FastPlus Backup checkpoint-800 at global step 800
Browse files- checkpoint-800/config.json +19 -0
- checkpoint-800/generation_config.json +6 -0
- checkpoint-800/model.safetensors +3 -0
- checkpoint-800/optimizer.pt +3 -0
- checkpoint-800/rng_state.pth +3 -0
- checkpoint-800/scaler.pt +3 -0
- checkpoint-800/scheduler.pt +3 -0
- checkpoint-800/trainer_state.json +674 -0
- checkpoint-800/training_args.bin +3 -0
checkpoint-800/config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"FastPlusForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"dtype": "float32",
|
| 6 |
+
"hidden_size": 768,
|
| 7 |
+
"initializer_range": 0.02,
|
| 8 |
+
"intermediate_size": 2304,
|
| 9 |
+
"kd_alpha": 0.4,
|
| 10 |
+
"kd_temperature": 2.5,
|
| 11 |
+
"max_position_embeddings": 512,
|
| 12 |
+
"model_type": "fastplus",
|
| 13 |
+
"num_attention_heads": 12,
|
| 14 |
+
"num_hidden_layers": 12,
|
| 15 |
+
"tie_word_embeddings": true,
|
| 16 |
+
"transformers_version": "5.12.0",
|
| 17 |
+
"use_cache": false,
|
| 18 |
+
"vocab_size": 31968
|
| 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": "5.12.0"
|
| 6 |
+
}
|
checkpoint-800/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6999825eba18d4941fe1585517635eaba727c4ac83f4f15c1809089a6178ffd1
|
| 3 |
+
size 395782848
|
checkpoint-800/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8a73350383781c5128157f682df11ec474f0a917d62c497898b047b5cc0df31b
|
| 3 |
+
size 444577931
|
checkpoint-800/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:61c19bab1174704a4a4441475683bf1270277af15d2e2c95e964789128e482c4
|
| 3 |
+
size 14645
|
checkpoint-800/scaler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2d189da8d02e62f8f1abe6f45fc36d5cf8e743384bca5066c612ef9b70473251
|
| 3 |
+
size 1383
|
checkpoint-800/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:210f5134da866230a019c191ef6aa90d387ba31993f11bae1ee0e4bc9c9acce8
|
| 3 |
+
size 6009
|
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": 0.02666666666666667,
|
| 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": 0.0003333333333333333,
|
| 14 |
+
"grad_norm": 243.78553771972656,
|
| 15 |
+
"learning_rate": 5.95e-06,
|
| 16 |
+
"loss": 166.0732177734375,
|
| 17 |
+
"original_ce_loss": 10.3789,
|
| 18 |
+
"step": 10
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"epoch": 0.0006666666666666666,
|
| 22 |
+
"grad_norm": 239.2476348876953,
|
| 23 |
+
"learning_rate": 1.445e-05,
|
| 24 |
+
"loss": 165.82977294921875,
|
| 25 |
+
"original_ce_loss": 10.3637,
|
| 26 |
+
"step": 20
|
| 27 |
+
},
|
| 28 |
+
{
|
| 29 |
+
"epoch": 0.001,
|
| 30 |
+
"grad_norm": 244.72525024414062,
|
| 31 |
+
"learning_rate": 2.295e-05,
|
| 32 |
+
"loss": 165.18355712890624,
|
| 33 |
+
"original_ce_loss": 10.3233,
|
| 34 |
+
"step": 30
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 0.0013333333333333333,
|
| 38 |
+
"grad_norm": 270.7417297363281,
|
| 39 |
+
"learning_rate": 3.145e-05,
|
| 40 |
+
"loss": 163.79927978515624,
|
| 41 |
+
"original_ce_loss": 10.2368,
|
| 42 |
+
"step": 40
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"epoch": 0.0016666666666666668,
|
| 46 |
+
"grad_norm": 326.618408203125,
|
| 47 |
+
"learning_rate": 3.9949999999999995e-05,
|
| 48 |
+
"loss": 160.3599365234375,
|
| 49 |
+
"original_ce_loss": 10.0218,
|
| 50 |
+
"step": 50
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"epoch": 0.002,
|
| 54 |
+
"grad_norm": 166.10595703125,
|
| 55 |
+
"learning_rate": 4.845e-05,
|
| 56 |
+
"loss": 151.0793212890625,
|
| 57 |
+
"original_ce_loss": 9.4418,
|
| 58 |
+
"step": 60
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.0023333333333333335,
|
| 62 |
+
"grad_norm": 54.51128387451172,
|
| 63 |
+
"learning_rate": 5.695e-05,
|
| 64 |
+
"loss": 142.18406982421874,
|
| 65 |
+
"original_ce_loss": 8.8858,
|
| 66 |
+
"step": 70
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.0026666666666666666,
|
| 70 |
+
"grad_norm": 45.69681930541992,
|
| 71 |
+
"learning_rate": 6.544999999999999e-05,
|
| 72 |
+
"loss": 135.513818359375,
|
| 73 |
+
"original_ce_loss": 8.4689,
|
| 74 |
+
"step": 80
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.003,
|
| 78 |
+
"grad_norm": 44.38578414916992,
|
| 79 |
+
"learning_rate": 7.395e-05,
|
| 80 |
+
"loss": 129.2341796875,
|
| 81 |
+
"original_ce_loss": 8.0765,
|
| 82 |
+
"step": 90
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"epoch": 0.0033333333333333335,
|
| 86 |
+
"grad_norm": 36.00300979614258,
|
| 87 |
+
"learning_rate": 8.245e-05,
|
| 88 |
+
"loss": 123.0133544921875,
|
| 89 |
+
"original_ce_loss": 7.6877,
|
| 90 |
+
"step": 100
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.0036666666666666666,
|
| 94 |
+
"grad_norm": 32.789207458496094,
|
| 95 |
+
"learning_rate": 9.094999999999999e-05,
|
| 96 |
+
"loss": 117.155908203125,
|
| 97 |
+
"original_ce_loss": 7.3216,
|
| 98 |
+
"step": 110
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"epoch": 0.004,
|
| 102 |
+
"grad_norm": 26.692514419555664,
|
| 103 |
+
"learning_rate": 9.86e-05,
|
| 104 |
+
"loss": 112.15670166015624,
|
| 105 |
+
"original_ce_loss": 7.0091,
|
| 106 |
+
"step": 120
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"epoch": 0.004333333333333333,
|
| 110 |
+
"grad_norm": 20.985740661621094,
|
| 111 |
+
"learning_rate": 0.0001071,
|
| 112 |
+
"loss": 108.44638671875,
|
| 113 |
+
"original_ce_loss": 6.7772,
|
| 114 |
+
"step": 130
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.004666666666666667,
|
| 118 |
+
"grad_norm": NaN,
|
| 119 |
+
"learning_rate": 0.00011475,
|
| 120 |
+
"loss": 105.64476318359375,
|
| 121 |
+
"original_ce_loss": 6.6021,
|
| 122 |
+
"step": 140
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.005,
|
| 126 |
+
"grad_norm": 47.759765625,
|
| 127 |
+
"learning_rate": 0.000119,
|
| 128 |
+
"loss": 104.857763671875,
|
| 129 |
+
"original_ce_loss": 6.5529,
|
| 130 |
+
"step": 150
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 0.005333333333333333,
|
| 134 |
+
"grad_norm": 164.30889892578125,
|
| 135 |
+
"learning_rate": 0.00012749999999999998,
|
| 136 |
+
"loss": 103.3579833984375,
|
| 137 |
+
"original_ce_loss": 6.4592,
|
| 138 |
+
"step": 160
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 0.005666666666666667,
|
| 142 |
+
"grad_norm": 11.574602127075195,
|
| 143 |
+
"learning_rate": 0.000136,
|
| 144 |
+
"loss": 103.28526611328125,
|
| 145 |
+
"original_ce_loss": 6.4547,
|
| 146 |
+
"step": 170
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.006,
|
| 150 |
+
"grad_norm": 13.660222053527832,
|
| 151 |
+
"learning_rate": 0.00014450000000000002,
|
| 152 |
+
"loss": 103.16549072265624,
|
| 153 |
+
"original_ce_loss": 6.4472,
|
| 154 |
+
"step": 180
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"epoch": 0.006333333333333333,
|
| 158 |
+
"grad_norm": 15.656424522399902,
|
| 159 |
+
"learning_rate": 0.00015299999999999998,
|
| 160 |
+
"loss": 102.62147216796875,
|
| 161 |
+
"original_ce_loss": 6.4132,
|
| 162 |
+
"step": 190
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"epoch": 0.006666666666666667,
|
| 166 |
+
"grad_norm": 18.646780014038086,
|
| 167 |
+
"learning_rate": 0.0001615,
|
| 168 |
+
"loss": 101.59368896484375,
|
| 169 |
+
"original_ce_loss": 6.3489,
|
| 170 |
+
"step": 200
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.007,
|
| 174 |
+
"grad_norm": 39.721866607666016,
|
| 175 |
+
"learning_rate": 0.00017,
|
| 176 |
+
"loss": 100.74405517578126,
|
| 177 |
+
"original_ce_loss": 6.2958,
|
| 178 |
+
"step": 210
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.007333333333333333,
|
| 182 |
+
"grad_norm": 16.912254333496094,
|
| 183 |
+
"learning_rate": 0.00017849999999999997,
|
| 184 |
+
"loss": 99.42055053710938,
|
| 185 |
+
"original_ce_loss": 6.2131,
|
| 186 |
+
"step": 220
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"epoch": 0.007666666666666666,
|
| 190 |
+
"grad_norm": 17.74278450012207,
|
| 191 |
+
"learning_rate": 0.000187,
|
| 192 |
+
"loss": 97.92646484375,
|
| 193 |
+
"original_ce_loss": 6.1197,
|
| 194 |
+
"step": 230
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"epoch": 0.008,
|
| 198 |
+
"grad_norm": 18.32769775390625,
|
| 199 |
+
"learning_rate": 0.0001955,
|
| 200 |
+
"loss": 96.8761962890625,
|
| 201 |
+
"original_ce_loss": 6.0541,
|
| 202 |
+
"step": 240
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"epoch": 0.008333333333333333,
|
| 206 |
+
"grad_norm": 16.8792781829834,
|
| 207 |
+
"learning_rate": 0.00020399999999999997,
|
| 208 |
+
"loss": 95.25513305664063,
|
| 209 |
+
"original_ce_loss": 5.9528,
|
| 210 |
+
"step": 250
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.008666666666666666,
|
| 214 |
+
"grad_norm": 12.205955505371094,
|
| 215 |
+
"learning_rate": 0.0002125,
|
| 216 |
+
"loss": 93.572900390625,
|
| 217 |
+
"original_ce_loss": 5.8476,
|
| 218 |
+
"step": 260
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"epoch": 0.009,
|
| 222 |
+
"grad_norm": 13.931303024291992,
|
| 223 |
+
"learning_rate": 0.000221,
|
| 224 |
+
"loss": 91.66510009765625,
|
| 225 |
+
"original_ce_loss": 5.7284,
|
| 226 |
+
"step": 270
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.009333333333333334,
|
| 230 |
+
"grad_norm": 26.242258071899414,
|
| 231 |
+
"learning_rate": 0.0002295,
|
| 232 |
+
"loss": 90.45693969726562,
|
| 233 |
+
"original_ce_loss": 5.6529,
|
| 234 |
+
"step": 280
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.009666666666666667,
|
| 238 |
+
"grad_norm": 16.811609268188477,
|
| 239 |
+
"learning_rate": 0.000238,
|
| 240 |
+
"loss": 91.2010009765625,
|
| 241 |
+
"original_ce_loss": 5.6994,
|
| 242 |
+
"step": 290
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"epoch": 0.01,
|
| 246 |
+
"grad_norm": 25.731582641601562,
|
| 247 |
+
"learning_rate": 0.00024565,
|
| 248 |
+
"loss": 97.92516479492187,
|
| 249 |
+
"original_ce_loss": 6.1197,
|
| 250 |
+
"step": 300
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.010333333333333333,
|
| 254 |
+
"grad_norm": 36.38114547729492,
|
| 255 |
+
"learning_rate": 0.00025414999999999997,
|
| 256 |
+
"loss": 98.45838623046875,
|
| 257 |
+
"original_ce_loss": 6.153,
|
| 258 |
+
"step": 310
|
| 259 |
+
},
|
| 260 |
+
{
|
| 261 |
+
"epoch": 0.010666666666666666,
|
| 262 |
+
"grad_norm": NaN,
|
| 263 |
+
"learning_rate": 0.00026179999999999997,
|
| 264 |
+
"loss": 95.69706420898437,
|
| 265 |
+
"original_ce_loss": 5.9804,
|
| 266 |
+
"step": 320
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 0.011,
|
| 270 |
+
"grad_norm": 51.799739837646484,
|
| 271 |
+
"learning_rate": 0.00026944999999999996,
|
| 272 |
+
"loss": 91.98704223632812,
|
| 273 |
+
"original_ce_loss": 5.7485,
|
| 274 |
+
"step": 330
|
| 275 |
+
},
|
| 276 |
+
{
|
| 277 |
+
"epoch": 0.011333333333333334,
|
| 278 |
+
"grad_norm": 22.107471466064453,
|
| 279 |
+
"learning_rate": 0.00027795,
|
| 280 |
+
"loss": 85.44754028320312,
|
| 281 |
+
"original_ce_loss": 5.3398,
|
| 282 |
+
"step": 340
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.011666666666666667,
|
| 286 |
+
"grad_norm": 30.49631118774414,
|
| 287 |
+
"learning_rate": 0.00028645,
|
| 288 |
+
"loss": 82.21802978515625,
|
| 289 |
+
"original_ce_loss": 5.138,
|
| 290 |
+
"step": 350
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.012,
|
| 294 |
+
"grad_norm": 46.481483459472656,
|
| 295 |
+
"learning_rate": 0.00029495,
|
| 296 |
+
"loss": 81.45587768554688,
|
| 297 |
+
"original_ce_loss": 5.0904,
|
| 298 |
+
"step": 360
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 0.012333333333333333,
|
| 302 |
+
"grad_norm": 35.435367584228516,
|
| 303 |
+
"learning_rate": 0.00030345,
|
| 304 |
+
"loss": 80.58915405273437,
|
| 305 |
+
"original_ce_loss": 5.0362,
|
| 306 |
+
"step": 370
|
| 307 |
+
},
|
| 308 |
+
{
|
| 309 |
+
"epoch": 0.012666666666666666,
|
| 310 |
+
"grad_norm": 23.500009536743164,
|
| 311 |
+
"learning_rate": 0.00031109999999999997,
|
| 312 |
+
"loss": 79.21079711914062,
|
| 313 |
+
"original_ce_loss": 4.95,
|
| 314 |
+
"step": 380
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.013,
|
| 318 |
+
"grad_norm": 36.313453674316406,
|
| 319 |
+
"learning_rate": 0.00031959999999999996,
|
| 320 |
+
"loss": 77.30345458984375,
|
| 321 |
+
"original_ce_loss": 4.8308,
|
| 322 |
+
"step": 390
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"epoch": 0.013333333333333334,
|
| 326 |
+
"grad_norm": 22.148427963256836,
|
| 327 |
+
"learning_rate": 0.0003281,
|
| 328 |
+
"loss": 75.77711181640625,
|
| 329 |
+
"original_ce_loss": 4.7354,
|
| 330 |
+
"step": 400
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.013666666666666667,
|
| 334 |
+
"grad_norm": 25.187255859375,
|
| 335 |
+
"learning_rate": 0.0003366,
|
| 336 |
+
"loss": 73.7842041015625,
|
| 337 |
+
"original_ce_loss": 4.6109,
|
| 338 |
+
"step": 410
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.014,
|
| 342 |
+
"grad_norm": 36.26912307739258,
|
| 343 |
+
"learning_rate": 0.0003451,
|
| 344 |
+
"loss": 71.966650390625,
|
| 345 |
+
"original_ce_loss": 4.4973,
|
| 346 |
+
"step": 420
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.014333333333333333,
|
| 350 |
+
"grad_norm": 43.76127624511719,
|
| 351 |
+
"learning_rate": 0.0003536,
|
| 352 |
+
"loss": 70.83294677734375,
|
| 353 |
+
"original_ce_loss": 4.4264,
|
| 354 |
+
"step": 430
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.014666666666666666,
|
| 358 |
+
"grad_norm": 26.379560470581055,
|
| 359 |
+
"learning_rate": 0.00036209999999999997,
|
| 360 |
+
"loss": 69.33712158203124,
|
| 361 |
+
"original_ce_loss": 4.333,
|
| 362 |
+
"step": 440
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 0.015,
|
| 366 |
+
"grad_norm": 34.02827453613281,
|
| 367 |
+
"learning_rate": 0.00037059999999999996,
|
| 368 |
+
"loss": 67.307373046875,
|
| 369 |
+
"original_ce_loss": 4.2061,
|
| 370 |
+
"step": 450
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.015333333333333332,
|
| 374 |
+
"grad_norm": 86.85484313964844,
|
| 375 |
+
"learning_rate": 0.0003791,
|
| 376 |
+
"loss": 65.93311767578125,
|
| 377 |
+
"original_ce_loss": 4.1202,
|
| 378 |
+
"step": 460
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"epoch": 0.015666666666666666,
|
| 382 |
+
"grad_norm": 39.64079666137695,
|
| 383 |
+
"learning_rate": 0.0003876,
|
| 384 |
+
"loss": 64.83297729492188,
|
| 385 |
+
"original_ce_loss": 4.0515,
|
| 386 |
+
"step": 470
|
| 387 |
+
},
|
| 388 |
+
{
|
| 389 |
+
"epoch": 0.016,
|
| 390 |
+
"grad_norm": 27.540979385375977,
|
| 391 |
+
"learning_rate": 0.0003961,
|
| 392 |
+
"loss": 63.43541259765625,
|
| 393 |
+
"original_ce_loss": 3.9641,
|
| 394 |
+
"step": 480
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.01633333333333333,
|
| 398 |
+
"grad_norm": 66.87540435791016,
|
| 399 |
+
"learning_rate": 0.00040459999999999997,
|
| 400 |
+
"loss": 62.5385986328125,
|
| 401 |
+
"original_ce_loss": 3.9081,
|
| 402 |
+
"step": 490
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.016666666666666666,
|
| 406 |
+
"grad_norm": 43.895957946777344,
|
| 407 |
+
"learning_rate": 0.00041309999999999996,
|
| 408 |
+
"loss": 61.425079345703125,
|
| 409 |
+
"original_ce_loss": 3.8385,
|
| 410 |
+
"step": 500
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.017,
|
| 414 |
+
"grad_norm": 29.485631942749023,
|
| 415 |
+
"learning_rate": 0.00042159999999999995,
|
| 416 |
+
"loss": 60.54182739257813,
|
| 417 |
+
"original_ce_loss": 3.7833,
|
| 418 |
+
"step": 510
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.017333333333333333,
|
| 422 |
+
"grad_norm": 39.4123649597168,
|
| 423 |
+
"learning_rate": 0.0004249999566202406,
|
| 424 |
+
"loss": 59.02940673828125,
|
| 425 |
+
"original_ce_loss": 3.6887,
|
| 426 |
+
"step": 520
|
| 427 |
+
},
|
| 428 |
+
{
|
| 429 |
+
"epoch": 0.017666666666666667,
|
| 430 |
+
"grad_norm": 29.954763412475586,
|
| 431 |
+
"learning_rate": 0.0004249996915217756,
|
| 432 |
+
"loss": 58.09852294921875,
|
| 433 |
+
"original_ce_loss": 3.6306,
|
| 434 |
+
"step": 530
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 0.018,
|
| 438 |
+
"grad_norm": 23.342824935913086,
|
| 439 |
+
"learning_rate": 0.00042499918542501206,
|
| 440 |
+
"loss": 57.1560546875,
|
| 441 |
+
"original_ce_loss": 3.5717,
|
| 442 |
+
"step": 540
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.018333333333333333,
|
| 446 |
+
"grad_norm": 28.458494186401367,
|
| 447 |
+
"learning_rate": 0.00042499843833052406,
|
| 448 |
+
"loss": 56.211798095703124,
|
| 449 |
+
"original_ce_loss": 3.5127,
|
| 450 |
+
"step": 550
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.018666666666666668,
|
| 454 |
+
"grad_norm": 22.47995948791504,
|
| 455 |
+
"learning_rate": 0.0004249974502391588,
|
| 456 |
+
"loss": 55.34743041992188,
|
| 457 |
+
"original_ce_loss": 3.4586,
|
| 458 |
+
"step": 560
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.019,
|
| 462 |
+
"grad_norm": 25.036964416503906,
|
| 463 |
+
"learning_rate": 0.00042499622115203697,
|
| 464 |
+
"loss": 54.5650390625,
|
| 465 |
+
"original_ce_loss": 3.4097,
|
| 466 |
+
"step": 570
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 0.019333333333333334,
|
| 470 |
+
"grad_norm": 19.473806381225586,
|
| 471 |
+
"learning_rate": 0.0004249947510705524,
|
| 472 |
+
"loss": 54.05086059570313,
|
| 473 |
+
"original_ce_loss": 3.3776,
|
| 474 |
+
"step": 580
|
| 475 |
+
},
|
| 476 |
+
{
|
| 477 |
+
"epoch": 0.019666666666666666,
|
| 478 |
+
"grad_norm": 31.730932235717773,
|
| 479 |
+
"learning_rate": 0.0004249930399963724,
|
| 480 |
+
"loss": 53.475189208984375,
|
| 481 |
+
"original_ce_loss": 3.3416,
|
| 482 |
+
"step": 590
|
| 483 |
+
},
|
| 484 |
+
{
|
| 485 |
+
"epoch": 0.02,
|
| 486 |
+
"grad_norm": 161.93760681152344,
|
| 487 |
+
"learning_rate": 0.0004249910879314376,
|
| 488 |
+
"loss": 52.975347900390624,
|
| 489 |
+
"original_ce_loss": 3.3104,
|
| 490 |
+
"step": 600
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.02033333333333333,
|
| 494 |
+
"grad_norm": 32.141876220703125,
|
| 495 |
+
"learning_rate": 0.0004249888948779616,
|
| 496 |
+
"loss": 52.947119140625,
|
| 497 |
+
"original_ce_loss": 3.3086,
|
| 498 |
+
"step": 610
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.020666666666666667,
|
| 502 |
+
"grad_norm": 47.48851013183594,
|
| 503 |
+
"learning_rate": 0.0004249864608384318,
|
| 504 |
+
"loss": 53.417730712890624,
|
| 505 |
+
"original_ce_loss": 3.338,
|
| 506 |
+
"step": 620
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.021,
|
| 510 |
+
"grad_norm": 32.532012939453125,
|
| 511 |
+
"learning_rate": 0.0004249837858156086,
|
| 512 |
+
"loss": 55.0054931640625,
|
| 513 |
+
"original_ce_loss": 3.4373,
|
| 514 |
+
"step": 630
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.021333333333333333,
|
| 518 |
+
"grad_norm": 36.4773063659668,
|
| 519 |
+
"learning_rate": 0.0004249808698125257,
|
| 520 |
+
"loss": 53.7179443359375,
|
| 521 |
+
"original_ce_loss": 3.3568,
|
| 522 |
+
"step": 640
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 0.021666666666666667,
|
| 526 |
+
"grad_norm": 53.40055465698242,
|
| 527 |
+
"learning_rate": 0.00042497771283249035,
|
| 528 |
+
"loss": 53.1301513671875,
|
| 529 |
+
"original_ce_loss": 3.3201,
|
| 530 |
+
"step": 650
|
| 531 |
+
},
|
| 532 |
+
{
|
| 533 |
+
"epoch": 0.022,
|
| 534 |
+
"grad_norm": 24.55984878540039,
|
| 535 |
+
"learning_rate": 0.0004249743148790827,
|
| 536 |
+
"loss": 52.9806396484375,
|
| 537 |
+
"original_ce_loss": 3.3107,
|
| 538 |
+
"step": 660
|
| 539 |
+
},
|
| 540 |
+
{
|
| 541 |
+
"epoch": 0.022333333333333334,
|
| 542 |
+
"grad_norm": 32.192604064941406,
|
| 543 |
+
"learning_rate": 0.0004249706759561566,
|
| 544 |
+
"loss": 51.710516357421874,
|
| 545 |
+
"original_ce_loss": 3.2314,
|
| 546 |
+
"step": 670
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"epoch": 0.02266666666666667,
|
| 550 |
+
"grad_norm": 29.168657302856445,
|
| 551 |
+
"learning_rate": 0.00042496679606783886,
|
| 552 |
+
"loss": 50.8844482421875,
|
| 553 |
+
"original_ce_loss": 3.1797,
|
| 554 |
+
"step": 680
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.023,
|
| 558 |
+
"grad_norm": 27.498979568481445,
|
| 559 |
+
"learning_rate": 0.00042496267521852977,
|
| 560 |
+
"loss": 50.27825012207031,
|
| 561 |
+
"original_ce_loss": 3.1418,
|
| 562 |
+
"step": 690
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 0.023333333333333334,
|
| 566 |
+
"grad_norm": 28.848630905151367,
|
| 567 |
+
"learning_rate": 0.0004249583134129028,
|
| 568 |
+
"loss": 49.64307861328125,
|
| 569 |
+
"original_ce_loss": 3.1022,
|
| 570 |
+
"step": 700
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.023666666666666666,
|
| 574 |
+
"grad_norm": 15.883838653564453,
|
| 575 |
+
"learning_rate": 0.0004249537106559047,
|
| 576 |
+
"loss": 49.61387329101562,
|
| 577 |
+
"original_ce_loss": 3.1003,
|
| 578 |
+
"step": 710
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.024,
|
| 582 |
+
"grad_norm": 17.885517120361328,
|
| 583 |
+
"learning_rate": 0.00042494886695275565,
|
| 584 |
+
"loss": 48.89302062988281,
|
| 585 |
+
"original_ce_loss": 3.0553,
|
| 586 |
+
"step": 720
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"epoch": 0.024333333333333332,
|
| 590 |
+
"grad_norm": 12.268507957458496,
|
| 591 |
+
"learning_rate": 0.0004249437823089488,
|
| 592 |
+
"loss": 48.71318054199219,
|
| 593 |
+
"original_ce_loss": 3.044,
|
| 594 |
+
"step": 730
|
| 595 |
+
},
|
| 596 |
+
{
|
| 597 |
+
"epoch": 0.024666666666666667,
|
| 598 |
+
"grad_norm": 56.60163497924805,
|
| 599 |
+
"learning_rate": 0.00042493845673025066,
|
| 600 |
+
"loss": 48.262673950195314,
|
| 601 |
+
"original_ce_loss": 3.0159,
|
| 602 |
+
"step": 740
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"epoch": 0.025,
|
| 606 |
+
"grad_norm": 30.987590789794922,
|
| 607 |
+
"learning_rate": 0.0004249328902227012,
|
| 608 |
+
"loss": 48.04598693847656,
|
| 609 |
+
"original_ce_loss": 3.0024,
|
| 610 |
+
"step": 750
|
| 611 |
+
},
|
| 612 |
+
{
|
| 613 |
+
"epoch": 0.025333333333333333,
|
| 614 |
+
"grad_norm": 26.050254821777344,
|
| 615 |
+
"learning_rate": 0.0004249270827926134,
|
| 616 |
+
"loss": 48.35114440917969,
|
| 617 |
+
"original_ce_loss": 3.0214,
|
| 618 |
+
"step": 760
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 0.025666666666666667,
|
| 622 |
+
"grad_norm": 17.595762252807617,
|
| 623 |
+
"learning_rate": 0.00042492103444657345,
|
| 624 |
+
"loss": 49.43587951660156,
|
| 625 |
+
"original_ce_loss": 3.0892,
|
| 626 |
+
"step": 770
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.026,
|
| 630 |
+
"grad_norm": 39.68498611450195,
|
| 631 |
+
"learning_rate": 0.000424914745191441,
|
| 632 |
+
"loss": 49.452764892578124,
|
| 633 |
+
"original_ce_loss": 3.0903,
|
| 634 |
+
"step": 780
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 0.026333333333333334,
|
| 638 |
+
"grad_norm": 19.96104621887207,
|
| 639 |
+
"learning_rate": 0.0004249095403375722,
|
| 640 |
+
"loss": 49.38257751464844,
|
| 641 |
+
"original_ce_loss": 3.0859,
|
| 642 |
+
"step": 790
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.02666666666666667,
|
| 646 |
+
"grad_norm": 14.854105949401855,
|
| 647 |
+
"learning_rate": 0.0004249028174642311,
|
| 648 |
+
"loss": 49.160150146484376,
|
| 649 |
+
"original_ce_loss": 3.072,
|
| 650 |
+
"step": 800
|
| 651 |
+
}
|
| 652 |
+
],
|
| 653 |
+
"logging_steps": 10,
|
| 654 |
+
"max_steps": 30000,
|
| 655 |
+
"num_input_tokens_seen": 0,
|
| 656 |
+
"num_train_epochs": 9223372036854775807,
|
| 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": 5.89491672711168e+16,
|
| 671 |
+
"train_batch_size": 16,
|
| 672 |
+
"trial_name": null,
|
| 673 |
+
"trial_params": null
|
| 674 |
+
}
|
checkpoint-800/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1f47ed938d04d7d144caa52ca167da9d45a43ccd6963b4147cc5e86aa6191d96
|
| 3 |
+
size 5137
|