Upload batch 14/24: sst2/mBERT/cls/seed_43 ... sst2/mBERT/gated_multi_branch/seed_43
Browse files- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/trainer_state.json +1245 -0
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8830275229357798,
|
| 4 |
+
"eval_loss": 0.4112645387649536,
|
| 5 |
+
"eval_runtime": 1.9029,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 458.254,
|
| 8 |
+
"eval_steps_per_second": 5.255,
|
| 9 |
+
"total_flos": 6195247341364224.0,
|
| 10 |
+
"train_loss": 0.23413483488593687,
|
| 11 |
+
"train_runtime": 601.3799,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 335.972,
|
| 14 |
+
"train_steps_per_second": 7.004
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_43/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8830275229357798,
|
| 4 |
+
"eval_loss": 0.4112645387649536,
|
| 5 |
+
"eval_runtime": 1.9029,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 458.254,
|
| 8 |
+
"eval_steps_per_second": 5.255
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/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/cls/seed_43 --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_43/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195247341364224.0,
|
| 4 |
+
"train_loss": 0.23413483488593687,
|
| 5 |
+
"train_runtime": 601.3799,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 335.972,
|
| 8 |
+
"train_steps_per_second": 7.004
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/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": 11.899322509765625,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6161,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 8.599114418029785,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5675,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 6.334699630737305,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4825,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 5.155947208404541,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.453,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 6.35404634475708,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4785,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 18.717365264892578,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4635,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 4.494163513183594,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4185,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 4.923633098602295,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.3977,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 4.342033386230469,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4455,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.251199245452881,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.3787,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 6.5207366943359375,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4183,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.768181324005127,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4177,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 5.529481410980225,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.4049,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.086929798126221,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3757,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 10.963387489318848,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3768,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 6.328497409820557,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.4057,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 8.082104682922363,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3812,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 5.179525375366211,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3748,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 6.2855706214904785,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3639,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 4.752938747406006,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3405,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 6.038226127624512,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3358,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 3.7510342597961426,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3293,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 5.231075286865234,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3269,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 5.223547458648682,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3295,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 4.623196601867676,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3527,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 3.5850796699523926,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3253,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 5.369431018829346,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3384,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.4506564140319824,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3274,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.685303211212158,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3152,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 6.3226165771484375,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3562,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.0952630043029785,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3391,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 4.458897590637207,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3176,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 6.301762580871582,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3198,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 3.6995792388916016,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.2873,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 6.185729026794434,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3454,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.4000091552734375,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.321,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 4.827035903930664,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.308,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 8.451615333557129,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3051,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.359392166137695,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2855,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 7.221246242523193,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2882,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 6.155463218688965,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2969,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.044367790222168,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.2945,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 2.831737756729126,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2703,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 4.712096214294434,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3199,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 6.547616004943848,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2884,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.526881217956543,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2937,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 6.0570502281188965,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.313,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.915360689163208,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2663,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 5.33869743347168,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.3204,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 3.361647129058838,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.314,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 3.440589427947998,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2915,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 9.250838279724121,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2927,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 6.006248474121094,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2641,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.5134968757629395,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2637,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.905615329742432,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2851,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 5.658141613006592,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2588,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8692660550458715,
|
| 406 |
+
"eval_loss": 0.37241309881210327,
|
| 407 |
+
"eval_runtime": 2.8951,
|
| 408 |
+
"eval_samples_per_second": 301.199,
|
| 409 |
+
"eval_steps_per_second": 3.454,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 5.2714385986328125,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.205,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 5.99589729309082,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2364,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.8661208152771,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2241,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 5.488052845001221,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.216,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 6.528749942779541,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2201,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 6.140528678894043,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2195,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.095637798309326,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2026,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 12.61596393585205,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2303,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 8.589193344116211,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.1944,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 4.71624755859375,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2502,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 4.837568759918213,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2211,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 3.9729413986206055,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2528,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 6.3291401863098145,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2252,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 4.722491264343262,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2124,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 2.203400135040283,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2148,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 4.980591773986816,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2165,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 6.768894672393799,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2318,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 6.314164638519287,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2028,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 6.2351579666137695,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2214,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 8.411465644836426,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.199,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 4.129683971405029,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2241,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 8.439371109008789,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2338,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 3.3914811611175537,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.225,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 8.710983276367188,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2219,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 7.651703834533691,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2362,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 5.300390243530273,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2076,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.031339645385742,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.193,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 4.650407314300537,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.1946,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 5.214559555053711,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2141,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 3.308401584625244,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.1984,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 8.41437816619873,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2209,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 5.631580829620361,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2025,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.83060884475708,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2217,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 7.915628910064697,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1995,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 10.310771942138672,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.228,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.9618258476257324,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2249,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 7.732817649841309,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.2447,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 6.119265079498291,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2022,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 9.79556655883789,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2022,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 3.7605247497558594,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2085,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 3.9718902111053467,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.182,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 8.677770614624023,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1992,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 3.8731489181518555,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1965,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 5.726711273193359,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2188,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 6.349492073059082,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2064,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.044182777404785,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1941,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 6.349427223205566,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2062,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 6.524653911590576,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1946,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 6.874235153198242,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1847,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 6.226429462432861,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1844,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 7.707487106323242,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.198,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 6.842658042907715,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2011,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.876976490020752,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.2,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 6.293125152587891,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1951,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 7.173577308654785,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1767,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 6.505352020263672,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1903,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8853211009174312,
|
| 807 |
+
"eval_loss": 0.3512602150440216,
|
| 808 |
+
"eval_runtime": 2.702,
|
| 809 |
+
"eval_samples_per_second": 322.727,
|
| 810 |
+
"eval_steps_per_second": 3.701,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 6.220231533050537,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.151,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 4.325716018676758,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1408,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 4.802373886108398,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1301,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 9.903068542480469,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1303,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 4.828006744384766,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1491,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 6.9948320388793945,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1423,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 8.504325866699219,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1195,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 8.715051651000977,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1687,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 7.790101528167725,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.145,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 5.997051239013672,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1412,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 6.252791881561279,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1439,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 4.620826244354248,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1572,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 8.212553024291992,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1533,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 6.339809894561768,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1388,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 9.398353576660156,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.168,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 5.217922687530518,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1196,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 1.509261131286621,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1339,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 5.639780044555664,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.184,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 4.317830562591553,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1482,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 10.58796501159668,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.148,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 6.989097595214844,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1335,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 4.497542858123779,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1234,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 6.298122406005859,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1369,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.804509162902832,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1331,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 7.867437839508057,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1391,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 9.102479934692383,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1347,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 6.831407070159912,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1363,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 5.384475231170654,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.153,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 11.92862319946289,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1326,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 5.533679008483887,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.177,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 8.05922794342041,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1725,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 8.144245147705078,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1178,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 2.8248932361602783,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1314,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 9.868926048278809,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1531,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 7.069158554077148,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1268,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 5.972555160522461,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.132,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 5.412944316864014,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1363,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 3.196437358856201,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.123,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.758090972900391,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1591,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 7.945244789123535,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1445,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 3.0366601943969727,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1493,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 7.185096263885498,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1239,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 8.388765335083008,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1442,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 6.044404029846191,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1401,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 6.591940879821777,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1385,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 10.55251407623291,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1302,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 8.754315376281738,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1378,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 13.67573070526123,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1375,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 8.554417610168457,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1288,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 7.53700590133667,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.15,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 5.9341840744018555,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.136,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 5.327037334442139,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1483,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 4.735034465789795,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1248,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 8.948857307434082,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.136,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 1.9421674013137817,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1221,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 11.712831497192383,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1416,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8830275229357798,
|
| 1208 |
+
"eval_loss": 0.4112645387649536,
|
| 1209 |
+
"eval_runtime": 2.5142,
|
| 1210 |
+
"eval_samples_per_second": 346.824,
|
| 1211 |
+
"eval_steps_per_second": 3.977,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195247341364224.0,
|
| 1218 |
+
"train_loss": 0.23413483488593687,
|
| 1219 |
+
"train_runtime": 601.3799,
|
| 1220 |
+
"train_samples_per_second": 335.972,
|
| 1221 |
+
"train_steps_per_second": 7.004
|
| 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": 6195247341364224.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8727064220183486,
|
| 4 |
+
"eval_loss": 0.42197221517562866,
|
| 5 |
+
"eval_runtime": 3.2736,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 266.377,
|
| 8 |
+
"eval_steps_per_second": 3.055,
|
| 9 |
+
"total_flos": 6187653780811776.0,
|
| 10 |
+
"train_loss": 0.23428635476095158,
|
| 11 |
+
"train_runtime": 594.1002,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 340.089,
|
| 14 |
+
"train_steps_per_second": 7.09
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/seed_44/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8727064220183486,
|
| 4 |
+
"eval_loss": 0.42197221517562866,
|
| 5 |
+
"eval_runtime": 3.2736,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 266.377,
|
| 8 |
+
"eval_steps_per_second": 3.055
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/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/cls/seed_44 --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_44/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6187653780811776.0,
|
| 4 |
+
"train_loss": 0.23428635476095158,
|
| 5 |
+
"train_runtime": 594.1002,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 340.089,
|
| 8 |
+
"train_steps_per_second": 7.09
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/cls/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": 4.544408798217773,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6962,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 11.815140724182129,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.6148,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 5.806680202484131,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.5426,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 4.855364799499512,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4955,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 5.595414161682129,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4755,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 9.313735008239746,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.421,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 6.8452301025390625,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4446,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 10.4423828125,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4104,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 13.461469650268555,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4436,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 8.394832611083984,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.417,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 7.565339088439941,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.3866,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 5.308486461639404,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4154,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 8.234648704528809,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3724,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.960317134857178,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3806,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 7.42978048324585,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3765,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 7.842386245727539,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3631,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 4.607180118560791,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3632,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 4.794679641723633,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3589,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 11.848945617675781,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3667,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 7.785835266113281,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3372,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 4.946593761444092,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3571,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 6.90305757522583,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.4005,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 7.453240871429443,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.328,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 7.070051670074463,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.367,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 10.425825119018555,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3662,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 6.277216911315918,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3437,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 10.400525093078613,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3134,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.789090394973755,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3083,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 5.684905052185059,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3314,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.152896404266357,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3323,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 4.695758819580078,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3352,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 3.5017030239105225,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3057,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 5.520479202270508,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.352,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 8.196359634399414,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3402,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 6.267371654510498,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3291,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 7.90025520324707,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3032,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 5.636307716369629,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3311,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 5.808764934539795,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3023,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 5.835859775543213,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.3005,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.862026691436768,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3064,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 4.802638530731201,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2894,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 5.215676307678223,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3267,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 4.993435859680176,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2914,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.710698127746582,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3213,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 6.524354934692383,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2718,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 4.122308254241943,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2931,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 5.939773082733154,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2793,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.6997945308685303,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.3038,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 8.996810913085938,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2617,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 6.043160915374756,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3034,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 6.564728736877441,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2674,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 5.635194301605225,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2985,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 3.969360828399658,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.302,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 3.2946078777313232,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2464,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.513924598693848,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.257,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 3.227766990661621,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2762,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8692660550458715,
|
| 406 |
+
"eval_loss": 0.34471234679222107,
|
| 407 |
+
"eval_runtime": 3.4101,
|
| 408 |
+
"eval_samples_per_second": 255.714,
|
| 409 |
+
"eval_steps_per_second": 2.933,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 8.288277626037598,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.221,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 8.34382438659668,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2234,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 6.666543006896973,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2186,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 7.285246849060059,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2113,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 5.339845657348633,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2225,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 5.701568126678467,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2173,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.603057861328125,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1841,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 3.8621342182159424,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2303,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 2.2362220287323,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2088,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 7.996912002563477,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2214,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 4.797756671905518,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2406,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 4.783480644226074,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2323,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 8.793172836303711,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2245,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 7.148190498352051,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2087,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 3.887162208557129,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2225,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 10.475890159606934,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2187,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 7.400747299194336,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.1916,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 3.959892988204956,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.259,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 7.272285461425781,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2158,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 8.383569717407227,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2492,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 4.818758964538574,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2133,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.851695537567139,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2328,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 10.17435359954834,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2158,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 9.315455436706543,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2044,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 8.105263710021973,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2019,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 5.378106594085693,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.1917,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 5.099005699157715,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2176,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 4.740331649780273,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2095,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 9.371432304382324,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2143,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 7.654178142547607,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2279,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.860955238342285,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2171,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 4.804610252380371,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1984,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 6.666094779968262,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2087,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 5.393036365509033,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.2109,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 9.913135528564453,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.1852,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.841836929321289,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.1913,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 5.76794958114624,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1937,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 7.626151084899902,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.161,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 6.066566467285156,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2246,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 6.049985885620117,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2361,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 5.326048851013184,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2102,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 8.22886848449707,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1759,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 7.372525215148926,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1793,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 6.495491027832031,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2184,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 6.921847343444824,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.1823,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.283856391906738,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1957,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 3.458707809448242,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2184,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 4.624142646789551,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.185,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 7.304952144622803,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.2003,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 5.552361488342285,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1938,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 4.119124889373779,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1985,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 7.9433794021606445,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.1981,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 3.933993339538574,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1934,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 4.338261127471924,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.2182,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 3.5682663917541504,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.2103,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 4.894927024841309,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1884,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8772935779816514,
|
| 807 |
+
"eval_loss": 0.3882327973842621,
|
| 808 |
+
"eval_runtime": 3.3663,
|
| 809 |
+
"eval_samples_per_second": 259.035,
|
| 810 |
+
"eval_steps_per_second": 2.971,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 5.493503570556641,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1596,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 5.289029598236084,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1635,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 3.988412380218506,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1292,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 10.415382385253906,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1389,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 7.213047027587891,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1465,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 2.96450138092041,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1217,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 9.286989212036133,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1247,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 6.475372314453125,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1227,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 5.394478797912598,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1202,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 1.415834903717041,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1447,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 2.4288244247436523,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1425,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 3.931483507156372,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1668,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 7.642755031585693,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1212,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 9.016899108886719,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1172,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 1.9267524480819702,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.133,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 6.930219650268555,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1381,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 4.029449462890625,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1476,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 4.7809038162231445,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1542,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 4.229599475860596,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.147,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 6.918832778930664,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1722,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 3.90094256401062,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1271,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 5.178628444671631,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1347,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 13.64695930480957,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1135,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 9.889863967895508,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1428,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 6.82381010055542,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1657,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 4.083231449127197,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1313,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 7.6138224601745605,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1449,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 9.568285942077637,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.135,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 4.168978691101074,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1339,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 7.447011947631836,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.121,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 3.299997091293335,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1294,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 8.263331413269043,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.153,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 7.747345447540283,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1371,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 8.644669532775879,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1348,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 8.861618995666504,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1261,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 7.180167198181152,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1452,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 9.721634864807129,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1449,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 4.829326629638672,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1128,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 5.182653903961182,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1419,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 10.403118133544922,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1546,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 4.860346794128418,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1402,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 4.410172462463379,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1294,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 5.24105978012085,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1407,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 3.597353935241699,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1242,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 8.975448608398438,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1347,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 10.143439292907715,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1247,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 52.937992095947266,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1561,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 5.693874359130859,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1572,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 5.083566188812256,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1294,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 5.132523536682129,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1288,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 7.246787071228027,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1625,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 5.641202926635742,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1287,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 6.785538673400879,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1531,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 6.58427095413208,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1387,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 9.866533279418945,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1264,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 6.858354568481445,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1228,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8727064220183486,
|
| 1208 |
+
"eval_loss": 0.42197221517562866,
|
| 1209 |
+
"eval_runtime": 3.1467,
|
| 1210 |
+
"eval_samples_per_second": 277.112,
|
| 1211 |
+
"eval_steps_per_second": 3.178,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6187653780811776.0,
|
| 1218 |
+
"train_loss": 0.23428635476095158,
|
| 1219 |
+
"train_runtime": 594.1002,
|
| 1220 |
+
"train_samples_per_second": 340.089,
|
| 1221 |
+
"train_steps_per_second": 7.09
|
| 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": 6187653780811776.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/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/gated_multi_branch/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy gated_multi_branch --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.43847355246543884,
|
| 5 |
+
"eval_runtime": 2.3072,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 377.941,
|
| 8 |
+
"eval_steps_per_second": 4.334,
|
| 9 |
+
"total_flos": 6393767311693476.0,
|
| 10 |
+
"train_loss": 0.23293366598041423,
|
| 11 |
+
"train_runtime": 549.6677,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 367.58,
|
| 14 |
+
"train_steps_per_second": 7.663
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.43847355246543884,
|
| 5 |
+
"eval_runtime": 2.3072,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 377.941,
|
| 8 |
+
"eval_steps_per_second": 4.334
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/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/gated_multi_branch/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy gated_multi_branch --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/seed_43/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6393767311693476.0,
|
| 4 |
+
"train_loss": 0.23293366598041423,
|
| 5 |
+
"train_runtime": 549.6677,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 367.58,
|
| 8 |
+
"train_steps_per_second": 7.663
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/gated_multi_branch/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": 12.341727256774902,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.5982,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 5.8127760887146,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5179,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 5.727067470550537,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4743,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 5.470449924468994,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4518,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 5.750891208648682,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4595,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 11.531625747680664,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4591,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.618504047393799,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4462,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 5.614153861999512,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4058,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 5.245266437530518,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4163,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.53200626373291,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.3961,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 6.879703521728516,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4211,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.392645835876465,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4104,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 4.904657363891602,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.4019,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 6.461519718170166,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3872,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 10.551005363464355,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3668,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 6.70601749420166,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3914,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 7.426308631896973,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3742,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 5.763049602508545,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3841,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 6.699713706970215,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.347,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 5.037260055541992,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3404,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 6.125556945800781,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3519,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.1446099281311035,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3473,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 3.1390116214752197,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3307,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 5.179666519165039,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.319,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.599213600158691,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3586,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 5.229942798614502,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3456,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 6.973955154418945,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3328,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.5575859546661377,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3374,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 4.2524638175964355,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3218,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 4.714740753173828,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3736,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 4.409247875213623,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3462,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 7.7085137367248535,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.331,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 6.445059299468994,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3021,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 5.634108543395996,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.2854,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 6.681398391723633,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.351,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 6.271320819854736,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3355,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 3.5563268661499023,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3076,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 8.871476173400879,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.2868,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 6.986723899841309,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2581,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.9431843757629395,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2891,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 4.208820343017578,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.299,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.107792377471924,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3098,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 3.8890693187713623,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2641,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.2277655601501465,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3124,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 8.017739295959473,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2933,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 6.566283226013184,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.289,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 6.2367072105407715,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.3125,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 2.873462438583374,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2737,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 3.463862419128418,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.3081,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 4.5434250831604,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3265,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 7.265514373779297,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2959,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 7.163388252258301,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2734,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 7.241029739379883,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2557,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.708783149719238,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.257,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.133213996887207,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2799,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 4.470254898071289,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2641,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8635321100917431,
|
| 406 |
+
"eval_loss": 0.3724311590194702,
|
| 407 |
+
"eval_runtime": 3.2057,
|
| 408 |
+
"eval_samples_per_second": 272.019,
|
| 409 |
+
"eval_steps_per_second": 3.119,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 5.714033126831055,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.1954,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 4.909178256988525,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2386,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.581356525421143,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2329,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 5.955012798309326,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2393,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 4.976151466369629,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2351,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 6.971107482910156,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2189,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 4.540194511413574,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1975,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 8.91683578491211,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2464,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 6.244039535522461,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.1985,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 4.486268997192383,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2351,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 5.804939270019531,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2122,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 4.952857494354248,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2529,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 9.430203437805176,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2181,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 4.528359413146973,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2222,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 5.020413398742676,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2149,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 4.092240333557129,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2251,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 11.186814308166504,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2279,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 5.696035385131836,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2038,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 5.0304646492004395,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.229,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 8.392675399780273,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.208,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 6.308480262756348,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2011,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 7.648500442504883,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2277,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.166980266571045,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2302,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 8.596280097961426,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2139,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 5.262944221496582,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2148,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 52.07601547241211,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.208,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.120980739593506,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.1897,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 3.347123146057129,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2106,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 4.8367180824279785,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1913,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 5.311898231506348,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2128,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 6.801087379455566,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2299,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 5.1527018547058105,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2089,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 5.476105690002441,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2162,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 7.262950420379639,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.2015,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 9.378935813903809,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2418,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.948502779006958,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2049,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 6.05908203125,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.231,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 5.070615291595459,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2003,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 7.837033271789551,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2013,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 4.444377422332764,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.1979,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 5.344629764556885,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.185,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 6.982371807098389,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2043,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 9.339909553527832,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1946,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 7.406871795654297,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2182,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 3.94970440864563,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2033,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 7.589832305908203,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.18,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 7.134607315063477,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.1989,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 8.96692180633545,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1953,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 2.87789249420166,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1838,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 4.450316429138184,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1927,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.040008544921875,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1891,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.559047698974609,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.1993,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.694094657897949,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.2163,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 6.131115913391113,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1879,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 6.2520952224731445,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1934,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 9.09149169921875,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1906,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8738532110091743,
|
| 807 |
+
"eval_loss": 0.35964497923851013,
|
| 808 |
+
"eval_runtime": 2.6419,
|
| 809 |
+
"eval_samples_per_second": 330.071,
|
| 810 |
+
"eval_steps_per_second": 3.785,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 9.36866569519043,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1372,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 4.772096157073975,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.15,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 1.3709602355957031,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1316,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 5.077552795410156,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.137,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 8.544183731079102,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1445,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 4.3138909339904785,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1182,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 8.031082153320312,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.124,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 7.775125026702881,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1434,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 11.894580841064453,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1456,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 5.693721771240234,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1407,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 10.361031532287598,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1606,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 4.53510856628418,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.154,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 6.4371538162231445,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1534,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 3.6855733394622803,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.134,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 11.527420043945312,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.146,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 7.941448211669922,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1277,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 2.830364465713501,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1281,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 6.681955337524414,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1728,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 9.449584007263184,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1471,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 9.082175254821777,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.173,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 11.798688888549805,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.143,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 7.8705058097839355,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1259,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 3.313554525375366,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1565,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.689424514770508,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1383,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 5.623819828033447,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1281,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 8.085272789001465,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1364,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 8.369499206542969,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1419,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 4.375655174255371,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1447,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 6.895667552947998,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1392,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 6.115300178527832,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1472,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 8.54385757446289,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1544,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 5.993228435516357,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.118,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 4.3533782958984375,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1364,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 2.8014254570007324,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1483,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 5.457911491394043,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1344,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 6.010044097900391,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1145,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.986755847930908,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1136,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 8.168453216552734,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1238,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 9.592988014221191,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1517,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 5.39243221282959,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.153,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 5.2420735359191895,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1341,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 6.862229347229004,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1289,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 5.822480201721191,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1312,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 7.827848434448242,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1433,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 9.657716751098633,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1325,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 7.985351085662842,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1341,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 7.562764644622803,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1513,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 8.586878776550293,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1202,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 14.69101619720459,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1492,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 6.459609031677246,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1243,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 3.9858133792877197,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1308,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 10.252853393554688,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1443,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 5.580255031585693,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1293,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 7.237607479095459,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1503,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 2.5081088542938232,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.118,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 6.452218532562256,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1443,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8784403669724771,
|
| 1208 |
+
"eval_loss": 0.43847355246543884,
|
| 1209 |
+
"eval_runtime": 2.3905,
|
| 1210 |
+
"eval_samples_per_second": 364.775,
|
| 1211 |
+
"eval_steps_per_second": 4.183,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6393767311693476.0,
|
| 1218 |
+
"train_loss": 0.23293366598041423,
|
| 1219 |
+
"train_runtime": 549.6677,
|
| 1220 |
+
"train_samples_per_second": 367.58,
|
| 1221 |
+
"train_steps_per_second": 7.663
|
| 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": 6393767311693476.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|