FastPlus40m Backup checkpoint-800 at global step 800
Browse files
checkpoint-800/config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"FastPlus40mForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"dtype": "float32",
|
| 6 |
+
"hidden_size": 512,
|
| 7 |
+
"initializer_range": 0.02,
|
| 8 |
+
"intermediate_size": 1536,
|
| 9 |
+
"kd_alpha": 0.4,
|
| 10 |
+
"kd_temperature": 2.5,
|
| 11 |
+
"max_position_embeddings": 512,
|
| 12 |
+
"model_type": "fastplus_40m",
|
| 13 |
+
"num_attention_heads": 8,
|
| 14 |
+
"num_hidden_layers": 11,
|
| 15 |
+
"tie_word_embeddings": true,
|
| 16 |
+
"transformers_version": "5.10.2",
|
| 17 |
+
"use_cache": false,
|
| 18 |
+
"vocab_size": 8160
|
| 19 |
+
}
|
checkpoint-800/generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"output_attentions": false,
|
| 4 |
+
"output_hidden_states": false,
|
| 5 |
+
"transformers_version": "5.10.2"
|
| 6 |
+
}
|
checkpoint-800/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
checkpoint-800/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"bos_token": "<s>",
|
| 4 |
+
"eos_token": "</s>",
|
| 5 |
+
"extra_special_tokens": [
|
| 6 |
+
"<|user|>",
|
| 7 |
+
"<|assistant|>"
|
| 8 |
+
],
|
| 9 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 10 |
+
"pad_token": "<pad>",
|
| 11 |
+
"tokenizer_class": "TokenizersBackend",
|
| 12 |
+
"unk_token": "<unk>"
|
| 13 |
+
}
|
checkpoint-800/trainer_state.json
ADDED
|
@@ -0,0 +1,674 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.4148705150228503,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 800,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.005185881437785629,
|
| 14 |
+
"grad_norm": 231.2135467529297,
|
| 15 |
+
"learning_rate": 2.5499999999999996e-05,
|
| 16 |
+
"loss": 145.025048828125,
|
| 17 |
+
"original_ce_loss": 9.0046,
|
| 18 |
+
"step": 10
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"epoch": 0.010371762875571258,
|
| 22 |
+
"grad_norm": 239.0395050048828,
|
| 23 |
+
"learning_rate": 6.8e-05,
|
| 24 |
+
"loss": 144.4317626953125,
|
| 25 |
+
"original_ce_loss": 8.9676,
|
| 26 |
+
"step": 20
|
| 27 |
+
},
|
| 28 |
+
{
|
| 29 |
+
"epoch": 0.015557644313356886,
|
| 30 |
+
"grad_norm": 382.8333435058594,
|
| 31 |
+
"learning_rate": 0.00010625,
|
| 32 |
+
"loss": 142.0313720703125,
|
| 33 |
+
"original_ce_loss": 8.8177,
|
| 34 |
+
"step": 30
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 0.020743525751142515,
|
| 38 |
+
"grad_norm": 247.25132751464844,
|
| 39 |
+
"learning_rate": 0.00014874999999999998,
|
| 40 |
+
"loss": 130.05081787109376,
|
| 41 |
+
"original_ce_loss": 8.0691,
|
| 42 |
+
"step": 40
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"epoch": 0.025929407188928142,
|
| 46 |
+
"grad_norm": 44.44298553466797,
|
| 47 |
+
"learning_rate": 0.000187,
|
| 48 |
+
"loss": 117.4221435546875,
|
| 49 |
+
"original_ce_loss": 7.2801,
|
| 50 |
+
"step": 50
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"epoch": 0.031115288626713773,
|
| 54 |
+
"grad_norm": 28.125579833984375,
|
| 55 |
+
"learning_rate": 0.0002295,
|
| 56 |
+
"loss": 111.31640625,
|
| 57 |
+
"original_ce_loss": 6.8989,
|
| 58 |
+
"step": 60
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.0363011700644994,
|
| 62 |
+
"grad_norm": 251.75486755371094,
|
| 63 |
+
"learning_rate": 0.000272,
|
| 64 |
+
"loss": 104.603857421875,
|
| 65 |
+
"original_ce_loss": 6.48,
|
| 66 |
+
"step": 70
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.04148705150228503,
|
| 70 |
+
"grad_norm": 16.69808006286621,
|
| 71 |
+
"learning_rate": 0.0003145,
|
| 72 |
+
"loss": 101.7426025390625,
|
| 73 |
+
"original_ce_loss": 6.3018,
|
| 74 |
+
"step": 80
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.04667293294007066,
|
| 78 |
+
"grad_norm": 12.393045425415039,
|
| 79 |
+
"learning_rate": 0.00034849999999999996,
|
| 80 |
+
"loss": 101.3855712890625,
|
| 81 |
+
"original_ce_loss": 6.2801,
|
| 82 |
+
"step": 90
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"epoch": 0.051858814377856284,
|
| 86 |
+
"grad_norm": 734.70263671875,
|
| 87 |
+
"learning_rate": 0.00038675,
|
| 88 |
+
"loss": 101.10267944335938,
|
| 89 |
+
"original_ce_loss": 6.2631,
|
| 90 |
+
"step": 100
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.057044695815641915,
|
| 94 |
+
"grad_norm": 116.42232513427734,
|
| 95 |
+
"learning_rate": 0.0004249996865262593,
|
| 96 |
+
"loss": 100.80792846679688,
|
| 97 |
+
"original_ce_loss": 6.2452,
|
| 98 |
+
"step": 110
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"epoch": 0.062230577253427545,
|
| 102 |
+
"grad_norm": 13.761225700378418,
|
| 103 |
+
"learning_rate": 0.000424962070796437,
|
| 104 |
+
"loss": 100.12683715820313,
|
| 105 |
+
"original_ce_loss": 6.203,
|
| 106 |
+
"step": 120
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"epoch": 0.06741645869121317,
|
| 110 |
+
"grad_norm": 27.79854965209961,
|
| 111 |
+
"learning_rate": 0.0004248617730346077,
|
| 112 |
+
"loss": 99.19635009765625,
|
| 113 |
+
"original_ce_loss": 6.1455,
|
| 114 |
+
"step": 130
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.0726023401289988,
|
| 118 |
+
"grad_norm": 33.40275192260742,
|
| 119 |
+
"learning_rate": 0.00042469882283131187,
|
| 120 |
+
"loss": 98.95189208984375,
|
| 121 |
+
"original_ce_loss": 6.1309,
|
| 122 |
+
"step": 140
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.07778822156678443,
|
| 126 |
+
"grad_norm": 13.005568504333496,
|
| 127 |
+
"learning_rate": 0.0004244732682612475,
|
| 128 |
+
"loss": 98.00054931640625,
|
| 129 |
+
"original_ce_loss": 6.0723,
|
| 130 |
+
"step": 150
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 0.08297410300457006,
|
| 134 |
+
"grad_norm": 17.792335510253906,
|
| 135 |
+
"learning_rate": 0.00042418517586908627,
|
| 136 |
+
"loss": 97.31512451171875,
|
| 137 |
+
"original_ce_loss": 6.0304,
|
| 138 |
+
"step": 160
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 0.08815998444235569,
|
| 142 |
+
"grad_norm": 18.2156925201416,
|
| 143 |
+
"learning_rate": 0.00042383463064984164,
|
| 144 |
+
"loss": 96.42101440429687,
|
| 145 |
+
"original_ce_loss": 5.9755,
|
| 146 |
+
"step": 170
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.09334586588014132,
|
| 150 |
+
"grad_norm": 19.725210189819336,
|
| 151 |
+
"learning_rate": 0.00042342173602379264,
|
| 152 |
+
"loss": 94.94822998046875,
|
| 153 |
+
"original_ce_loss": 5.8844,
|
| 154 |
+
"step": 180
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"epoch": 0.09853174731792694,
|
| 158 |
+
"grad_norm": 16.279741287231445,
|
| 159 |
+
"learning_rate": 0.0004229466138059723,
|
| 160 |
+
"loss": 93.37152099609375,
|
| 161 |
+
"original_ce_loss": 5.7867,
|
| 162 |
+
"step": 190
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"epoch": 0.10371762875571257,
|
| 166 |
+
"grad_norm": 13.233877182006836,
|
| 167 |
+
"learning_rate": 0.00042240940417022886,
|
| 168 |
+
"loss": 91.78927001953124,
|
| 169 |
+
"original_ce_loss": 5.6886,
|
| 170 |
+
"step": 200
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.1089035101934982,
|
| 174 |
+
"grad_norm": 16.279666900634766,
|
| 175 |
+
"learning_rate": 0.00042181026560787066,
|
| 176 |
+
"loss": 90.180810546875,
|
| 177 |
+
"original_ce_loss": 5.589,
|
| 178 |
+
"step": 210
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.11408939163128383,
|
| 182 |
+
"grad_norm": 17.527315139770508,
|
| 183 |
+
"learning_rate": 0.000421149374880907,
|
| 184 |
+
"loss": 88.39944458007812,
|
| 185 |
+
"original_ce_loss": 5.4785,
|
| 186 |
+
"step": 220
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"epoch": 0.11927527306906946,
|
| 190 |
+
"grad_norm": 17.70699119567871,
|
| 191 |
+
"learning_rate": 0.00042042692696989874,
|
| 192 |
+
"loss": 86.77321166992188,
|
| 193 |
+
"original_ce_loss": 5.3776,
|
| 194 |
+
"step": 230
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"epoch": 0.12446115450685509,
|
| 198 |
+
"grad_norm": 13.845442771911621,
|
| 199 |
+
"learning_rate": 0.00041964313501643334,
|
| 200 |
+
"loss": 85.32883911132812,
|
| 201 |
+
"original_ce_loss": 5.2882,
|
| 202 |
+
"step": 240
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"epoch": 0.12964703594464072,
|
| 206 |
+
"grad_norm": 15.751480102539062,
|
| 207 |
+
"learning_rate": 0.0004188854638487832,
|
| 208 |
+
"loss": 84.12140502929688,
|
| 209 |
+
"original_ce_loss": 5.2135,
|
| 210 |
+
"step": 250
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.13483291738242634,
|
| 214 |
+
"grad_norm": NaN,
|
| 215 |
+
"learning_rate": 0.00041798577018333747,
|
| 216 |
+
"loss": 82.77281494140625,
|
| 217 |
+
"original_ce_loss": 5.13,
|
| 218 |
+
"step": 260
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"epoch": 0.14001879882021198,
|
| 222 |
+
"grad_norm": 21.7114315032959,
|
| 223 |
+
"learning_rate": 0.0004171242045703103,
|
| 224 |
+
"loss": 81.66472778320312,
|
| 225 |
+
"original_ce_loss": 5.0614,
|
| 226 |
+
"step": 270
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.1452046802579976,
|
| 230 |
+
"grad_norm": 17.575490951538086,
|
| 231 |
+
"learning_rate": 0.0004161095673476243,
|
| 232 |
+
"loss": 80.130908203125,
|
| 233 |
+
"original_ce_loss": 4.9662,
|
| 234 |
+
"step": 280
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.15039056169578324,
|
| 238 |
+
"grad_norm": 45.984519958496094,
|
| 239 |
+
"learning_rate": 0.0004150348598198265,
|
| 240 |
+
"loss": 77.93204956054687,
|
| 241 |
+
"original_ce_loss": 4.8296,
|
| 242 |
+
"step": 290
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"epoch": 0.15557644313356886,
|
| 246 |
+
"grad_norm": 18.64594078063965,
|
| 247 |
+
"learning_rate": 0.00041401652462566313,
|
| 248 |
+
"loss": 77.3225341796875,
|
| 249 |
+
"original_ce_loss": 4.7922,
|
| 250 |
+
"step": 300
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.16076232457135448,
|
| 254 |
+
"grad_norm": 19.417156219482422,
|
| 255 |
+
"learning_rate": 0.00041282857161178824,
|
| 256 |
+
"loss": 76.28924560546875,
|
| 257 |
+
"original_ce_loss": 4.7283,
|
| 258 |
+
"step": 310
|
| 259 |
+
},
|
| 260 |
+
{
|
| 261 |
+
"epoch": 0.16594820600914012,
|
| 262 |
+
"grad_norm": 19.997188568115234,
|
| 263 |
+
"learning_rate": 0.0004118356381184281,
|
| 264 |
+
"loss": 74.18194580078125,
|
| 265 |
+
"original_ce_loss": 4.5972,
|
| 266 |
+
"step": 320
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 0.17113408744692574,
|
| 270 |
+
"grad_norm": 18.9919376373291,
|
| 271 |
+
"learning_rate": 0.00041054156498848937,
|
| 272 |
+
"loss": 72.67985229492187,
|
| 273 |
+
"original_ce_loss": 4.504,
|
| 274 |
+
"step": 330
|
| 275 |
+
},
|
| 276 |
+
{
|
| 277 |
+
"epoch": 0.17631996888471138,
|
| 278 |
+
"grad_norm": 20.276124954223633,
|
| 279 |
+
"learning_rate": 0.0004091890642640685,
|
| 280 |
+
"loss": 70.95518798828125,
|
| 281 |
+
"original_ce_loss": 4.397,
|
| 282 |
+
"step": 340
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.181505850322497,
|
| 286 |
+
"grad_norm": 21.962339401245117,
|
| 287 |
+
"learning_rate": 0.000407778534969299,
|
| 288 |
+
"loss": 69.48126831054688,
|
| 289 |
+
"original_ce_loss": 4.3056,
|
| 290 |
+
"step": 350
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.18669173176028264,
|
| 294 |
+
"grad_norm": 15.767695426940918,
|
| 295 |
+
"learning_rate": 0.0004063103932483052,
|
| 296 |
+
"loss": 68.26876220703124,
|
| 297 |
+
"original_ce_loss": 4.2305,
|
| 298 |
+
"step": 360
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 0.19187761319806826,
|
| 302 |
+
"grad_norm": 19.607179641723633,
|
| 303 |
+
"learning_rate": 0.0004047850722424284,
|
| 304 |
+
"loss": 66.68286743164063,
|
| 305 |
+
"original_ce_loss": 4.1321,
|
| 306 |
+
"step": 370
|
| 307 |
+
},
|
| 308 |
+
{
|
| 309 |
+
"epoch": 0.19706349463585388,
|
| 310 |
+
"grad_norm": 23.764141082763672,
|
| 311 |
+
"learning_rate": 0.0004032030219624389,
|
| 312 |
+
"loss": 65.70160522460938,
|
| 313 |
+
"original_ce_loss": 4.0714,
|
| 314 |
+
"step": 380
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.20224937607363952,
|
| 318 |
+
"grad_norm": 21.18793296813965,
|
| 319 |
+
"learning_rate": 0.0004015647091557703,
|
| 320 |
+
"loss": 64.89686279296875,
|
| 321 |
+
"original_ce_loss": 4.0218,
|
| 322 |
+
"step": 390
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"epoch": 0.20743525751142514,
|
| 326 |
+
"grad_norm": 22.754533767700195,
|
| 327 |
+
"learning_rate": 0.0003998706171688167,
|
| 328 |
+
"loss": 63.323309326171874,
|
| 329 |
+
"original_ce_loss": 3.9241,
|
| 330 |
+
"step": 400
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.21262113894921078,
|
| 334 |
+
"grad_norm": 58.86216354370117,
|
| 335 |
+
"learning_rate": 0.0003981212458043321,
|
| 336 |
+
"loss": 62.618524169921876,
|
| 337 |
+
"original_ce_loss": 3.8807,
|
| 338 |
+
"step": 410
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.2178070203869964,
|
| 342 |
+
"grad_norm": 35.05159378051758,
|
| 343 |
+
"learning_rate": 0.0003963171111739758,
|
| 344 |
+
"loss": 62.59974365234375,
|
| 345 |
+
"original_ce_loss": 3.8801,
|
| 346 |
+
"step": 420
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.22299290182478204,
|
| 350 |
+
"grad_norm": 27.153701782226562,
|
| 351 |
+
"learning_rate": 0.000394458745546045,
|
| 352 |
+
"loss": 62.70927734375,
|
| 353 |
+
"original_ce_loss": 3.8874,
|
| 354 |
+
"step": 430
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.22817878326256766,
|
| 358 |
+
"grad_norm": 21.153301239013672,
|
| 359 |
+
"learning_rate": 0.0003925466971884414,
|
| 360 |
+
"loss": 61.38936157226563,
|
| 361 |
+
"original_ce_loss": 3.8054,
|
| 362 |
+
"step": 440
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 0.23336466470035327,
|
| 366 |
+
"grad_norm": 20.64156723022461,
|
| 367 |
+
"learning_rate": 0.00039058153020691774,
|
| 368 |
+
"loss": 61.003076171875,
|
| 369 |
+
"original_ce_loss": 3.7819,
|
| 370 |
+
"step": 450
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.23855054613813892,
|
| 374 |
+
"grad_norm": 21.125715255737305,
|
| 375 |
+
"learning_rate": 0.000388563824378651,
|
| 376 |
+
"loss": 59.52923583984375,
|
| 377 |
+
"original_ce_loss": 3.6904,
|
| 378 |
+
"step": 460
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"epoch": 0.24373642757592454,
|
| 382 |
+
"grad_norm": 26.057849884033203,
|
| 383 |
+
"learning_rate": 0.00038649417498119276,
|
| 384 |
+
"loss": 58.688787841796874,
|
| 385 |
+
"original_ce_loss": 3.6385,
|
| 386 |
+
"step": 470
|
| 387 |
+
},
|
| 388 |
+
{
|
| 389 |
+
"epoch": 0.24892230901371018,
|
| 390 |
+
"grad_norm": 16.029029846191406,
|
| 391 |
+
"learning_rate": 0.00038458758312086766,
|
| 392 |
+
"loss": 58.049749755859374,
|
| 393 |
+
"original_ce_loss": 3.5991,
|
| 394 |
+
"step": 480
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.2541081904514958,
|
| 398 |
+
"grad_norm": 21.53459358215332,
|
| 399 |
+
"learning_rate": 0.00038242093567210356,
|
| 400 |
+
"loss": 57.9917236328125,
|
| 401 |
+
"original_ce_loss": 3.5961,
|
| 402 |
+
"step": 490
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.25929407188928144,
|
| 406 |
+
"grad_norm": 20.32674217224121,
|
| 407 |
+
"learning_rate": 0.0003802041569717304,
|
| 408 |
+
"loss": 56.76655883789063,
|
| 409 |
+
"original_ce_loss": 3.52,
|
| 410 |
+
"step": 500
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.2644799533270671,
|
| 414 |
+
"grad_norm": 20.9243106842041,
|
| 415 |
+
"learning_rate": 0.00037793790102915776,
|
| 416 |
+
"loss": 56.13651733398437,
|
| 417 |
+
"original_ce_loss": 3.4812,
|
| 418 |
+
"step": 510
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.2696658347648527,
|
| 422 |
+
"grad_norm": 20.96311378479004,
|
| 423 |
+
"learning_rate": 0.00037562283645091373,
|
| 424 |
+
"loss": 56.05534057617187,
|
| 425 |
+
"original_ce_loss": 3.4767,
|
| 426 |
+
"step": 520
|
| 427 |
+
},
|
| 428 |
+
{
|
| 429 |
+
"epoch": 0.2748517162026383,
|
| 430 |
+
"grad_norm": 17.55072784423828,
|
| 431 |
+
"learning_rate": 0.0003732596462433884,
|
| 432 |
+
"loss": 55.78792114257813,
|
| 433 |
+
"original_ce_loss": 3.4605,
|
| 434 |
+
"step": 530
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 0.28003759764042396,
|
| 438 |
+
"grad_norm": 18.758989334106445,
|
| 439 |
+
"learning_rate": 0.00037084902761132834,
|
| 440 |
+
"loss": 54.729583740234375,
|
| 441 |
+
"original_ce_loss": 3.3949,
|
| 442 |
+
"step": 540
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.28522347907820955,
|
| 446 |
+
"grad_norm": NaN,
|
| 447 |
+
"learning_rate": 0.0003683916917521425,
|
| 448 |
+
"loss": 54.55269775390625,
|
| 449 |
+
"original_ce_loss": 3.3844,
|
| 450 |
+
"step": 550
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.2904093605159952,
|
| 454 |
+
"grad_norm": 14.696817398071289,
|
| 455 |
+
"learning_rate": 0.00036614074516450577,
|
| 456 |
+
"loss": 54.1279541015625,
|
| 457 |
+
"original_ce_loss": 3.3583,
|
| 458 |
+
"step": 560
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.29559524195378084,
|
| 462 |
+
"grad_norm": 17.74142837524414,
|
| 463 |
+
"learning_rate": 0.00036385308259816797,
|
| 464 |
+
"loss": 53.192022705078124,
|
| 465 |
+
"original_ce_loss": 3.3002,
|
| 466 |
+
"step": 570
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 0.3007811233915665,
|
| 470 |
+
"grad_norm": 17.893028259277344,
|
| 471 |
+
"learning_rate": 0.00036126883857104554,
|
| 472 |
+
"loss": 52.832635498046876,
|
| 473 |
+
"original_ce_loss": 3.2782,
|
| 474 |
+
"step": 580
|
| 475 |
+
},
|
| 476 |
+
{
|
| 477 |
+
"epoch": 0.3059670048293521,
|
| 478 |
+
"grad_norm": 19.497455596923828,
|
| 479 |
+
"learning_rate": 0.00035864070373053703,
|
| 480 |
+
"loss": 52.86646728515625,
|
| 481 |
+
"original_ce_loss": 3.2808,
|
| 482 |
+
"step": 590
|
| 483 |
+
},
|
| 484 |
+
{
|
| 485 |
+
"epoch": 0.3111528862671377,
|
| 486 |
+
"grad_norm": 18.8817081451416,
|
| 487 |
+
"learning_rate": 0.00035596945344719004,
|
| 488 |
+
"loss": 52.08366088867187,
|
| 489 |
+
"original_ce_loss": 3.2324,
|
| 490 |
+
"step": 600
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.31633876770492336,
|
| 494 |
+
"grad_norm": 16.73759651184082,
|
| 495 |
+
"learning_rate": 0.0003532558758117689,
|
| 496 |
+
"loss": 51.96683959960937,
|
| 497 |
+
"original_ce_loss": 3.2255,
|
| 498 |
+
"step": 610
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.32152464914270895,
|
| 502 |
+
"grad_norm": 18.244924545288086,
|
| 503 |
+
"learning_rate": 0.00035050077140274654,
|
| 504 |
+
"loss": 51.41688232421875,
|
| 505 |
+
"original_ce_loss": 3.1916,
|
| 506 |
+
"step": 620
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.3267105305804946,
|
| 510 |
+
"grad_norm": 15.633984565734863,
|
| 511 |
+
"learning_rate": 0.0003477049530501123,
|
| 512 |
+
"loss": 50.827294921875,
|
| 513 |
+
"original_ce_loss": 3.1552,
|
| 514 |
+
"step": 630
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.33189641201828024,
|
| 518 |
+
"grad_norm": 17.32355499267578,
|
| 519 |
+
"learning_rate": 0.0003448692455955654,
|
| 520 |
+
"loss": 50.5419189453125,
|
| 521 |
+
"original_ce_loss": 3.1378,
|
| 522 |
+
"step": 640
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 0.3370822934560659,
|
| 526 |
+
"grad_norm": 17.346134185791016,
|
| 527 |
+
"learning_rate": 0.000341994485649164,
|
| 528 |
+
"loss": 50.77027893066406,
|
| 529 |
+
"original_ce_loss": 3.1525,
|
| 530 |
+
"step": 650
|
| 531 |
+
},
|
| 532 |
+
{
|
| 533 |
+
"epoch": 0.3422681748938515,
|
| 534 |
+
"grad_norm": 14.735562324523926,
|
| 535 |
+
"learning_rate": 0.00033908152134250305,
|
| 536 |
+
"loss": 50.90714111328125,
|
| 537 |
+
"original_ce_loss": 3.1615,
|
| 538 |
+
"step": 660
|
| 539 |
+
},
|
| 540 |
+
{
|
| 541 |
+
"epoch": 0.3474540563316371,
|
| 542 |
+
"grad_norm": 15.107223510742188,
|
| 543 |
+
"learning_rate": 0.00033613121207849256,
|
| 544 |
+
"loss": 50.397952270507815,
|
| 545 |
+
"original_ce_loss": 3.13,
|
| 546 |
+
"step": 670
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"epoch": 0.35263993776942276,
|
| 550 |
+
"grad_norm": 17.90372657775879,
|
| 551 |
+
"learning_rate": 0.0003331444282778105,
|
| 552 |
+
"loss": 50.302572631835936,
|
| 553 |
+
"original_ce_loss": 3.1245,
|
| 554 |
+
"step": 680
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.35782581920720835,
|
| 558 |
+
"grad_norm": 16.361705780029297,
|
| 559 |
+
"learning_rate": 0.00033012205112210554,
|
| 560 |
+
"loss": 49.57408142089844,
|
| 561 |
+
"original_ce_loss": 3.0793,
|
| 562 |
+
"step": 690
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 0.363011700644994,
|
| 566 |
+
"grad_norm": 15.112071990966797,
|
| 567 |
+
"learning_rate": 0.00032706497229402433,
|
| 568 |
+
"loss": 49.512637329101565,
|
| 569 |
+
"original_ce_loss": 3.0759,
|
| 570 |
+
"step": 700
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.36819758208277964,
|
| 574 |
+
"grad_norm": 15.508453369140625,
|
| 575 |
+
"learning_rate": 0.00032397409371414114,
|
| 576 |
+
"loss": 48.785379028320314,
|
| 577 |
+
"original_ce_loss": 3.0308,
|
| 578 |
+
"step": 710
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.3733834635205653,
|
| 582 |
+
"grad_norm": 16.45890235900879,
|
| 583 |
+
"learning_rate": 0.0003208503272748677,
|
| 584 |
+
"loss": 48.78138732910156,
|
| 585 |
+
"original_ce_loss": 3.0309,
|
| 586 |
+
"step": 720
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"epoch": 0.3785693449583509,
|
| 590 |
+
"grad_norm": 15.485218048095703,
|
| 591 |
+
"learning_rate": 0.0003176945945714194,
|
| 592 |
+
"loss": 48.64510803222656,
|
| 593 |
+
"original_ce_loss": 3.0227,
|
| 594 |
+
"step": 730
|
| 595 |
+
},
|
| 596 |
+
{
|
| 597 |
+
"epoch": 0.3837552263961365,
|
| 598 |
+
"grad_norm": 15.203038215637207,
|
| 599 |
+
"learning_rate": 0.00031450782662992055,
|
| 600 |
+
"loss": 48.32204895019531,
|
| 601 |
+
"original_ce_loss": 3.0028,
|
| 602 |
+
"step": 740
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"epoch": 0.38894110783392216,
|
| 606 |
+
"grad_norm": 14.665837287902832,
|
| 607 |
+
"learning_rate": 0.00031129096363272617,
|
| 608 |
+
"loss": 48.31576843261719,
|
| 609 |
+
"original_ce_loss": 3.0028,
|
| 610 |
+
"step": 750
|
| 611 |
+
},
|
| 612 |
+
{
|
| 613 |
+
"epoch": 0.39412698927170775,
|
| 614 |
+
"grad_norm": 14.794254302978516,
|
| 615 |
+
"learning_rate": 0.0003080449546410435,
|
| 616 |
+
"loss": 47.79297790527344,
|
| 617 |
+
"original_ce_loss": 2.9704,
|
| 618 |
+
"step": 760
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 0.3993128707094934,
|
| 622 |
+
"grad_norm": 28.44912338256836,
|
| 623 |
+
"learning_rate": 0.0003047707573149325,
|
| 624 |
+
"loss": 48.0023193359375,
|
| 625 |
+
"original_ce_loss": 2.9838,
|
| 626 |
+
"step": 770
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.40449875214727904,
|
| 630 |
+
"grad_norm": 18.549545288085938,
|
| 631 |
+
"learning_rate": 0.0003014693376307713,
|
| 632 |
+
"loss": 47.87590026855469,
|
| 633 |
+
"original_ce_loss": 2.9762,
|
| 634 |
+
"step": 780
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 0.4096846335850647,
|
| 638 |
+
"grad_norm": 13.932229042053223,
|
| 639 |
+
"learning_rate": 0.0002981416695962664,
|
| 640 |
+
"loss": 47.00486755371094,
|
| 641 |
+
"original_ce_loss": 2.9221,
|
| 642 |
+
"step": 790
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.4148705150228503,
|
| 646 |
+
"grad_norm": 15.301485061645508,
|
| 647 |
+
"learning_rate": 0.0002947887349630945,
|
| 648 |
+
"loss": 46.82394104003906,
|
| 649 |
+
"original_ce_loss": 2.9111,
|
| 650 |
+
"step": 800
|
| 651 |
+
}
|
| 652 |
+
],
|
| 653 |
+
"logging_steps": 10,
|
| 654 |
+
"max_steps": 1929,
|
| 655 |
+
"num_input_tokens_seen": 0,
|
| 656 |
+
"num_train_epochs": 1,
|
| 657 |
+
"save_steps": 200,
|
| 658 |
+
"stateful_callbacks": {
|
| 659 |
+
"TrainerControl": {
|
| 660 |
+
"args": {
|
| 661 |
+
"should_epoch_stop": false,
|
| 662 |
+
"should_evaluate": false,
|
| 663 |
+
"should_log": false,
|
| 664 |
+
"should_save": true,
|
| 665 |
+
"should_training_stop": false
|
| 666 |
+
},
|
| 667 |
+
"attributes": {}
|
| 668 |
+
}
|
| 669 |
+
},
|
| 670 |
+
"total_flos": 8479331323625472.0,
|
| 671 |
+
"train_batch_size": 16,
|
| 672 |
+
"trial_name": null,
|
| 673 |
+
"trial_params": null
|
| 674 |
+
}
|