Upload batch 13/24: sst2/mBERT/attention/seed_42 ... sst2/mBERT/cls/seed_42
Browse files- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/trainer_state.json +1245 -0
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8738532110091743,
|
| 4 |
+
"eval_loss": 0.42198362946510315,
|
| 5 |
+
"eval_runtime": 2.7152,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 321.161,
|
| 8 |
+
"eval_steps_per_second": 3.683,
|
| 9 |
+
"total_flos": 6195653766242742.0,
|
| 10 |
+
"train_loss": 0.2329871662718165,
|
| 11 |
+
"train_runtime": 614.6462,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 328.721,
|
| 14 |
+
"train_steps_per_second": 6.853
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8738532110091743,
|
| 4 |
+
"eval_loss": 0.42198362946510315,
|
| 5 |
+
"eval_runtime": 2.7152,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 321.161,
|
| 8 |
+
"eval_steps_per_second": 3.683
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/run_command.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
/usr/local/bin/python /workspace/run_glue_MHA_gated.py --model_name_or_path bert-base-multilingual-cased --dataset_name tmnam20/ViGLUE --task_name sst2 --max_seq_length 128 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 25 --report_to none --num_train_epochs 3 --learning_rate 3e-05 --per_device_train_batch_size 48 --per_device_eval_batch_size 96 --gradient_accumulation_steps 1 --dataloader_num_workers 1 --seed 42 --output_dir result_ablation_mbert/sst2/mBERT/attention/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy attention --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195653766242742.0,
|
| 4 |
+
"train_loss": 0.2329871662718165,
|
| 5 |
+
"train_runtime": 614.6462,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 328.721,
|
| 8 |
+
"train_steps_per_second": 6.853
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_42/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 6.190553665161133,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6233,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 6.925804615020752,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5194,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 7.219901084899902,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4933,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 3.3414974212646484,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4425,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 5.254892826080322,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4529,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 11.317140579223633,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4673,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.350831985473633,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4333,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 5.936102867126465,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.435,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 4.3392229080200195,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4261,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 3.9682154655456543,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4132,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 3.9904191493988037,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.42,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 5.818790435791016,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4081,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 8.263792037963867,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3671,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.332169055938721,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3912,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 6.198218822479248,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3877,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 10.373798370361328,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3612,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 5.906569004058838,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3569,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 6.637474536895752,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3956,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 6.43071985244751,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3373,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 3.827481746673584,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3408,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 5.9641499519348145,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3333,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 5.058157920837402,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3666,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 7.681595325469971,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3118,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 7.121781349182129,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.315,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 8.08523941040039,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3496,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 8.6000337600708,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3281,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 4.43425989151001,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3559,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 5.389403820037842,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3216,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 7.843454837799072,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3129,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 7.1295552253723145,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3045,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 6.759848117828369,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3455,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 7.8054728507995605,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.2897,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 4.309993743896484,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3282,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.47447395324707,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3308,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 5.381199359893799,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.299,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.10048246383667,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3428,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 7.194526195526123,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2882,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 3.7840445041656494,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3129,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 5.296388149261475,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.3103,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 8.04012393951416,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3282,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 7.637207508087158,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2666,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.8317315578460693,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3205,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 6.972060203552246,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.291,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 7.21154260635376,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2882,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 7.036592960357666,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2887,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 4.199458599090576,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.3202,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 3.8784468173980713,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2946,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 8.992252349853516,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2804,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 7.883377552032471,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2568,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 7.557548522949219,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3299,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 8.649031639099121,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2936,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 5.636533260345459,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2733,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 5.311301231384277,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2535,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 4.173188209533691,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2923,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 5.092861175537109,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.3012,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 4.731616973876953,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2835,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8772935779816514,
|
| 406 |
+
"eval_loss": 0.3283161222934723,
|
| 407 |
+
"eval_runtime": 2.8133,
|
| 408 |
+
"eval_samples_per_second": 309.956,
|
| 409 |
+
"eval_steps_per_second": 3.555,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 4.966166973114014,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2496,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 4.677845001220703,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2567,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.538151264190674,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2447,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 7.1953444480896,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2182,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 7.372513294219971,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2163,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 4.767239570617676,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2059,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 4.543787956237793,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2218,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 6.764569282531738,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2137,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 7.984580039978027,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2052,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 4.718881130218506,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2193,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 4.635927200317383,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2218,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 6.2887749671936035,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2334,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 3.9413135051727295,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2168,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 2.905959129333496,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2157,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 6.1196980476379395,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2166,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 6.198342323303223,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2186,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 6.144821643829346,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2297,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 5.098264217376709,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2001,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 4.906731605529785,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.1915,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 4.632498264312744,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2109,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 6.075517654418945,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.219,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 5.030157566070557,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.1874,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 3.2884023189544678,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.1908,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 5.920406341552734,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2054,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 3.6458442211151123,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2028,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 5.1585540771484375,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2187,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.3051066398620605,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2098,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 6.22515869140625,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2235,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 8.557690620422363,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1803,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 3.5899055004119873,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.208,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.730844020843506,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2255,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 9.438525199890137,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2002,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 4.9582438468933105,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2112,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 11.508864402770996,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1812,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 4.604296684265137,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2321,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 5.378977298736572,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2254,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 5.7168073654174805,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1898,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 4.259668350219727,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2022,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 2.2321457862854004,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1736,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 3.9781229496002197,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.1954,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 6.1012864112854,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2012,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 6.998119831085205,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2288,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 7.157221794128418,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1677,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 6.2332048416137695,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2166,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 4.947112560272217,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2234,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.558706760406494,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.212,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 5.970888614654541,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.1963,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 5.78853178024292,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.2055,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 4.062685489654541,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1929,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 7.280600547790527,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2263,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.848437786102295,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.2016,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.420389175415039,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.191,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 5.929506301879883,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1815,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 7.437278747558594,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1783,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 5.994002342224121,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1944,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 2.336717367172241,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1847,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8646788990825688,
|
| 807 |
+
"eval_loss": 0.39354804158210754,
|
| 808 |
+
"eval_runtime": 2.7391,
|
| 809 |
+
"eval_samples_per_second": 318.351,
|
| 810 |
+
"eval_steps_per_second": 3.651,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 10.470025062561035,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1419,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 2.0128817558288574,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1518,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 4.478729724884033,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1597,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 4.27400016784668,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1397,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 1.771393895149231,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1375,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 3.486797571182251,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1489,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 6.954610347747803,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1476,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 3.9687387943267822,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1635,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 2.03997802734375,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.143,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 6.041903018951416,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1528,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 10.33615779876709,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1461,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 3.014862060546875,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1655,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 7.641251564025879,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1496,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 9.256719589233398,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1323,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 7.038164138793945,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1573,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 3.353912115097046,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.154,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 7.4979987144470215,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1215,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 6.864070892333984,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1467,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 5.322312355041504,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1654,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 7.702466011047363,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1304,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 8.480555534362793,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1376,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 4.206344127655029,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1575,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 6.281627655029297,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1453,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 8.912609100341797,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1533,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 3.313055992126465,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1494,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 6.231218338012695,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1274,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 6.147752285003662,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1272,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 5.486332416534424,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1068,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 8.742708206176758,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1285,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 4.387041091918945,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1615,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 6.3914289474487305,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1597,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 6.12612771987915,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1403,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 6.846413612365723,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.134,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 6.113450050354004,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.138,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 8.001541137695312,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1579,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 4.307462215423584,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1421,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 3.094508171081543,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1655,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 6.463779926300049,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1199,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 5.655273914337158,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1214,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 8.402252197265625,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1484,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 6.291233062744141,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1303,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 10.19725227355957,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1421,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 6.500277042388916,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1379,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 6.249205589294434,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1531,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 5.652390480041504,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.125,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 5.818241596221924,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1304,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 2.3665356636047363,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1453,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 3.836926221847534,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1357,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 9.464919090270996,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1288,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 5.046365261077881,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1221,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 3.713693141937256,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1218,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 6.709402084350586,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1264,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 5.188994407653809,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.123,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 5.253087043762207,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1519,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 7.782111167907715,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1223,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 5.651302814483643,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1424,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8738532110091743,
|
| 1208 |
+
"eval_loss": 0.42198362946510315,
|
| 1209 |
+
"eval_runtime": 2.7408,
|
| 1210 |
+
"eval_samples_per_second": 318.157,
|
| 1211 |
+
"eval_steps_per_second": 3.649,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195653766242742.0,
|
| 1218 |
+
"train_loss": 0.2329871662718165,
|
| 1219 |
+
"train_runtime": 614.6462,
|
| 1220 |
+
"train_samples_per_second": 328.721,
|
| 1221 |
+
"train_steps_per_second": 6.853
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6195653766242742.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8876146788990825,
|
| 4 |
+
"eval_loss": 0.37215080857276917,
|
| 5 |
+
"eval_runtime": 2.6232,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 332.421,
|
| 8 |
+
"eval_steps_per_second": 3.812,
|
| 9 |
+
"total_flos": 6195302711288262.0,
|
| 10 |
+
"train_loss": 0.2262992448503362,
|
| 11 |
+
"train_runtime": 614.8564,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 328.608,
|
| 14 |
+
"train_steps_per_second": 6.85
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8876146788990825,
|
| 4 |
+
"eval_loss": 0.37215080857276917,
|
| 5 |
+
"eval_runtime": 2.6232,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 332.421,
|
| 8 |
+
"eval_steps_per_second": 3.812
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/run_command.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
/usr/local/bin/python /workspace/run_glue_MHA_gated.py --model_name_or_path bert-base-multilingual-cased --dataset_name tmnam20/ViGLUE --task_name sst2 --max_seq_length 128 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 25 --report_to none --num_train_epochs 3 --learning_rate 3e-05 --per_device_train_batch_size 48 --per_device_eval_batch_size 96 --gradient_accumulation_steps 1 --dataloader_num_workers 1 --seed 43 --output_dir result_ablation_mbert/sst2/mBERT/attention/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy attention --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195302711288262.0,
|
| 4 |
+
"train_loss": 0.2262992448503362,
|
| 5 |
+
"train_runtime": 614.8564,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 328.608,
|
| 8 |
+
"train_steps_per_second": 6.85
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_43/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 10.542654991149902,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6093,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 9.98283576965332,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5208,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 9.943350791931152,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4818,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 7.55079460144043,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4978,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 9.282877922058105,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.511,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 12.591513633728027,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4884,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.470808982849121,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4336,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 6.542767524719238,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4065,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 4.82924747467041,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4235,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.465608596801758,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4086,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 7.445090293884277,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4061,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.214295864105225,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4055,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 6.417483329772949,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3994,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 4.549795150756836,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3773,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 9.889904975891113,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3606,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 6.260658264160156,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3905,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 6.75949239730835,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3739,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 4.298714637756348,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3711,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 7.4904375076293945,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3451,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 4.1577372550964355,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3391,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 5.753899097442627,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3305,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 5.21770715713501,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3092,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 4.58365535736084,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3221,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 5.746969223022461,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3072,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 4.481151103973389,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.345,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 3.4031097888946533,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3163,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 4.172615051269531,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3375,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.7140543460845947,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3261,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.4524738788604736,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.2996,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.079586029052734,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3411,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.373317241668701,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3245,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 5.607375144958496,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3189,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 5.149340629577637,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3204,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 3.4397289752960205,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.2784,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 4.670861721038818,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3415,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 5.538418292999268,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.316,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 8.973953247070312,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2966,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 5.755313396453857,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.267,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.531044960021973,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2727,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 6.303153038024902,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.277,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 5.841059684753418,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2957,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.7818610668182373,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.2889,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 3.824685573577881,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2656,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.192707538604736,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2995,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 6.990862846374512,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2666,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.588795185089111,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2953,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 7.11063289642334,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.3026,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 5.198852062225342,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2524,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 6.312851905822754,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.304,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 3.281928539276123,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3073,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 4.181487560272217,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2892,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 8.573134422302246,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2738,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 6.456146717071533,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2592,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 6.0996551513671875,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2493,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 3.991407632827759,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2734,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 4.094855308532715,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2566,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8646788990825688,
|
| 406 |
+
"eval_loss": 0.3782247006893158,
|
| 407 |
+
"eval_runtime": 2.7736,
|
| 408 |
+
"eval_samples_per_second": 314.397,
|
| 409 |
+
"eval_steps_per_second": 3.605,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 4.360231876373291,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2075,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 5.285398006439209,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2223,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 3.872520923614502,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2219,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 6.159001350402832,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2046,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 4.8034515380859375,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2286,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 7.053504467010498,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2244,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 3.2571218013763428,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1838,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 8.169988632202148,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2345,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 5.1919474601745605,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.1883,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 2.8392348289489746,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2328,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 8.967991828918457,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2172,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 6.2017822265625,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2413,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 5.301101207733154,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.215,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 5.351938724517822,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2051,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 4.041431427001953,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2171,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 5.350503921508789,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2037,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 8.642088890075684,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2131,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 8.335514068603516,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.1906,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 6.64874792098999,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2077,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 9.018375396728516,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.199,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 5.139105319976807,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.1982,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 5.301655292510986,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.235,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 5.445256233215332,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.1918,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 8.137327194213867,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2046,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 8.391393661499023,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2287,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 5.696448802947998,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.1785,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 4.959506511688232,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.174,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 5.475242614746094,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.189,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 5.73848819732666,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2176,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 5.012078762054443,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.1905,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 5.193539142608643,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2071,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 6.435009002685547,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1836,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.936493873596191,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.208,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 4.981485366821289,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1992,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 10.91718578338623,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2197,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 2.5237295627593994,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2074,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 7.2508134841918945,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.2332,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 4.596436977386475,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.1943,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 11.022151947021484,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1886,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 4.71660041809082,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2094,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 4.282094478607178,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.1794,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 8.960872650146484,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1953,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 6.301848411560059,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.2026,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 5.984243869781494,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2225,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 5.029277324676514,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2011,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 3.369697093963623,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1781,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 5.92061185836792,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.1943,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 6.218677520751953,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1939,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 1.640540599822998,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.167,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 6.22006368637085,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1771,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.115149974822998,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1957,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 4.984256744384766,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.1949,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 8.123810768127441,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1971,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 4.800151348114014,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1813,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 8.057966232299805,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1891,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 4.981904029846191,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1885,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8864678899082569,
|
| 807 |
+
"eval_loss": 0.3301945626735687,
|
| 808 |
+
"eval_runtime": 2.6507,
|
| 809 |
+
"eval_samples_per_second": 328.967,
|
| 810 |
+
"eval_steps_per_second": 3.773,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 4.930946350097656,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1388,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 4.403440475463867,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1321,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 2.765547513961792,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.121,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 7.422887325286865,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1021,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 5.689745903015137,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1449,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 5.132945537567139,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1387,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 9.208808898925781,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1075,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 9.15725040435791,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1566,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 9.371588706970215,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1191,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 4.342051029205322,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.142,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 10.187299728393555,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1461,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 9.254781723022461,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1516,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 6.7898030281066895,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1513,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 7.200855731964111,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1227,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 11.505096435546875,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1421,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 8.079204559326172,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1116,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 5.551544189453125,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1273,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 4.623546600341797,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1744,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 4.765681743621826,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1499,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 6.318047046661377,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1527,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 7.2246222496032715,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1174,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 6.583380222320557,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1183,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 3.616935968399048,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1252,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.645325660705566,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1276,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 6.770578384399414,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1338,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 6.095039367675781,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1314,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 5.728174209594727,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1318,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 6.517219543457031,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1396,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 10.088645935058594,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1349,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 6.337277412414551,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1559,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 43.75395202636719,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1486,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 9.157793998718262,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1235,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 2.1910250186920166,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1428,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 5.6312642097473145,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1339,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 2.3120150566101074,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1341,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 4.181162357330322,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1146,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 6.430379867553711,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1321,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 7.1609272956848145,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1118,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.545069694519043,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1427,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 6.437716484069824,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1432,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 1.733481764793396,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1445,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 4.317609786987305,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1109,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 7.470036506652832,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1261,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 5.958364486694336,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1449,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 7.367522239685059,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1139,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 12.50982666015625,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1315,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 10.28714656829834,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1404,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 6.238487243652344,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1135,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 5.367002964019775,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1227,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 7.356612682342529,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1339,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 5.527950763702393,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1231,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 5.052021026611328,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1417,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 4.3053789138793945,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.114,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 10.383979797363281,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1296,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 1.6954126358032227,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1097,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 10.368294715881348,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1335,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8876146788990825,
|
| 1208 |
+
"eval_loss": 0.37215080857276917,
|
| 1209 |
+
"eval_runtime": 2.5664,
|
| 1210 |
+
"eval_samples_per_second": 339.771,
|
| 1211 |
+
"eval_steps_per_second": 3.896,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195302711288262.0,
|
| 1218 |
+
"train_loss": 0.2262992448503362,
|
| 1219 |
+
"train_runtime": 614.8564,
|
| 1220 |
+
"train_samples_per_second": 328.608,
|
| 1221 |
+
"train_steps_per_second": 6.85
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6195302711288262.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8681192660550459,
|
| 4 |
+
"eval_loss": 0.4196762442588806,
|
| 5 |
+
"eval_runtime": 2.7038,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 322.513,
|
| 8 |
+
"eval_steps_per_second": 3.699,
|
| 9 |
+
"total_flos": 6187709082868488.0,
|
| 10 |
+
"train_loss": 0.22888659196927902,
|
| 11 |
+
"train_runtime": 614.6558,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 328.716,
|
| 14 |
+
"train_steps_per_second": 6.853
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8681192660550459,
|
| 4 |
+
"eval_loss": 0.4196762442588806,
|
| 5 |
+
"eval_runtime": 2.7038,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 322.513,
|
| 8 |
+
"eval_steps_per_second": 3.699
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/run_command.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
/usr/local/bin/python /workspace/run_glue_MHA_gated.py --model_name_or_path bert-base-multilingual-cased --dataset_name tmnam20/ViGLUE --task_name sst2 --max_seq_length 128 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 25 --report_to none --num_train_epochs 3 --learning_rate 3e-05 --per_device_train_batch_size 48 --per_device_eval_batch_size 96 --gradient_accumulation_steps 1 --dataloader_num_workers 1 --seed 44 --output_dir result_ablation_mbert/sst2/mBERT/attention/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy attention --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6187709082868488.0,
|
| 4 |
+
"train_loss": 0.22888659196927902,
|
| 5 |
+
"train_runtime": 614.6558,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 328.716,
|
| 8 |
+
"train_steps_per_second": 6.853
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/attention/seed_44/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 9.166430473327637,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.673,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 5.473770618438721,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.563,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 5.838556289672852,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4768,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 4.572579860687256,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4686,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 4.164381504058838,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4655,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 4.3329758644104,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4164,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 6.549654483795166,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4331,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 8.307929992675781,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4216,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 7.342564105987549,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.437,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 9.498054504394531,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4054,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 5.651376247406006,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.3664,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 6.842675685882568,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4038,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 7.150350570678711,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3597,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 7.095651149749756,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3766,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 6.428597450256348,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3735,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 5.938681602478027,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.363,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 5.656927108764648,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3564,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 4.999871253967285,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3594,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 10.75291633605957,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3804,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 6.314740180969238,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3449,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 4.632686614990234,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3607,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.203431606292725,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3843,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 5.610105514526367,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3276,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 4.6883344650268555,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3469,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.791014194488525,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3536,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 6.000133037567139,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3185,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 11.348575592041016,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3129,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 5.260585784912109,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3102,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.894665002822876,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3431,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 7.454975128173828,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3139,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.206374645233154,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3394,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 4.992528438568115,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.2891,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 8.107696533203125,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3446,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.0048956871032715,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3241,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 7.798484802246094,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3172,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.952391147613525,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3062,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 6.3346943855285645,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3322,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 3.890843629837036,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.2912,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 35.86193084716797,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.294,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 5.479144096374512,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2944,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 3.612403392791748,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2737,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 5.270524024963379,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3182,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 4.343075752258301,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2961,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.056978702545166,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3158,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 7.2179460525512695,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2678,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 4.711992263793945,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2848,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 5.709394454956055,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2778,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.6055846214294434,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2982,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 7.002932548522949,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2491,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 6.08748197555542,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.2822,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 4.948605537414551,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2672,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 7.480295181274414,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2983,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 3.2178902626037598,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2908,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 4.5441179275512695,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2448,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 3.6685950756073,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2544,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 3.038529634475708,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2849,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8669724770642202,
|
| 406 |
+
"eval_loss": 0.35786205530166626,
|
| 407 |
+
"eval_runtime": 2.7721,
|
| 408 |
+
"eval_samples_per_second": 314.56,
|
| 409 |
+
"eval_steps_per_second": 3.607,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 6.551161289215088,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2173,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 7.9860029220581055,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2164,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 7.9271745681762695,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2179,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 5.643150806427002,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2135,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 5.300719738006592,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2108,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 4.370731353759766,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2163,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 4.780885696411133,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1905,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 6.059784889221191,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2151,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 2.67641282081604,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2074,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 8.191490173339844,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2121,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 6.611438274383545,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2301,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 4.290998935699463,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2185,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 7.976767063140869,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2116,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 7.508364200592041,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.1961,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 2.9899652004241943,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2197,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 6.865063667297363,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2084,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 4.736561298370361,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.1945,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 5.538402080535889,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2402,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 5.8651041984558105,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2042,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 6.733060359954834,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2294,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 4.538701057434082,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2125,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.635342597961426,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.226,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.724644660949707,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.208,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 8.434202194213867,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2085,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 8.490874290466309,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2006,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 5.654787063598633,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.1952,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 8.111847877502441,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2211,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 4.258538722991943,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2157,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 5.8487043380737305,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2064,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 7.06566858291626,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2317,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.033429145812988,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2051,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 6.757253646850586,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1892,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 4.245517253875732,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.207,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 6.168206214904785,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.2096,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 7.274152755737305,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.1956,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.2115416526794434,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.1738,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 5.1700592041015625,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1855,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 9.513809204101562,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.1604,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 6.011676788330078,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2115,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 3.034172773361206,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2103,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 5.8933210372924805,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2079,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 3.766124963760376,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1812,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 10.100558280944824,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1828,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 7.019357681274414,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2185,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 6.346127986907959,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.1795,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 2.605994701385498,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1925,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 3.7986857891082764,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2255,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 3.616391658782959,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1847,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 7.280228614807129,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1888,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 4.260396480560303,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1936,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 5.613341331481934,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1908,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.046325206756592,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2132,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 5.225821495056152,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.183,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 5.279199123382568,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.2125,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 3.75113582611084,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1981,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 6.852717876434326,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.2,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8727064220183486,
|
| 807 |
+
"eval_loss": 0.3726809322834015,
|
| 808 |
+
"eval_runtime": 2.7303,
|
| 809 |
+
"eval_samples_per_second": 319.38,
|
| 810 |
+
"eval_steps_per_second": 3.663,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 5.271040916442871,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1618,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 2.8803491592407227,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1554,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 5.512966156005859,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1285,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 2.8406527042388916,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1507,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 7.727383136749268,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1362,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 1.186449408531189,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1268,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 4.949362277984619,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1214,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 6.7746357917785645,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1271,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 8.22618293762207,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1151,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 2.2353920936584473,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1439,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 2.509779691696167,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1431,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 7.4952778816223145,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1658,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 8.376375198364258,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1278,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 7.1031599044799805,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.0987,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 4.170283317565918,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1497,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 7.59684419631958,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1482,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 3.4206981658935547,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1386,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 7.247228622436523,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1416,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 5.494585990905762,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1431,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 5.741605281829834,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1579,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 3.8412177562713623,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1243,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 5.621340274810791,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1095,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 4.624086380004883,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1145,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 8.109993934631348,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1182,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 7.137166500091553,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1595,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 5.1795830726623535,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1248,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 6.151353359222412,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1594,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 7.759547233581543,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.124,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 2.2276389598846436,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1353,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 4.8944783210754395,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1153,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 3.277752637863159,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1217,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 3.847231388092041,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1429,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 8.18724250793457,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1399,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 6.286890506744385,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1322,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 7.081835746765137,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1285,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 7.356555461883545,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1495,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.841353893280029,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1446,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 2.8718764781951904,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1048,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 9.222522735595703,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1422,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 7.3638596534729,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1566,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 3.8745789527893066,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1294,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 6.600473880767822,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1224,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 3.235243797302246,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1409,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 4.209414958953857,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1295,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 6.498126029968262,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.145,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 8.389975547790527,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1121,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 8.208123207092285,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1556,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 8.633599281311035,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1522,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 4.9335832595825195,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1197,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 5.0071330070495605,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1286,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 7.061375141143799,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1453,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 6.917630672454834,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1214,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 7.20908784866333,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1375,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.880325794219971,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.137,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 7.890770435333252,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1233,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 4.7497124671936035,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1225,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8681192660550459,
|
| 1208 |
+
"eval_loss": 0.4196762442588806,
|
| 1209 |
+
"eval_runtime": 2.8738,
|
| 1210 |
+
"eval_samples_per_second": 303.431,
|
| 1211 |
+
"eval_steps_per_second": 3.48,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6187709082868488.0,
|
| 1218 |
+
"train_loss": 0.22888659196927902,
|
| 1219 |
+
"train_runtime": 614.6558,
|
| 1220 |
+
"train_samples_per_second": 328.716,
|
| 1221 |
+
"train_steps_per_second": 6.853
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6187709082868488.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8761467889908257,
|
| 4 |
+
"eval_loss": 0.4319080710411072,
|
| 5 |
+
"eval_runtime": 2.8294,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 308.195,
|
| 8 |
+
"eval_steps_per_second": 3.534,
|
| 9 |
+
"total_flos": 6195598393181184.0,
|
| 10 |
+
"train_loss": 0.23663618221015786,
|
| 11 |
+
"train_runtime": 603.5382,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 334.771,
|
| 14 |
+
"train_steps_per_second": 6.979
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8761467889908257,
|
| 4 |
+
"eval_loss": 0.4319080710411072,
|
| 5 |
+
"eval_runtime": 2.8294,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 308.195,
|
| 8 |
+
"eval_steps_per_second": 3.534
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/run_command.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
/usr/local/bin/python /workspace/run_glue_MHA_gated.py --model_name_or_path bert-base-multilingual-cased --dataset_name tmnam20/ViGLUE --task_name sst2 --max_seq_length 128 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 25 --report_to none --num_train_epochs 3 --learning_rate 3e-05 --per_device_train_batch_size 48 --per_device_eval_batch_size 96 --gradient_accumulation_steps 1 --dataloader_num_workers 1 --seed 42 --output_dir result_ablation_mbert/sst2/mBERT/cls/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy cls --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195598393181184.0,
|
| 4 |
+
"train_loss": 0.23663618221015786,
|
| 5 |
+
"train_runtime": 603.5382,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 334.771,
|
| 8 |
+
"train_steps_per_second": 6.979
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_42/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 3.4892871379852295,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6912,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 7.7788004875183105,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5639,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 9.356425285339355,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.5109,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 5.531455993652344,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4495,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 10.970504760742188,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4726,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 6.736781120300293,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4587,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 4.1514058113098145,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4564,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 7.306417942047119,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4393,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 3.756854295730591,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4196,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.12280797958374,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4352,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 7.415233612060547,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4245,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.382491111755371,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4267,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 6.075983047485352,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3615,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 4.707869052886963,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3888,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 6.1774702072143555,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3949,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 5.345212459564209,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3792,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 4.547158241271973,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3593,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 7.507021427154541,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3961,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 6.312687397003174,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3479,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 3.5032427310943604,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3631,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 5.260069847106934,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3472,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.14707088470459,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3639,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 7.177229404449463,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.307,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 8.337231636047363,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3307,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 8.042259216308594,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3606,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 6.068445205688477,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3343,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 3.6793155670166016,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3594,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 5.841829776763916,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3248,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 7.230782985687256,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3193,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.786648273468018,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3065,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 7.400082588195801,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3385,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 7.303816795349121,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.298,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 3.867819309234619,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3234,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.512134075164795,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3176,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 4.083412170410156,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.2976,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.65538215637207,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3344,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 6.460114002227783,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2938,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 3.923708915710449,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3188,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 5.781052112579346,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.326,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.427947998046875,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3255,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 9.613182067871094,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2802,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 6.398096084594727,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3298,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 6.773290634155273,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2931,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 7.815273284912109,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2969,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 7.773916244506836,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2934,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 3.440981864929199,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.3208,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 4.047163963317871,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2982,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 8.276764869689941,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2816,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 7.157919406890869,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2505,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 7.40455961227417,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3221,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 7.556107997894287,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2844,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 4.182618141174316,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2846,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 6.533888816833496,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2526,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 6.314499855041504,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2972,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.895680904388428,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.3083,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 5.753303527832031,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.278,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8623853211009175,
|
| 406 |
+
"eval_loss": 0.32925349473953247,
|
| 407 |
+
"eval_runtime": 3.2809,
|
| 408 |
+
"eval_samples_per_second": 265.778,
|
| 409 |
+
"eval_steps_per_second": 3.048,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 5.784731388092041,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2489,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 3.505681276321411,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2689,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.55968713760376,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2405,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 5.579002380371094,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2069,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 6.1483473777771,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2203,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 3.7870450019836426,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2116,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.129452705383301,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2265,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 6.672730922698975,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2208,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 6.742493629455566,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2195,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 6.029114246368408,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2359,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 4.291238307952881,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2323,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 7.634119987487793,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2322,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 3.733006000518799,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2295,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 4.223710536956787,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2423,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 6.308727264404297,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2246,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 8.297652244567871,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2227,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 6.339427471160889,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2222,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 4.139725685119629,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.1942,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 4.268884658813477,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.1782,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 5.899971961975098,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.234,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 5.680488109588623,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2162,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 3.350731611251831,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2019,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 3.556905508041382,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2025,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 7.302080154418945,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.205,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 7.075467109680176,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2018,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 7.821969509124756,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2161,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.186368465423584,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2291,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 6.79318380355835,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.218,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 9.108755111694336,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1886,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 4.722745418548584,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2093,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.428115367889404,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2178,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 5.834876537322998,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1914,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 4.404437065124512,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.1968,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 5.458004951477051,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1703,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 3.8215408325195312,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2219,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 4.616006374359131,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2383,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 4.510752201080322,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1899,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 8.315879821777344,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2183,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 4.295562267303467,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1819,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 3.9789421558380127,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.207,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 5.019663333892822,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2099,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 5.3795084953308105,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2221,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 5.306377410888672,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1856,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 4.707239151000977,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2127,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 4.833738327026367,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2234,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 7.220142364501953,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.2029,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 6.233399391174316,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2113,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 7.503040790557861,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.2022,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 3.6452744007110596,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.208,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 6.353243827819824,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2217,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 5.499780654907227,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.2024,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 4.213687896728516,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2004,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.887466907501221,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1819,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 8.314630508422852,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1743,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 3.1232941150665283,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1986,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 4.603719234466553,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1776,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8738532110091743,
|
| 807 |
+
"eval_loss": 0.4023585319519043,
|
| 808 |
+
"eval_runtime": 3.2911,
|
| 809 |
+
"eval_samples_per_second": 264.96,
|
| 810 |
+
"eval_steps_per_second": 3.039,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 7.295980930328369,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1441,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 2.6215474605560303,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1557,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 8.68816089630127,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1588,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 8.143495559692383,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1428,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 1.6307039260864258,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1511,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 4.316905498504639,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1449,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 5.853209972381592,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1362,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 5.798532485961914,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1689,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 6.802812099456787,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1406,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 9.124958992004395,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1719,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 3.147346258163452,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.142,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 1.5045090913772583,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1658,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 10.917342185974121,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1515,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 5.975603103637695,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1413,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 5.302867889404297,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.147,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 9.168828964233398,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1696,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 11.544958114624023,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1215,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 8.240396499633789,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1631,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 4.7147698402404785,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1656,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 3.747054100036621,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1266,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 9.234180450439453,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1403,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 3.2716071605682373,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1513,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 6.231365203857422,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1497,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 11.44874382019043,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1498,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 2.6117000579833984,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1607,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 5.92045259475708,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1297,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 7.554933547973633,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1296,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 3.0191569328308105,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1041,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 11.349215507507324,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1353,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 5.1535186767578125,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1537,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 6.665163040161133,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1654,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 6.151921272277832,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1357,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 9.49363899230957,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1295,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 6.906139373779297,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1479,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 6.709246635437012,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1546,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 3.2899770736694336,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1509,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 5.124023914337158,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1615,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 2.2731356620788574,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1287,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 2.539703845977783,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1272,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 6.450427532196045,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1484,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 6.159839630126953,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.12,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 7.354176044464111,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1377,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 7.894482135772705,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1463,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 7.725776195526123,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1472,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 7.50349235534668,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1302,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 5.284519195556641,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1459,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 3.4219322204589844,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.137,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 2.7474920749664307,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1351,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 7.327629566192627,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1297,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 6.449563503265381,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1082,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 4.632013320922852,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1275,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 8.410384178161621,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1329,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 6.041531562805176,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1297,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 6.039644241333008,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1418,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 8.809195518493652,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1224,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 8.934418678283691,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1363,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8761467889908257,
|
| 1208 |
+
"eval_loss": 0.4319080710411072,
|
| 1209 |
+
"eval_runtime": 2.7126,
|
| 1210 |
+
"eval_samples_per_second": 321.466,
|
| 1211 |
+
"eval_steps_per_second": 3.687,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195598393181184.0,
|
| 1218 |
+
"train_loss": 0.23663618221015786,
|
| 1219 |
+
"train_runtime": 603.5382,
|
| 1220 |
+
"train_samples_per_second": 334.771,
|
| 1221 |
+
"train_steps_per_second": 6.979
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6195598393181184.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|