm_pre_clip": 1.1754314155675447, "gradient_norm_post_clip": 1.1754314155675447, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1165.0}
+{"global_step": 59, "epoch": 0, "batch_id": 3775, "num_supervised_tokens": 35230, "learning_rate": 1e-05, "batch_loss": 0.3431599554949708, "token_entropy": 0.371388406236504, "gradient_norm_pre_clip": 1.2830303272613002, "gradient_norm_post_clip": 1.2830303272613002, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 822.0}
+{"global_step": 60, "epoch": 0, "batch_id": 3839, "num_supervised_tokens": 38789, "learning_rate": 1e-05, "batch_loss": 0.3329281333882721, "token_entropy": 0.3585336901490098, "gradient_norm_pre_clip": 1.1874897641946844, "gradient_norm_post_clip": 1.1874897641946844, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 949.0, "parameter_l2_norm": 2564.8725468259304, "parameter_l2_from_init": 4.310021390205154, "relative_parameter_l2_from_init": 0.001680391239416158, "update_norm": 0.7696166181237579, "relative_update_norm": 0.0003000603749594382}
+{"global_step": 61, "epoch": 0, "batch_id": 3903, "num_supervised_tokens": 39372, "learning_rate": 1e-05, "batch_loss": 0.5233446117745052, "token_entropy": 0.5440462944312127, "gradient_norm_pre_clip": 1.2230855109544123, "gradient_norm_post_clip": 1.2230855109544123, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 39.91840314865112, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 2779.0}
+{"global_step": 62, "epoch": 0, "batch_id": 3967, "num_supervised_tokens": 42472, "learning_rate": 1e-05, "batch_loss": 0.6765156513468388, "token_entropy": 0.6836658881913698, "gradient_norm_pre_clip": 1.1653133299966727, "gradient_norm_post_clip": 1.1653133299966727, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 40.605780601501465, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 3194.0}
+{"global_step": 63, "epoch": 0, "batch_id": 4031, "num_supervised_tokens": 39498, "learning_rate": 1e-05, "batch_loss": 0.4540258361734674, "token_entropy": 0.4626636327179216, "gradient_norm_pre_clip": 1.2167812208368673, "gradient_norm_post_clip": 1.2167812208368673, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.66124439239502, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 2020.0}
+{"global_step": 64, "epoch": 0, "batch_id": 4095, "num_supervised_tokens": 36428, "learning_rate": 1e-05, "batch_loss": 0.3676329950099062, "token_entropy": 0.3834996238110342, "gradient_norm_pre_clip": 1.2629767575795694, "gradient_norm_post_clip": 1.2629767575795694, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1420.0}
+{"global_step": 65, "epoch": 0, "batch_id": 4159, "num_supervised_tokens": 40012, "learning_rate": 1e-05, "batch_loss": 0.7496747684577599, "token_entropy": 0.7573721720601968, "gradient_norm_pre_clip": 1.1699678556395612, "gradient_norm_post_clip": 1.1699678556395612, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 40.62565898895264, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 3206.0}
+{"global_step": 66, "epoch": 0, "batch_id": 4223, "num_supervised_tokens": 39825, "learning_rate": 1e-05, "batch_loss": 0.4570504533345136, "token_entropy": 0.48038996843145243, "gradient_norm_pre_clip": 1.180489722905441, "gradient_norm_post_clip": 1.180489722905441, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 940.0}
+{"global_step": 67, "epoch": 0, "batch_id": 4287, "num_supervised_tokens": 39291, "learning_rate": 1e-05, "batch_loss": 0.317471748684846, "token_entropy": 0.35598751960454006, "gradient_norm_pre_clip": 1.190442990872425, "gradient_norm_post_clip": 1.190442990872425, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 804.0}
+{"global_step": 68, "epoch": 0, "batch_id": 4351, "num_supervised_tokens": 35329, "learning_rate": 1e-05, "batch_loss": 0.3798398624212178, "token_entropy": 0.4144585976366743, "gradient_norm_pre_clip": 1.2516332673852122, "gradient_norm_post_clip": 1.2516332673852122, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 852.0}
+{"global_step": 69, "epoch": 0, "batch_id": 4415, "num_supervised_tokens": 39135, "learning_rate": 1e-05, "batch_loss": 0.40428939048851475, "token_entropy": 0.4325886920720845, "gradient_norm_pre_clip": 1.1986547030839962, "gradient_norm_post_clip": 1.1986547030839962, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1372.0}
+{"global_step": 70, "epoch": 0, "batch_id": 4479, "num_supervised_tokens": 35374, "learning_rate": 1e-05, "batch_loss": 0.40254540275008305, "token_entropy": 0.4365511286175796, "gradient_norm_pre_clip": 1.2930102570737187, "gradient_norm_post_clip": 1.2930102570737187, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 1132.0, "parameter_l2_norm": 2564.8702289100593, "parameter_l2_from_init": 4.710242132453206, "relative_parameter_l2_from_init": 0.0018364293116713507, "update_norm": 0.734483461951192, "relative_update_norm": 0.0002863628162050563}
+{"global_step": 71, "epoch": 0, "batch_id": 4543, "num_supervised_tokens": 38830, "learning_rate": 1e-05, "batch_loss": 0.35527964321377886, "token_entropy": 0.3795889584022854, "gradient_norm_pre_clip": 1.2086695154850853, "gradient_norm_post_clip": 1.2086695154850853, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.615234375, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.615234375, "max_sequence_length": 852.0}
+{"global_step": 72, "epoch": 0, "batch_id": 4607, "num_supervised_tokens": 39732, "learning_rate": 1e-05, "batch_loss": 0.7415623589819935, "token_entropy": 0.7645009383686742, "gradient_norm_pre_clip": 1.216221916732849, "gradient_norm_post_clip": 1.216221916732849, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.70184850692749, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3252.0}
+{"global_step": 73, "epoch": 0, "batch_id": 4671, "num_supervised_tokens": 40644, "learning_rate": 1e-05, "batch_loss": 0.6571113305305132, "token_entropy": 0.6808921543940141, "gradient_norm_pre_clip": 1.2247850107756983, "gradient_norm_post_clip": 1.2247850107756983, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.45339918136597, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3102.0}
+{"global_step": 74, "epoch": 0, "batch_id": 4735, "num_supervised_tokens": 37231, "learning_rate": 1e-05, "batch_loss": 0.335500991186499, "token_entropy": 0.36843365204618905, "gradient_norm_pre_clip": 1.2127346162004924, "gradient_norm_post_clip": 1.2127346162004924, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 889.0}
+{"global_step": 75, "epoch": 0, "batch_id": 4799, "num_supervised_tokens": 35415, "learning_rate": 1e-05, "batch_loss": 0.3435840227049605, "token_entropy": 0.3751828176834212, "gradient_norm_pre_clip": 1.2496687108046414, "gradient_norm_post_clip": 1.2496687108046414, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 874.0}
+{"global_step": 76, "epoch": 0, "batch_id": 4863, "num_supervised_tokens": 39288, "learning_rate": 1e-05, "batch_loss": 0.5012429471004802, "token_entropy": 0.5244210489265692, "gradient_norm_pre_clip": 1.2291341424273583, "gradient_norm_post_clip": 1.2291341424273583, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1257.0}
+{"global_step": 77, "epoch": 0, "batch_id": 4927, "num_supervised_tokens": 41469, "learning_rate": 1e-05, "batch_loss": 1.0037667628415585, "token_entropy": 1.0293630408730183, "gradient_norm_pre_clip": 1.1979851940465354, "gradient_norm_post_clip": 1.1979851940465354, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.69356727600098, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3247.0}
+{"global_step": 78, "epoch": 0, "batch_id": 4991, "num_supervised_tokens": 36321, "learning_rate": 1e-05, "batch_loss": 0.4086013517885927, "token_entropy": 0.45013082877171956, "gradient_norm_pre_clip": 1.275631506463141, "gradient_norm_post_clip": 1.275631506463141, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 956.0}
+{"global_step": 79, "epoch": 0, "batch_id": 5055, "num_supervised_tokens": 42191, "learning_rate": 1e-05, "batch_loss": 0.4090270907513585, "token_entropy": 0.4345601039511343, "gradient_norm_pre_clip": 1.1349553422546, "gradient_norm_post_clip": 1.1349553422546, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 894.0}
+{"global_step": 80, "epoch": 0, "batch_id": 5119, "num_supervised_tokens": 35651, "learning_rate": 1e-05, "batch_loss": 0.32013294119559754, "token_entropy": 0.3624018752883552, "gradient_norm_pre_clip": 1.3024845181742815, "gradient_norm_post_clip": 1.3024845181742815, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 798.0, "parameter_l2_norm": 2564.868663096049, "parameter_l2_from_init": 5.141903628895947, "relative_parameter_l2_from_init": 0.002004725505900012, "update_norm": 0.7689322510930778, "relative_update_norm": 0.00029979400589069574}
+{"global_step": 81, "epoch": 0, "batch_id": 5183, "num_supervised_tokens": 46633, "learning_rate": 1e-05, "batch_loss": 0.7157380362081137, "token_entropy": 0.7316534938177878, "gradient_norm_pre_clip": 1.0697065827007788, "gradient_norm_post_clip": 1.0697065827007788, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.61737775802612, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3201.0}
+{"global_step": 82, "epoch": 0, "batch_id": 5247, "num_supervised_tokens": 37956, "learning_rate": 1e-05, "batch_loss": 0.40888613869844087, "token_entropy": 0.43365279868651846, "gradient_norm_pre_clip": 1.2160421041195841, "gradient_norm_post_clip": 1.2160421041195841, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 960.0}
+{"global_step": 83, "epoch": 0, "batch_id": 5311, "num_supervised_tokens": 45863, "learning_rate": 1e-05, "batch_loss": 0.6041303461282029, "token_entropy": 0.6306276754133862, "gradient_norm_pre_clip": 1.0505752333899532, "gradient_norm_post_clip": 1.0505752333899532, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.650503635406494, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3221.0}
+{"global_step": 84, "epoch": 0, "batch_id": 5375, "num_supervised_tokens": 42951, "learning_rate": 1e-05, "batch_loss": 0.8829925069231787, "token_entropy": 0.9257842825026993, "gradient_norm_pre_clip": 1.1702303064128194, "gradient_norm_post_clip": 1.1702303064128194, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.65547227859497, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3224.0}
+{"global_step": 85, "epoch": 0, "batch_id": 5439, "num_supervised_tokens": 38855, "learning_rate": 1e-05, "batch_loss": 0.39335928056281233, "token_entropy": 0.43011091080369707, "gradient_norm_pre_clip": 1.2005732714114803, "gradient_norm_post_clip": 1.2005732714114803, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1012.0}
+{"global_step": 86, "epoch": 0, "batch_id": 5503, "num_supervised_tokens": 36758, "learning_rate": 1e-05, "batch_loss": 0.2966990740184596, "token_entropy": 0.33127248234662965, "gradient_norm_pre_clip": 1.2430146077163848, "gradient_norm_post_clip": 1.2430146077163848, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 793.0}
+{"global_step": 87, "epoch": 0, "batch_id": 5567, "num_supervised_tokens": 39019, "learning_rate": 1e-05, "batch_loss": 0.3832843240323561, "token_entropy": 0.42371242626042266, "gradient_norm_pre_clip": 1.170002871161156, "gradient_norm_post_clip": 1.170002871161156, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1255.0}
+{"global_step": 88, "epoch": 0, "batch_id": 5631, "num_supervised_tokens": 37292, "learning_rate": 1e-05, "batch_loss": 0.4114828224481291, "token_entropy": 0.4605441379352763, "gradient_norm_pre_clip": 1.2293546084048508, "gradient_norm_post_clip": 1.2293546084048508, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1658.0}
+{"global_step": 89, "epoch": 0, "batch_id": 5695, "num_supervised_tokens": 39271, "learning_rate": 1e-05, "batch_loss": 0.35884744506592064, "token_entropy": 0.38474147164555406, "gradient_norm_pre_clip": 1.1564628917591127, "gradient_norm_post_clip": 1.1564628917591127, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1043.0}
+{"global_step": 90, "epoch": 0, "batch_id": 5759, "num_supervised_tokens": 38784, "learning_rate": 1e-05, "batch_loss": 0.32387491487056513, "token_entropy": 0.353547760042617, "gradient_norm_pre_clip": 1.1926757068738056, "gradient_norm_post_clip": 1.1926757068738056, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 898.0, "parameter_l2_norm": 2564.8671186106585, "parameter_l2_from_init": 5.594595885775977, "relative_parameter_l2_from_init": 0.0021812211734949525, "update_norm": 0.7830019421921601, "relative_update_norm": 0.00030527973028727427}
+{"global_step": 91, "epoch": 0, "batch_id": 5823, "num_supervised_tokens": 37076, "learning_rate": 1e-05, "batch_loss": 0.4044827641509309, "token_entropy": 0.4331861588417075, "gradient_norm_pre_clip": 1.2118217769553639, "gradient_norm_post_clip": 1.2118217769553639, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 772.0}
+{"global_step": 92, "epoch": 0, "batch_id": 5887, "num_supervised_tokens": 39431, "learning_rate": 1e-05, "batch_loss": 0.6820326731283878, "token_entropy": 0.7076275158138884, "gradient_norm_pre_clip": 1.212270120215968, "gradient_norm_post_clip": 1.212270120215968, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.339112281799316, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3033.0}
+{"global_step": 93, "epoch": 0, "batch_id": 5951, "num_supervised_tokens": 40804, "learning_rate": 1e-05, "batch_loss": 0.5383636713394148, "token_entropy": 0.5705826782897435, "gradient_norm_pre_clip": 1.092711592828796, "gradient_norm_post_clip": 1.092711592828796, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1586.0}
+{"global_step": 94, "epoch": 0, "batch_id": 6015, "num_supervised_tokens": 39627, "learning_rate": 1e-05, "batch_loss": 0.42269078910262053, "token_entropy": 0.4583130438984137, "gradient_norm_pre_clip": 1.1107661792418098, "gradient_norm_post_clip": 1.1107661792418098, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1154.0}
+{"global_step": 95, "epoch": 0, "batch_id": 6079, "num_supervised_tokens": 42976, "learning_rate": 1e-05, "batch_loss": 0.8760041393284503, "token_entropy": 0.8917625249541702, "gradient_norm_pre_clip": 1.1580061301599698, "gradient_norm_post_clip": 1.1580061301599698, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.59915542602539, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3190.0}
+{"global_step": 96, "epoch": 0, "batch_id": 6143, "num_supervised_tokens": 45956, "learning_rate": 1e-05, "batch_loss": 0.8498273864428997, "token_entropy": 0.8583646602549267, "gradient_norm_pre_clip": 1.1244867290850622, "gradient_norm_post_clip": 1.1244867290850622, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 40.61240577697754, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 3198.0}
+{"global_step": 97, "epoch": 0, "batch_id": 6207, "num_supervised_tokens": 40479, "learning_rate": 1e-05, "batch_loss": 0.3897647004287955, "token_entropy": 0.41895050864560224, "gradient_norm_pre_clip": 1.1002130224291178, "gradient_norm_post_clip": 1.1002130224291178, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1690.0}
+{"global_step": 98, "epoch": 0, "batch_id": 6271, "num_supervised_tokens": 37374, "learning_rate": 1e-05, "batch_loss": 0.31758297478765496, "token_entropy": 0.346343934252103, "gradient_norm_pre_clip": 1.1768752195606218, "gradient_norm_post_clip": 1.1768752195606218, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 840.0}
+{"global_step": 99, "epoch": 0, "batch_id": 6335, "num_supervised_tokens": 38498, "learning_rate": 1e-05, "batch_loss": 0.43533767390084394, "token_entropy": 0.4635246647593944, "gradient_norm_pre_clip": 1.1262502541630197, "gradient_norm_post_clip": 1.1262502541630197, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1018.0}
+{"global_step": 100, "epoch": 0, "batch_id": 6399, "num_supervised_tokens": 37740, "learning_rate": 1e-05, "batch_loss": 0.3824177592709734, "token_entropy": 0.4106542695434117, "gradient_norm_pre_clip": 1.146226491984302, "gradient_norm_post_clip": 1.146226491984302, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1245.0, "parameter_l2_norm": 2564.8656841744523, "parameter_l2_from_init": 6.027126069733591, "relative_parameter_l2_from_init": 0.0023498560516320276, "update_norm": 0.7641784881838767, "relative_update_norm": 0.0002979409381547561}
+{"global_step": 101, "epoch": 0, "batch_id": 6463, "num_supervised_tokens": 39465, "learning_rate": 1e-05, "batch_loss": 0.34525836199360604, "token_entropy": 0.37905116635221925, "gradient_norm_pre_clip": 0.9950431949460669, "gradient_norm_post_clip": 0.9950431949460669, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 1003.0}
+{"global_step": 102, "epoch": 0, "batch_id": 6481, "num_supervised_tokens": 10574, "learning_rate": 1e-05, "batch_loss": 0.2563313147814457, "token_entropy": 0.3039479456801078, "gradient_norm_pre_clip": 1.5617508157023872, "gradient_norm_post_clip": 1.5617508157023872, "cuda_current_allocated_gib": 25.70121479034424, "cuda_current_reserved_gib": 42.712890625, "cuda_peak_allocated_gib": 38.52049732208252, "cuda_peak_reserved_gib": 42.712890625, "max_sequence_length": 768.0, "parameter_l2_norm": 2564.8653875256623, "parameter_l2_from_init": 6.107013460842562, "relative_parameter_l2_from_init": 0.0023810025495274675, "update_norm": 0.16154742015296286, "relative_update_norm": 6.298475582330986e-05}

training/q3_1p7b_sft_shuffled/offline_wandb_ready.json

@@ -0,0 +1,9 @@
+{
+  "generation": 3,
+  "offline_dir": "wandb/offline-run-20260925_071649-7y8jx2s3",
+  "run_file": "wandb/offline-run-20260925_071649-7y8jx2s3/run-7y8jx2s3.wandb",
+  "run_file_bytes": 998745,
+  "schema_version": 1,
+  "status": "offline_wandb_ready_for_sync",
+  "wandb_run_id": "7y8jx2s3"
+}

training/q3_1p7b_sft_shuffled/provenance.json

@@ -0,0 +1,27 @@
+{
+  "base_model": "/mnt/mystorageoutput/projects/rl-forgetting/base_models/Qwen3-1.7B-Base-ea980cb0",
+  "dataset": "qwen3_1p7b_s500_code_sft_data",
+  "dataset_sha256": "08974359f21d570e60fe801ee98c139f5bd943d0c8411ec03bec40b4ca2a96a6",
+  "durable_jsonl_sha256": "757270437abe9b280a7104cb5b13223f32d782444e42bda77e5907c4c1b5d7e5",
+  "effective_batch_size": 256,
+  "max_seq_length": 4096,
+  "metrics_log_every": 10,
+  "micro_batch_size": 1,
+  "model_name": "qwen3_1p7b_s500",
+  "num_checkpoints": 10,
+  "offline_wandb_sync_required": true,
+  "order": "shuffled",
+  "replay_lambda": 0.0,
+  "replay_strategy": "none",
+  "resume_generation": 3,
+  "resume_helper_sha256": "7aa836fe797652a772a8f0b22724f3ec9a20bd3ed7b834e9d4b3e1756e125d68",
+  "resume_step": 51,
+  "run_sft_sha256": "f4ee1ff82e7da742db079178b8d3695d2142f1db42096b0f1dcb5d837baea263",
+  "run_sweep_sha256": "8713a47d266379e301de2f2b6f20517954cecb10184611cfd9577f8fcbc429dd",
+  "save_optimizer_every_n_checkpoints": 5,
+  "save_optimizer_first_checkpoint": true,
+  "schema_version": 1,
+  "wandb_mode": "offline",
+  "wandb_run_id_file": "/mnt/mystorageoutput/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925/wandb_run_id_generation_3.txt",
+  "world_size": 4
+}

training/q3_1p7b_sft_shuffled/sft_config.json

@@ -0,0 +1,56 @@
+{
+  "generation_logs_dir": null,
+  "dataset": "qwen3_1p7b_s500_code_sft_data",
+  "positives_file": "/mnt/mystorageoutput/projects/rl-forgetting/sft_data/code/qwen3_1p7b_s500_code_sft_data/qwen3_1p7b_s500_code_sft_data.jsonl",
+  "model_path": "/mnt/mystorageoutput/projects/rl-forgetting/base_models/Qwen3-1.7B-Base-ea980cb0",
+  "output_dir": "/mnt/mystorageoutput/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925",
+  "lr": 1e-05,
+  "batch_size": 1,
+  "effective_batch_size": 256,
+  "weight_decay": 0.01,
+  "warmup_ratio": 0.0,
+  "schedule": "constant",
+  "max_seq_length": 4096,
+  "seed": 42,
+  "bf16": true,
+  "attn_impl": "auto",
+  "precision": "mixed_bf16",
+  "grad_clip": 0.0,
+  "loss_agg_mode": "token-mean",
+  "ordered": false,
+  "within_step_shuffle": false,
+  "system_prompt_mode": "from_record",
+  "score_threshold": 0.0,
+  "drop_all_pass_groups": false,
+  "max_rl_step": null,
+  "num_epochs": 1,
+  "replay_strategy": "none",
+  "replay_lambda": 0.0,
+  "replay_batch_size": 1,
+  "replay_cooldown_step": 128.0,
+  "replay_ce_tau": 1.0,
+  "replay_ce_init": "insert",
+  "replay_group_size": 8,
+  "replay_seed": null,
+  "max_train_steps": null,
+  "num_checkpoints": 10,
+  "checkpoint_after_resume_steps": null,
+  "save_optimizer_state": true,
+  "prune_old_optimizer": true,
+  "save_optimizer_every_n_ckpts": 5,
+  "save_optimizer_first_ckpt": true,
+  "snapshot_dtype": "float32",
+  "snapshot_device": "cpu",
+  "metrics_log_every": 10,
+  "ddp_timeout_seconds": 3600,
+  "memory_canary_longest_first": false,
+  "log_cuda_memory": true,
+  "save_final_model": true,
+  "wandb_project": "rl_forgetting_sft",
+  "wandb_entity": "jingyanshen-new-york-university",
+  "wandb_run_name": "qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925_segment_g3",
+  "wandb_group": "code_sft_20260922_qwen3_1p7b_s500",
+  "wandb_run_id_file": "/mnt/mystorageoutput/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925/wandb_run_id_generation_3.txt",
+  "resume_from": "/mnt/mystorageoutput/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925/checkpoints/step_51",
+  "grad_accum_steps": 64
+}

training/q3_1p7b_sft_shuffled/training_complete.json

@@ -0,0 +1,11 @@
+{
+  "azure_run_id": "mango_button_p5zgp0zwbm",
+  "completed_at": "2026-09-25T07:33:13.812772+00:00",
+  "host": "node-0",
+  "offline_wandb_sync_required": true,
+  "schema_version": 1,
+  "status": "training_complete",
+  "wandb_generation": 3,
+  "wandb_mode": "offline",
+  "wandb_run_id_file": "/mnt/mystorageoutput/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925/wandb_run_id_generation_3.txt"
+}

training/q3_1p7b_sft_shuffled/wandb/offline-run-20260925_071649-7y8jx2s3/files/requirements.txt

@@ -0,0 +1,266 @@
+aiohttp==3.14.3
+rpds-py==0.30.0
+cupy-cuda12x==14.2.0
+pyfiglet==1.0.2
+fastapi-cli==0.0.32
+flash_attn==2.7.4.post1
+websockets==16.1.1
+xformers==0.0.29.post2
+astor==0.8.1
+python-multipart==0.0.32
+anyio==4.15.1
+pytz==2026.3.post1
+jsonschema-specifications==2025.9.1
+graphene==3.4.3
+multiprocess==0.70.19
+vllm==0.8.5
+airportsdata==20260315
+blake3==1.0.9
+Deprecated==1.3.1
+numba==0.61.2
+python-dateutil==2.9.0.post0
+opentelemetry-exporter-otlp-proto-common==1.26.0
+nvidia-cuda-nvrtc-cu12==12.4.127
+graphql-core==3.2.12
+outlines_core==0.1.26
+virtualenv==21.7.9
+shellingham==1.5.4
+nvidia-cufft-cu12==11.2.1.3
+truststore==0.10.4
+pydantic-settings==2.15.0
+huggingface_hub==0.36.2
+typer==0.27.2
+zipp==4.1.0
+requests==2.34.2
+uvloop==0.22.1
+setuptools==80.9.0
+fsspec==2026.6.0
+arckit==0.1.0
+filelock==3.32.6
+safetensors==0.8.0
+opencv-python-headless==5.0.0.93
+llguidance==0.7.30
+pillow==12.3.0
+cffi==2.1.1
+Pygments==2.21.0
+opentelemetry-sdk==1.26.0
+smmap==5.0.3
+opentelemetry-exporter-prometheus==0.47b0
+mdurl==0.1.2
+itsdangerous==2.2.0
+gunicorn==26.2.0
+aiohttp-cors==0.8.1
+zss==1.2.0
+pycountry==26.2.16
+opencensus-context==0.1.3
+pip==26.2.1
+tabulate==0.9.0
+markdown-it-py==4.2.0
+lm-format-enforcer==0.10.12
+triton==3.2.0
+pyasn1==0.6.4
+ninja==1.13.2
+tensordict==0.6.2
+Jinja2==3.1.6
+fastapi==0.141.1
+pyasn1_modules==0.4.2
+scipy==1.15.3
+importlib_metadata==8.0.0
+sympy==1.13.1
+pycparser==3.0
+mpmath==1.3.0
+nvidia-cusparse-cu12==12.3.1.170
+hydra-core==1.3.6
+Flask==3.1.3
+scikit-learn==1.7.2
+datasets==5.0.1
+opentelemetry-proto==1.26.0
+nvidia-cusparselt-cu12==0.6.2
+PyYAML==6.0.3
+attrs==26.1.0
+nvidia-curand-cu12==10.3.5.147
+nvidia-cudnn-cu12==9.1.0.70
+SQLAlchemy==2.0.52
+mlflow-skinny==3.16.0
+rich-toolkit==0.20.5
+xxhash==4.0.1
+urllib3==2.7.0
+charset-normalizer==3.5.1
+protobuf==4.25.9
+platformdirs==4.11.8
+httpx==0.28.1
+wandb==0.28.0
+pyarrow==25.0.1
+openai==3.13.0
+peft==0.20.0
+packaging==26.3
+certifi==2026.7.22
+Mako==1.4.1
+prometheus_client==0.26.0
+exceptiongroup==1.3.1
+nvidia-cuda-cupti-cu12==12.4.127
+idna==3.19
+cachetools==7.1.8
+python-json-logger==4.2.0
+distlib==0.4.3
+pydantic_core==2.46.5
+mistral_common==1.11.7
+orjson==3.12.0
+fastar==0.12.0
+tokenizers==0.22.2
+watchfiles==1.2.0
+nvidia-cusolver-cu12==11.6.1.9
+smart_open==8.0.1
+wheel==0.47.0
+magiccube==0.3.0
+conda-pack==0.9.2
+tiktoken==0.14.0
+accelerate==1.15.0
+prettytable==3.18.0
+jsonschema==4.26.0
+prometheus-fastapi-instrumentator==8.1.0
+mlflow-tracing==3.16.0
+python-dotenv==1.2.3
+aiosignal==1.4.0
+py-spy==0.4.2
+tqdm==4.70.1
+opentelemetry-exporter-otlp==1.26.0
+MarkupSafe==3.0.3
+interegular==0.3.3
+networkx==3.4.2
+contourpy==1.3.2
+gguf==0.19.0
+async-timeout==5.0.1
+h11==0.16.0
+xgrammar==0.1.18
+httpx2==2.12.0
+matplotlib==3.10.9
+huey==3.4.0
+einops==0.8.2
+opencensus==0.11.4
+msgpack==1.2.2
+pandas==2.3.3
+llvmlite==0.44.0
+sniffio==1.3.1
+GitPython==3.1.62
+opentelemetry-exporter-otlp-proto-grpc==1.26.0
+greenlet==3.5.5
+six==1.17.0
+partial-json-parser==0.2.1.1.post7
+outlines==0.1.11
+google-auth==2.58.0
+joblib==1.6.0
+mlflow==3.16.0
+nvidia-cuda-runtime-cu12==12.4.127
+pydantic-extra-types==2.11.1
+databricks-sdk==0.138.0
+referencing==0.37.0
+propcache==0.5.2
+nest-asyncio==1.6.0
+torch==2.6.0
+math-verify==0.9.0
+compressed-tensors==0.9.3
+googleapis-common-protos==1.75.0
+yarl==1.24.5
+cellpylib==2.4.0
+nvidia-nvjitlink-cu12==12.4.127
+opentelemetry-semantic-conventions-ai==0.4.13
+ray==2.47.1
+sqlparse==0.6.0
+jiter==0.16.0
+drawsvg==2.4.2
+email-validator==2.3.0
+opentelemetry-exporter-otlp-proto-http==1.26.0
+hf-xet==1.6.0
+torchdata==0.11.0
+transformers==4.57.6
+codetiming==1.4.0
+httpcore2==2.12.0
+pydantic==2.13.5
+cryptography==50.0.1
+torchvision==0.21.0
+psutil==7.2.2
+starlette==1.6.0
+uvicorn==0.52.4
+diskcache==5.6.3
+nvidia-nccl-cu12==2.21.5
+detect-installer==0.2.1
+typing-inspection==0.4.4
+tomli==2.4.1
+opentelemetry-semantic-conventions==0.47b0
+numpy==2.2.6
+alembic==1.19.2
+cycler==0.12.1
+gitdb==4.0.12
+google-api-core==2.30.3
+rich==15.0.0
+reasoning_gym==0.1.25
+httptools==0.8.0
+httpcore==1.0.9
+rignore==0.8.1
+nvidia-cublas-cu12==12.4.5.8
+annotated-doc==0.0.5
+pyzmq==27.2.0
+antlr4-python3-runtime==4.9.3
+agent-detector==2.0.0
+fonttools==4.65.0
+latex2sympy2_extended==1.11.0
+threadpoolctl==3.6.0
+regex==2026.9.10
+lark==1.2.2
+verl==0.4.1.dev0
+blinker==1.9.0
+flask-cors==6.0.5
+grpcio==1.83.1
+colorful==0.5.8
+skops==0.14.0
+tzdata==2026.3
+dnspython==2.8.0
+wcwidth==0.8.3
+sentencepiece==0.2.2
+python-discovery==1.6.0
+sentry-sdk==2.69.1
+bfi==1.0.4
+Werkzeug==3.1.8
+nvidia-nvtx-cu12==12.4.127
+click==8.5.0
+annotated-types==0.8.0
+aiohappyeyeballs==2.7.1
+cloudpickle==3.1.2
+fastapi-cloud-cli==0.26.0
+msgspec==0.21.1
+pylatexenc==2.11
+proto-plus==1.28.2
+dill==0.4.1
+py-cpuinfo==9.0.0
+pyparsing==3.3.2
+typing_extensions==4.16.0
+wrapt==2.4.1
+depyf==0.18.0
+kiwisolver==1.5.1
+cuda-pathfinder==1.8.1
+pybind11==3.1.0
+opentelemetry-api==1.26.0
+frozenlist==1.8.0
+pycosat==0.6.6
+graphql-relay==3.2.0
+docker==7.2.0
+multidict==6.8.0
+torchaudio==2.6.0
+omegaconf==2.3.1
+platformdirs==4.2.2
+tomli==2.0.1
+autocommand==2.2.2
+zipp==3.19.2
+more-itertools==10.3.0
+typing_extensions==4.12.2
+jaraco.functools==4.0.1
+importlib_metadata==8.0.0
+jaraco.text==3.12.1
+wheel==0.45.1
+jaraco.context==5.3.0
+jaraco.collections==5.1.0
+backports.tarfile==1.2.0
+packaging==24.2
+inflect==7.3.1
+typeguard==4.3.0

training/q3_1p7b_sft_shuffled/wandb/run-20260925_060938-ade2qena/files/requirements.txt

@@ -0,0 +1,266 @@
+tqdm==4.70.1
+sqlparse==0.6.0
+huggingface_hub==0.36.2
+rich-toolkit==0.20.5
+torch==2.6.0
+threadpoolctl==3.6.0
+numpy==2.2.6
+nvidia-curand-cu12==10.3.5.147
+httpx==0.28.1
+anyio==4.15.1
+vllm==0.8.5
+gunicorn==26.2.0
+prettytable==3.18.0
+mlflow==3.16.0
+importlib_metadata==8.0.0
+cryptography==50.0.1
+graphql-relay==3.2.0
+nvidia-cudnn-cu12==9.1.0.70
+ray==2.47.1
+psutil==7.2.2
+tomli==2.4.1
+omegaconf==2.3.1
+Jinja2==3.1.6
+nvidia-cuda-cupti-cu12==12.4.127
+hf-xet==1.6.0
+pyarrow==25.0.1
+pyasn1_modules==0.4.2
+rich==15.0.0
+jsonschema-specifications==2025.9.1
+einops==0.8.2
+nvidia-cufft-cu12==11.2.1.3
+tiktoken==0.14.0
+tabulate==0.9.0
+aiosignal==1.4.0
+sentencepiece==0.2.2
+cellpylib==2.4.0
+nvidia-cuda-nvrtc-cu12==12.4.127
+wheel==0.47.0
+referencing==0.37.0
+opentelemetry-exporter-otlp-proto-common==1.26.0
+rpds-py==0.30.0
+yarl==1.24.5
+nest-asyncio==1.6.0
+reasoning_gym==0.1.25
+flash_attn==2.7.4.post1
+kiwisolver==1.5.1
+pandas==2.3.3
+antlr4-python3-runtime==4.9.3
+databricks-sdk==0.138.0
+jsonschema==4.26.0
+llvmlite==0.44.0
+partial-json-parser==0.2.1.1.post7
+aiohttp-cors==0.8.1
+python-multipart==0.0.32
+fastapi==0.141.1
+virtualenv==21.7.9
+pycparser==3.0
+peft==0.20.0
+networkx==3.4.2
+py-cpuinfo==9.0.0
+grpcio==1.83.1
+opencensus-context==0.1.3
+conda-pack==0.9.2
+multiprocess==0.70.19
+SQLAlchemy==2.0.52
+opentelemetry-api==1.26.0
+nvidia-cusolver-cu12==11.6.1.9
+pybind11==3.1.0
+xformers==0.0.29.post2
+greenlet==3.5.5
+dnspython==2.8.0
+uvicorn==0.52.4
+sniffio==1.3.1
+httpx2==2.12.0
+joblib==1.6.0
+mlflow-tracing==3.16.0
+requests==2.34.2
+email-validator==2.3.0
+pydantic-extra-types==2.11.1
+proto-plus==1.28.2
+attrs==26.1.0
+safetensors==0.8.0
+skops==0.14.0
+h11==0.16.0
+huey==3.4.0
+PyYAML==6.0.3
+httpcore2==2.12.0
+pip==26.2.1
+xgrammar==0.1.18
+pydantic_core==2.46.5
+graphql-core==3.2.12
+blake3==1.0.9
+fastar==0.12.0
+markdown-it-py==4.2.0
+pyfiglet==1.0.2
+triton==3.2.0
+pyparsing==3.3.2
+torchvision==0.21.0
+magiccube==0.3.0
+itsdangerous==2.2.0
+httpcore==1.0.9
+websockets==16.1.1
+verl==0.4.1.dev0
+cachetools==7.1.8
+mistral_common==1.11.7
+openai==3.13.0
+cupy-cuda12x==14.2.0
+exceptiongroup==1.3.1
+aiohttp==3.14.3
+smmap==5.0.3
+frozenlist==1.8.0
+depyf==0.18.0
+wandb==0.28.0
+certifi==2026.7.22
+matplotlib==3.10.9
+google-auth==2.58.0
+python-json-logger==4.2.0
+annotated-doc==0.0.5
+accelerate==1.15.0
+compressed-tensors==0.9.3
+six==1.17.0
+nvidia-nvjitlink-cu12==12.4.127
+opentelemetry-semantic-conventions==0.47b0
+scikit-learn==1.7.2
+mlflow-skinny==3.16.0
+opentelemetry-sdk==1.26.0
+mdurl==0.1.2
+uvloop==0.22.1
+pydantic==2.13.5
+cycler==0.12.1
+math-verify==0.9.0
+transformers==4.57.6
+prometheus-fastapi-instrumentator==8.1.0
+filelock==3.32.6
+jiter==0.16.0
+gitdb==4.0.12
+smart_open==8.0.1
+torchdata==0.11.0
+contourpy==1.3.2
+opentelemetry-proto==1.26.0
+hydra-core==1.3.6
+alembic==1.19.2
+cloudpickle==3.1.2
+sympy==1.13.1
+astor==0.8.1
+Mako==1.4.1
+MarkupSafe==3.0.3
+pydantic-settings==2.15.0
+starlette==1.6.0
+colorful==0.5.8
+diskcache==5.6.3
+bfi==1.0.4
+shellingham==1.5.4
+typing_extensions==4.16.0
+wrapt==2.4.1
+propcache==0.5.2
+httptools==0.8.0
+nvidia-cublas-cu12==12.4.5.8
+codetiming==1.4.0
+nvidia-nccl-cu12==2.21.5
+pytz==2026.3.post1
+charset-normalizer==3.5.1
+Pygments==2.21.0
+zss==1.2.0
+blinker==1.9.0
+llguidance==0.7.30
+flask-cors==6.0.5
+opentelemetry-exporter-prometheus==0.47b0
+GitPython==3.1.62
+tzdata==2026.3
+pycosat==0.6.6
+opentelemetry-exporter-otlp-proto-http==1.26.0
+lm-format-enforcer==0.10.12
+googleapis-common-protos==1.75.0
+msgpack==1.2.2
+xxhash==4.0.1
+agent-detector==2.0.0
+outlines==0.1.11
+torchaudio==2.6.0
+airportsdata==20260315
+rignore==0.8.1
+mpmath==1.3.0
+dill==0.4.1
+async-timeout==5.0.1
+lark==1.2.2
+fastapi-cloud-cli==0.26.0
+sentry-sdk==2.69.1
+graphene==3.4.3
+orjson==3.12.0
+typing-inspection==0.4.4
+protobuf==4.25.9
+urllib3==2.7.0
+drawsvg==2.4.2
+opencensus==0.11.4
+tokenizers==0.22.2
+google-api-core==2.30.3
+pycountry==26.2.16
+interegular==0.3.3
+truststore==0.10.4
+annotated-types==0.8.0
+opentelemetry-exporter-otlp-proto-grpc==1.26.0
+regex==2026.9.10
+nvidia-cuda-runtime-cu12==12.4.127
+aiohappyeyeballs==2.7.1
+numba==0.61.2
+zipp==4.1.0
+tensordict==0.6.2
+pyasn1==0.6.4
+ninja==1.13.2
+packaging==26.3
+click==8.5.0
+py-spy==0.4.2
+nvidia-cusparse-cu12==12.3.1.170
+docker==7.2.0
+scipy==1.15.3
+prometheus_client==0.26.0
+detect-installer==0.2.1
+wcwidth==0.8.3
+opencv-python-headless==5.0.0.93
+arckit==0.1.0
+gguf==0.19.0
+Deprecated==1.3.1
+fastapi-cli==0.0.32
+distlib==0.4.3
+nvidia-cusparselt-cu12==0.6.2
+fsspec==2026.6.0
+cuda-pathfinder==1.8.1
+nvidia-nvtx-cu12==12.4.127
+multidict==6.8.0
+datasets==5.0.1
+python-discovery==1.6.0
+cffi==2.1.1
+opentelemetry-semantic-conventions-ai==0.4.13
+watchfiles==1.2.0
+platformdirs==4.11.8
+msgspec==0.21.1
+python-dotenv==1.2.3
+pylatexenc==2.11
+outlines_core==0.1.26
+Werkzeug==3.1.8
+Flask==3.1.3
+latex2sympy2_extended==1.11.0
+setuptools==80.9.0
+python-dateutil==2.9.0.post0
+fonttools==4.65.0
+opentelemetry-exporter-otlp==1.26.0
+idna==3.19
+pillow==12.3.0
+typer==0.27.2
+pyzmq==27.2.0
+backports.tarfile==1.2.0
+importlib_metadata==8.0.0
+jaraco.collections==5.1.0
+packaging==24.2
+typing_extensions==4.12.2
+more-itertools==10.3.0
+inflect==7.3.1
+jaraco.context==5.3.0
+platformdirs==4.2.2
+autocommand==2.2.2
+jaraco.text==3.12.1
+typeguard==4.3.0
+wheel==0.45.1
+tomli==2.0.1
+jaraco.functools==4.0.1
+zipp==3.19.2

training/q3_1p7b_sft_shuffled/wandb/run-20260925_060938-ade2qena/files/wandb-metadata.json

@@ -0,0 +1,112 @@
+{
+  "os": "Linux-6.8.0-1062-azure-x86_64-with-glibc2.35",
+  "python": "CPython 3.10.21",
+  "startedAt": "2026-09-25T06:09:38.934083Z",
+  "args": [
+    "--dataset",
+    "qwen3_1p7b_s500_code_sft_data",
+    "--model_path",
+    "/mnt/mystorageoutput/projects/rl-forgetting/base_models/Qwen3-1.7B-Base-ea980cb0",
+    "--output_dir",
+    "/mnt/output/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925",
+    "--system_prompt_mode",
+    "from_record",
+    "--lr",
+    "1e-5",
+    "--batch_size",
+    "1",
+    "--effective_batch_size",
+    "256",
+    "--max_seq_length",
+    "4096",
+    "--ddp_timeout_seconds",
+    "3600",
+    "--schedule",
+    "constant",
+    "--warmup_ratio",
+    "0",
+    "--num_epochs",
+    "1",
+    "--seed",
+    "42",
+    "--log_cuda_memory",
+    "--num_checkpoints",
+    "10",
+    "--save_optimizer_every_n_ckpts",
+    "5",
+    "--prune_old_optimizer",
+    "--save_optimizer_first_ckpt",
+    "--metrics_log_every",
+    "10",
+    "--precision",
+    "mixed_bf16",
+    "--grad_clip",
+    "0",
+    "--snapshot_device",
+    "cpu",
+    "--snapshot_dtype",
+    "float32",
+    "--wandb_project",
+    "rl_forgetting_sft",
+    "--wandb_entity",
+    "jingyanshen-new-york-university",
+    "--wandb_group",
+    "code_sft_20260922_qwen3_1p7b_s500",
+    "--wandb_run_name",
+    "qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925_segment_g1",
+    "--wandb_run_id_file",
+    "/mnt/output/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925/wandb_run_id_generation_1.txt"
+  ],
+  "program": "/scratch/amlt_code/sft/run_sft.py",
+  "codePath": "sft/run_sft.py",
+  "codePathLocal": "sft/run_sft.py",
+  "email": "js15262@nyu.edu",
+  "root": "/mnt/output/projects/rl-forgetting/amlt-results/7209658985.41605-70a912e4-65e2/qwen3_1p7b_s500_code_shuffled_lr1e-5_bs256_seed42_gcr_rerun_20260925",
+  "host": "node-0",
+  "executable": "/tmp/rlf-code-runtime/bin/python",
+  "cpu_count": 96,
+  "cpu_count_logical": 96,
+  "gpu": "NVIDIA H100 80GB HBM3",
+  "gpu_count": 4,
+  "disk": {
+    "/": {
+      "total": "30601604947968",
+      "used": "843751456768"
+    }
+  },
+  "memory": {
+    "total": "2008079089664"
+  },
+  "gpu_nvidia": [
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper",
+      "uuid": "GPU-bdd68df3-74c2-3c86-5bb3-2fb37a8467aa"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper",
+      "uuid": "GPU-0eb031cd-bc39-e289-aa5f-491995612f2c"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper",
+      "uuid": "GPU-dc94e8ac-c4bf-3968-1ffb-653db2b301e3"
+    },
+    {
+      "name": "NVIDIA H100 80GB HBM3",
+      "memoryTotal": "85520809984",
+      "cudaCores": 16896,
+      "architecture": "Hopper",
+      "uuid": "GPU-69e7739a-0c44-75df-af9f-811136725212"
+    }
+  ],
+  "cudaVersion": "13.0",
+  "writerId": "rapeft57noznsok8ugkhxhgjypwccziq"
+}

training/q3_1p7b_sft_shuffled/wandb_run_id_generation_1.txt

@@ -0,0 +1 @@
+ade2qena

training/q3_1p7b_sft_shuffled/wandb_run_id_generation_2.txt

@@ -0,0 +1 @@
+c1cb400b

training/q3_1p7b_sft_shuffled/wandb_run_id_generation_3.txt

@@ -0,0 +1 @@
+7y8jx2s3

training/q3_1p7b_sft_shuffled/wandb_segments/generation_1.json

@@ -0,0 +1,8 @@
+{
+  "attempt_id": "bold_rod_d75dnm49lp|node-0|a15a53f32e7e4e7090ef98b58648857c",
+  "created_at": "2026-09-25T06:08:29.584953+00:00",
+  "generation": 1,
+  "resume_step": 0,
+  "schema_version": 1,
+  "wandb_run_id": "ade2qena"
+}

training/q3_1p7b_sft_shuffled/wandb_segments/generation_2.json

@@ -0,0 +1,8 @@
+{
+  "attempt_id": "bold_rod_d75dnm49lp|node-0|e45ce8697e4348c4b105e610d34b812c",
+  "created_at": "2026-09-25T06:37:37.462035+00:00",
+  "generation": 2,
+  "resume_step": 51,
+  "schema_version": 1,
+  "wandb_run_id": "c1cb400b"
+}

training/q3_1p7b_sft_shuffled/wandb_segments/generation_3.json

@@ -0,0 +1,8 @@
+{
+  "attempt_id": "mango_button_p5zgp0zwbm|node-0|c150453c68ac4618ab0856ce0b8a899e",
+  "created_at": "2026-09-25T07:15:00.825913+00:00",
+  "generation": 3,
+  "resume_step": 51,
+  "schema_version": 1,
+  "wandb_run_id": "7y8jx2s3"
+}