Upload batch 16/24: sst2/mBERT/max/seed_42 ... sst2/mBERT/mean/seed_42
Browse files- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/trainer_state.json +1245 -0
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8715596330275229,
|
| 4 |
+
"eval_loss": 0.44055378437042236,
|
| 5 |
+
"eval_runtime": 1.9862,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 439.038,
|
| 8 |
+
"eval_steps_per_second": 5.035,
|
| 9 |
+
"total_flos": 6195598393181184.0,
|
| 10 |
+
"train_loss": 0.23420973027307196,
|
| 11 |
+
"train_runtime": 611.5724,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 330.373,
|
| 14 |
+
"train_steps_per_second": 6.887
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8715596330275229,
|
| 4 |
+
"eval_loss": 0.44055378437042236,
|
| 5 |
+
"eval_runtime": 1.9862,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 439.038,
|
| 8 |
+
"eval_steps_per_second": 5.035
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/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/max/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy max --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195598393181184.0,
|
| 4 |
+
"train_loss": 0.23420973027307196,
|
| 5 |
+
"train_runtime": 611.5724,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 330.373,
|
| 8 |
+
"train_steps_per_second": 6.887
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_42/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 4.572626113891602,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6556,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 5.642662048339844,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.531,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 5.865051746368408,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4909,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 4.161923885345459,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4493,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 4.692380905151367,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4633,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 9.864856719970703,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4522,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 4.178947448730469,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4322,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 5.8239359855651855,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4346,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 5.256783485412598,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4222,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 5.055699348449707,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4092,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 4.693850517272949,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4161,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 5.928548336029053,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.408,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 8.231057167053223,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.362,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 4.433553218841553,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3742,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 5.139867305755615,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3762,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 7.152890682220459,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3639,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 5.7032318115234375,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3649,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 6.602631092071533,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3779,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 6.3586106300354,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3418,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 3.4253218173980713,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3506,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 5.91262674331665,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3412,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.994682788848877,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3776,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 8.793638229370117,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3092,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 6.566744804382324,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.318,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.559112071990967,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3517,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 7.370287895202637,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3211,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 5.444779872894287,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3672,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 7.23353910446167,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3221,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 6.892766952514648,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3274,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.45596170425415,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3028,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 6.6439948081970215,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3236,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 6.414327144622803,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.2812,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 4.234644889831543,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3449,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 7.113037586212158,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3195,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 5.406424045562744,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3025,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 5.059304714202881,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3326,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 5.244286060333252,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2988,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 2.8773767948150635,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3044,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.933010101318359,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.3033,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.674722194671631,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3348,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 11.249885559082031,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2793,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 4.0882487297058105,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3166,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 5.73197078704834,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2865,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 7.728004455566406,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2864,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 5.918716907501221,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.294,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 4.736082553863525,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.3126,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 4.772009372711182,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2953,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 8.847880363464355,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.271,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 8.24442195892334,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2644,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 8.77250862121582,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.323,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 7.804999828338623,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2838,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 3.2050528526306152,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2819,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 5.419010162353516,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2609,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.360693454742432,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2882,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.282514572143555,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.3117,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 4.647046089172363,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2697,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8543577981651376,
|
| 406 |
+
"eval_loss": 0.3604632318019867,
|
| 407 |
+
"eval_runtime": 2.9218,
|
| 408 |
+
"eval_samples_per_second": 298.447,
|
| 409 |
+
"eval_steps_per_second": 3.423,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 6.0091376304626465,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2365,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 3.118197202682495,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.263,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.978915214538574,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2416,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 5.657942295074463,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2268,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 6.424144268035889,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2194,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 4.324857711791992,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.1908,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.596123218536377,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2254,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 6.009299278259277,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2246,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 6.5087666511535645,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2099,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 3.807996988296509,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2086,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 4.7029876708984375,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2271,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 8.97614860534668,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2336,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 3.313724994659424,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.233,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 4.740766525268555,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2359,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 5.9057416915893555,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2192,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 9.381734848022461,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2254,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 5.515054225921631,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2271,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 5.436347007751465,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.1955,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 4.496245861053467,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.1803,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 5.424933433532715,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2227,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 7.030899524688721,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2177,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 5.324127197265625,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2039,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 3.8943467140197754,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2032,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 4.613260746002197,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2095,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 7.127693176269531,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.1997,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 6.521519660949707,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2195,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 5.181880950927734,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2142,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 5.8275580406188965,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.217,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 6.496601581573486,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1974,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 5.465758323669434,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.406064033508301,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2137,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 6.331623554229736,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1931,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 3.5731804370880127,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2058,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 12.332954406738281,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1819,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 7.745568752288818,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.228,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 5.354550838470459,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2297,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 5.07284688949585,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.2,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 8.026025772094727,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2147,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 4.400514602661133,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.178,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 5.277686595916748,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.1942,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 7.074545383453369,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2094,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 6.552160263061523,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2159,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 5.432361602783203,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1711,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 6.4135518074035645,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2121,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 5.134510517120361,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2247,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 5.401628017425537,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.2093,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 5.999505043029785,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2044,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 5.34188175201416,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.201,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 3.8825325965881348,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.2093,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 10.037313461303711,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2205,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 5.914582252502441,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.2138,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 4.430501937866211,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2063,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.060576438903809,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1787,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 10.095841407775879,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1924,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 4.202109336853027,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1961,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 2.330782651901245,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1968,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8600917431192661,
|
| 807 |
+
"eval_loss": 0.42205187678337097,
|
| 808 |
+
"eval_runtime": 2.8601,
|
| 809 |
+
"eval_samples_per_second": 304.889,
|
| 810 |
+
"eval_steps_per_second": 3.496,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 4.521354675292969,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1412,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 1.9365651607513428,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.159,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 6.060638427734375,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1567,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 4.342278003692627,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1386,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 2.665410280227661,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1577,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 5.060510158538818,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1598,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 4.072982311248779,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1436,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 4.2295355796813965,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.158,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 4.787444591522217,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1395,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 6.345071792602539,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1667,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 5.390758037567139,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1544,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 5.116415023803711,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1685,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 8.77122688293457,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.137,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 5.480506896972656,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1298,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 7.008491039276123,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1453,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 5.334630012512207,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1559,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 11.077719688415527,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1266,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 6.5304155349731445,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1537,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 8.476485252380371,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1598,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 6.34197998046875,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1261,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 8.567544937133789,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1489,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 4.856594085693359,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.153,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 4.469034194946289,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1409,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.7454376220703125,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1554,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 4.455084800720215,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1695,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 6.617997646331787,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1243,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 8.007392883300781,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1292,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 4.090811252593994,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1113,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 6.6732330322265625,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1291,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 4.921858787536621,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1721,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 4.100010871887207,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1608,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 8.19946575164795,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.128,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 6.919534683227539,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1312,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 12.261134147644043,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1446,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 7.7642998695373535,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1602,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 4.801642417907715,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1433,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.496195316314697,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1598,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 3.9703192710876465,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1213,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 4.7190704345703125,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1153,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 6.287429332733154,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1611,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 6.068782806396484,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1178,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 7.431860446929932,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1343,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 7.969870567321777,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1367,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 10.325299263000488,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1546,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 6.497237205505371,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1353,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 5.315041542053223,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1358,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 2.1374499797821045,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1407,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 2.6150763034820557,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1383,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 11.517195701599121,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1245,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 4.165342330932617,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1138,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 7.029419898986816,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1348,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 6.481456279754639,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1309,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 2.5296356678009033,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1307,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.275231838226318,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1485,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 3.98374342918396,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1131,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 5.982683181762695,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1452,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8715596330275229,
|
| 1208 |
+
"eval_loss": 0.44055378437042236,
|
| 1209 |
+
"eval_runtime": 2.0845,
|
| 1210 |
+
"eval_samples_per_second": 418.316,
|
| 1211 |
+
"eval_steps_per_second": 4.797,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195598393181184.0,
|
| 1218 |
+
"train_loss": 0.23420973027307196,
|
| 1219 |
+
"train_runtime": 611.5724,
|
| 1220 |
+
"train_samples_per_second": 330.373,
|
| 1221 |
+
"train_steps_per_second": 6.887
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6195598393181184.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.45231905579566956,
|
| 5 |
+
"eval_runtime": 2.9696,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 293.642,
|
| 8 |
+
"eval_steps_per_second": 3.367,
|
| 9 |
+
"total_flos": 6195247341364224.0,
|
| 10 |
+
"train_loss": 0.2331274027837987,
|
| 11 |
+
"train_runtime": 583.7709,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 346.107,
|
| 14 |
+
"train_steps_per_second": 7.215
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.45231905579566956,
|
| 5 |
+
"eval_runtime": 2.9696,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 293.642,
|
| 8 |
+
"eval_steps_per_second": 3.367
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/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/max/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy max --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195247341364224.0,
|
| 4 |
+
"train_loss": 0.2331274027837987,
|
| 5 |
+
"train_runtime": 583.7709,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 346.107,
|
| 8 |
+
"train_steps_per_second": 7.215
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_43/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 10.533888816833496,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6141,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 6.087037563323975,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5242,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 7.794853687286377,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4732,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 4.941077709197998,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4738,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 11.120641708374023,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.5189,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 9.229264259338379,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4969,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.0303802490234375,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.444,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 5.746410369873047,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.3957,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 4.253859996795654,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4371,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.077849388122559,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4091,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 4.020836353302002,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4167,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.866827487945557,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4038,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 4.986913681030273,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.4117,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 3.6944093704223633,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3811,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 11.268630981445312,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3654,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 6.277194976806641,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.4182,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 8.13642406463623,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3661,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 5.497155666351318,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3715,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 5.32804012298584,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3509,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 4.229322910308838,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3421,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 8.549903869628906,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3296,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.3918538093566895,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3423,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 7.303037166595459,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3336,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 6.252259254455566,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3168,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.800606727600098,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3498,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 3.7319788932800293,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3191,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 7.328004837036133,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.34,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 2.73371958732605,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3246,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.143244504928589,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3167,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 4.831525802612305,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3529,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 3.9505984783172607,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3354,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 7.3627190589904785,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3121,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 5.624882698059082,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3044,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 4.087714672088623,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.285,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 5.534547805786133,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3425,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 5.4497971534729,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3182,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 6.556898593902588,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3047,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 12.05080509185791,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.2988,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.211089611053467,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2872,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 7.8275957107543945,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2916,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 5.235176086425781,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2916,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.2151002883911133,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.2996,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 3.5429224967956543,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2695,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.168999195098877,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3228,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 7.725827217102051,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2818,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.353954315185547,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.292,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 7.166689872741699,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.3063,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 4.725644588470459,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2686,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 5.433608531951904,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.3164,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 4.89896821975708,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3199,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 4.294034481048584,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2993,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 9.50782299041748,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2729,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 6.67723274230957,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2808,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.270874500274658,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2549,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.390813827514648,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.3065,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 5.145437717437744,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2583,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8704128440366973,
|
| 406 |
+
"eval_loss": 0.3591882586479187,
|
| 407 |
+
"eval_runtime": 3.0112,
|
| 408 |
+
"eval_samples_per_second": 289.589,
|
| 409 |
+
"eval_steps_per_second": 3.321,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 5.855111598968506,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.198,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 5.23792839050293,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2347,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 3.466867208480835,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2332,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 4.278041362762451,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2257,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 4.987913608551025,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2272,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 10.01087474822998,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.22,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.252983570098877,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1933,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 10.657994270324707,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2417,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 4.6842217445373535,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2031,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 6.226792335510254,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2402,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 7.891331672668457,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2358,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 5.509846210479736,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.245,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 5.382606506347656,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2134,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 5.331611633300781,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2099,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 3.2548651695251465,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2228,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 3.7445404529571533,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2204,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 7.570513725280762,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2236,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 6.178224563598633,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2042,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 7.369462490081787,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.212,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 7.667268753051758,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2111,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 7.186551094055176,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2145,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.26616907119751,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2345,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.855222702026367,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2123,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 6.575676918029785,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2185,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 6.971273422241211,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2373,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 9.297901153564453,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2022,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 4.781652927398682,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.1945,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 5.836018085479736,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.1906,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 5.54315185546875,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2179,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 5.965412139892578,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2018,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 5.88109827041626,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2252,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 4.788642406463623,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1926,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.205417633056641,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2285,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 9.735586166381836,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1961,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 7.466604232788086,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2272,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.185211658477783,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2134,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 7.947246074676514,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.2469,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 4.1298017501831055,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2024,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 7.961837291717529,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1916,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 5.81777286529541,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2115,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 3.4709064960479736,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.178,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 5.54365348815918,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1945,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 3.3070201873779297,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1944,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 6.19324254989624,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2163,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 6.503810405731201,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.1985,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.444545745849609,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1912,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 8.013023376464844,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.1804,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 5.5860066413879395,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.2041,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 4.696243762969971,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1798,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 6.9738640785217285,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1803,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 4.754683971405029,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1997,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.981789588928223,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2039,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 8.075098991394043,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.2026,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 6.301675319671631,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1808,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 6.82260799407959,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1798,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 5.5957255363464355,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1855,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8864678899082569,
|
| 807 |
+
"eval_loss": 0.366256445646286,
|
| 808 |
+
"eval_runtime": 3.0647,
|
| 809 |
+
"eval_samples_per_second": 284.53,
|
| 810 |
+
"eval_steps_per_second": 3.263,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 11.356785774230957,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1483,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 5.76664924621582,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1369,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 7.5251007080078125,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1345,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 9.718250274658203,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1275,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 5.473138332366943,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1487,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 5.799095630645752,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1331,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 6.039730548858643,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1053,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 7.5315046310424805,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1713,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 13.346685409545898,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1277,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 6.536266803741455,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1611,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 7.972631454467773,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1487,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 5.839498519897461,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1677,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 11.375701904296875,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1463,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 3.1823294162750244,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1551,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 9.296834945678711,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1588,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 8.869190216064453,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1164,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 1.2889273166656494,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1376,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 5.465572834014893,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1861,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 5.926852703094482,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.164,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 8.36825942993164,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1499,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 6.375903129577637,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1351,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 4.985777854919434,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.121,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 5.228201866149902,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1456,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 7.188004493713379,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1227,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 6.393586158752441,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1388,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 5.2751617431640625,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1281,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 6.462953090667725,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1454,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 5.000539779663086,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1433,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 11.502623558044434,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1452,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 9.126944541931152,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1489,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 11.926445960998535,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1628,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 12.38862419128418,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1154,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 5.735890865325928,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1303,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 5.791765213012695,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1407,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 9.762746810913086,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1355,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 3.998180866241455,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1256,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 7.290683746337891,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1208,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 5.552916049957275,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1262,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.960696220397949,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1512,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 4.352052688598633,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.138,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 3.2729454040527344,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1471,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 3.7950143814086914,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1292,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 8.923233985900879,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1232,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 5.807420253753662,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1541,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 7.672214508056641,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1256,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 7.12013578414917,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.121,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 5.2720723152160645,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1508,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 4.515068054199219,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1209,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 5.374870777130127,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.12,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 6.523993015289307,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1476,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 5.954843521118164,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.129,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 6.340404987335205,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1443,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 5.985660552978516,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1227,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 3.0158228874206543,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1538,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 3.819382429122925,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1209,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 14.777352333068848,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1369,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8784403669724771,
|
| 1208 |
+
"eval_loss": 0.45231905579566956,
|
| 1209 |
+
"eval_runtime": 2.9669,
|
| 1210 |
+
"eval_samples_per_second": 293.911,
|
| 1211 |
+
"eval_steps_per_second": 3.371,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195247341364224.0,
|
| 1218 |
+
"train_loss": 0.2331274027837987,
|
| 1219 |
+
"train_runtime": 583.7709,
|
| 1220 |
+
"train_samples_per_second": 346.107,
|
| 1221 |
+
"train_steps_per_second": 7.215
|
| 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/max/seed_44/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.42756688594818115,
|
| 5 |
+
"eval_runtime": 3.0471,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 286.174,
|
| 8 |
+
"eval_steps_per_second": 3.282,
|
| 9 |
+
"total_flos": 6187653780811776.0,
|
| 10 |
+
"train_loss": 0.23267899721096724,
|
| 11 |
+
"train_runtime": 602.3932,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 335.407,
|
| 14 |
+
"train_steps_per_second": 6.992
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.42756688594818115,
|
| 5 |
+
"eval_runtime": 3.0471,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 286.174,
|
| 8 |
+
"eval_steps_per_second": 3.282
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/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/max/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy max --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/seed_44/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6187653780811776.0,
|
| 4 |
+
"train_loss": 0.23267899721096724,
|
| 5 |
+
"train_runtime": 602.3932,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 335.407,
|
| 8 |
+
"train_steps_per_second": 6.992
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/max/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": 2.4613089561462402,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6673,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 9.352470397949219,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5424,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 6.129732608795166,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.5156,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 6.724287509918213,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.509,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 4.349117755889893,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4712,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 7.8638834953308105,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.424,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.144013404846191,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4315,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 7.804922580718994,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4179,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 6.48636531829834,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4432,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 6.120077610015869,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.3917,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 5.933201789855957,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.3906,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 4.916437149047852,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4093,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 6.760597229003906,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.364,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 4.732594966888428,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3793,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 4.7152323722839355,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3656,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 5.935976028442383,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3542,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 4.335594177246094,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3688,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 4.671463966369629,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3593,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 8.532559394836426,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3651,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 8.62664794921875,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3292,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 4.108664035797119,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3589,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 3.7505433559417725,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3872,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 5.499462604522705,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3278,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 7.6328325271606445,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3479,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.765148162841797,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3539,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 5.427891254425049,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3403,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 12.525001525878906,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3058,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 4.497734546661377,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3176,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 5.479736328125,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3404,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 6.656313896179199,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3238,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 6.293766021728516,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3364,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 3.2362658977508545,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3037,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 6.181748390197754,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3383,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 7.096750736236572,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3165,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 7.634105205535889,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3337,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 5.226578712463379,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3009,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 5.2742204666137695,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3238,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 4.9047465324401855,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3059,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 5.629724025726318,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.3173,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.653864860534668,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2942,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 4.941677570343018,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2839,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 6.002717971801758,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3144,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 4.273251056671143,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2756,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 4.769801139831543,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3291,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 6.7584028244018555,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2821,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.414306640625,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2863,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 5.301296234130859,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2771,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.481055974960327,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.3076,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 5.969762325286865,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.253,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 6.241204261779785,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.298,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 4.602875709533691,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2754,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 4.113981246948242,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.3135,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 4.297750949859619,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2932,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 2.358107805252075,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.243,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 4.090692520141602,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2661,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 5.151073455810547,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2775,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8646788990825688,
|
| 406 |
+
"eval_loss": 0.3502647280693054,
|
| 407 |
+
"eval_runtime": 3.2868,
|
| 408 |
+
"eval_samples_per_second": 265.306,
|
| 409 |
+
"eval_steps_per_second": 3.042,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 9.066251754760742,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2258,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 8.02142333984375,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2275,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 6.521266460418701,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2251,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 8.209929466247559,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2219,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 5.013307094573975,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2196,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 4.47982931137085,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2263,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 3.648218870162964,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2033,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 2.921250104904175,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2224,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 1.6407908201217651,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2076,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 5.652299880981445,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2204,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 3.0928351879119873,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2337,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 4.4479451179504395,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2102,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 11.119450569152832,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2216,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 7.519162178039551,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2088,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 4.343770980834961,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2217,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 9.132966041564941,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2193,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 8.013995170593262,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.1954,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 4.957169532775879,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.242,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 4.039794445037842,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2274,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 8.598875999450684,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2422,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 5.111716270446777,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2218,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.959665775299072,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2191,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.033991813659668,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.21,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 8.66789436340332,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2128,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 9.918187141418457,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.1932,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 4.617578983306885,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2004,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.3206682205200195,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2106,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 5.957472324371338,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2205,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 8.056126594543457,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2064,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 6.829466342926025,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.225,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 3.8213694095611572,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2082,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 7.349196434020996,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2028,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.094357967376709,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2066,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 11.346162796020508,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.2067,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 6.7377028465271,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2033,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.634881019592285,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.1913,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 5.957315921783447,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1921,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 5.096012115478516,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.1695,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 4.20332145690918,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2121,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 7.939763069152832,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2306,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 5.450836658477783,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2053,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 6.280238628387451,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1818,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 6.123204708099365,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1948,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 7.73919153213501,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2255,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 7.03324031829834,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.1862,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.0663347244262695,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1917,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 3.621750831604004,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2186,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 4.200271129608154,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1761,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 6.466761112213135,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1878,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 5.006765365600586,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1913,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 3.351299285888672,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.189,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 6.840633392333984,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.1991,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.051394462585449,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.188,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 5.012309551239014,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.212,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 2.834137439727783,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.2102,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 4.473265171051025,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.2118,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8807339449541285,
|
| 807 |
+
"eval_loss": 0.34572434425354004,
|
| 808 |
+
"eval_runtime": 3.1593,
|
| 809 |
+
"eval_samples_per_second": 276.011,
|
| 810 |
+
"eval_steps_per_second": 3.165,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 5.904903888702393,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1599,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 2.59757137298584,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1537,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 5.345468521118164,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.132,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 4.515627384185791,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1488,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 7.912057399749756,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1458,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 1.8391884565353394,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1251,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 5.444867134094238,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1317,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 6.585514545440674,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1174,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 6.712602138519287,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.12,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 0.7592522501945496,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1507,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 4.0161967277526855,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1498,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 3.9227731227874756,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1549,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 6.278088569641113,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1177,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 8.586590766906738,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1218,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 2.6703341007232666,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1511,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 11.908092498779297,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1473,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 5.823470115661621,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1481,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 8.432798385620117,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1519,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 4.938522815704346,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1377,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 6.975924491882324,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1683,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 4.234536170959473,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1245,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 12.742045402526855,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1241,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 8.45624828338623,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1215,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 14.034553527832031,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1313,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 7.723348140716553,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1643,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 4.675960540771484,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1311,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 7.044912338256836,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1546,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 11.560933113098145,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1315,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 1.0812780857086182,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1338,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 7.997935771942139,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1205,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 3.060136556625366,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1341,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 5.3422675132751465,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1519,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 9.257694244384766,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1356,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 7.157827854156494,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1406,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 8.246857643127441,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.134,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 5.715890407562256,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1428,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 5.1389899253845215,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.129,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 7.9778265953063965,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1159,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.096800804138184,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1347,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 8.13379955291748,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1366,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 5.279202461242676,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1324,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 7.436024188995361,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.115,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 7.5911545753479,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1593,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 4.309954643249512,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1251,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 11.659241676330566,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1438,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 9.77135181427002,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1178,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 8.901510238647461,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1434,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 6.801113128662109,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1625,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 4.763524055480957,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1101,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 7.274860858917236,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.135,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 6.5425496101379395,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1543,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 5.345522403717041,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1324,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 6.087719917297363,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1593,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.432441234588623,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1533,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 9.946971893310547,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1267,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 3.9437642097473145,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.126,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8784403669724771,
|
| 1208 |
+
"eval_loss": 0.42756688594818115,
|
| 1209 |
+
"eval_runtime": 3.0086,
|
| 1210 |
+
"eval_samples_per_second": 289.831,
|
| 1211 |
+
"eval_steps_per_second": 3.324,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6187653780811776.0,
|
| 1218 |
+
"train_loss": 0.23267899721096724,
|
| 1219 |
+
"train_runtime": 602.3932,
|
| 1220 |
+
"train_samples_per_second": 335.407,
|
| 1221 |
+
"train_steps_per_second": 6.992
|
| 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/mean/seed_42/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8692660550458715,
|
| 4 |
+
"eval_loss": 0.4306594431400299,
|
| 5 |
+
"eval_runtime": 3.0758,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 283.504,
|
| 8 |
+
"eval_steps_per_second": 3.251,
|
| 9 |
+
"total_flos": 6195598393181184.0,
|
| 10 |
+
"train_loss": 0.23150781453618072,
|
| 11 |
+
"train_runtime": 607.3299,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 332.681,
|
| 14 |
+
"train_steps_per_second": 6.935
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8692660550458715,
|
| 4 |
+
"eval_loss": 0.4306594431400299,
|
| 5 |
+
"eval_runtime": 3.0758,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 283.504,
|
| 8 |
+
"eval_steps_per_second": 3.251
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/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/mean/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy mean --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6195598393181184.0,
|
| 4 |
+
"train_loss": 0.23150781453618072,
|
| 5 |
+
"train_runtime": 607.3299,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 332.681,
|
| 8 |
+
"train_steps_per_second": 6.935
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mean/seed_42/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 6.04003381729126,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6446,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 8.885544776916504,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5425,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 6.380282878875732,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4724,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 4.395967960357666,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4302,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 5.154633522033691,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4498,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 9.002699851989746,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4527,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 4.869556427001953,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4307,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 6.635958671569824,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4244,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 4.126345157623291,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4031,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 3.699632167816162,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.414,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 4.352095127105713,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4211,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 5.182868480682373,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4127,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 8.215089797973633,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3583,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.067569255828857,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3813,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 7.131149768829346,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3879,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 5.645003318786621,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.3502,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 6.8105340003967285,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3552,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 8.141983032226562,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3966,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 5.963000297546387,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3313,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 3.7655272483825684,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3518,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 6.811331748962402,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3344,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 4.010414123535156,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3761,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 9.180388450622559,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3036,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 6.658926963806152,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.328,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 7.350872993469238,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3501,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 7.3605055809021,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3145,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 4.212459564208984,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3637,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 7.717599868774414,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3165,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 6.80646276473999,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3088,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 7.617321014404297,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.2957,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 6.6909332275390625,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3323,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 7.510091304779053,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.2842,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 3.5400657653808594,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3299,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.912785530090332,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3115,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 5.293907165527344,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3095,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 3.7852420806884766,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3405,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 3.7965898513793945,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2952,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 3.3105015754699707,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.309,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.677436828613281,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.315,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 5.102982044219971,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3257,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 8.322901725769043,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.279,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 4.952832221984863,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3179,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 7.331367492675781,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2918,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 4.884490013122559,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2854,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 7.155500888824463,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2934,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 4.2931694984436035,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.3199,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 3.8412227630615234,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.3078,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 5.696902275085449,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2861,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 7.062839508056641,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2733,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 7.343217849731445,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3191,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 7.958421230316162,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2712,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 5.715513229370117,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2711,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 9.136781692504883,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2466,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.686506748199463,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2951,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 5.792591571807861,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.295,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 4.701540946960449,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2737,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8635321100917431,
|
| 406 |
+
"eval_loss": 0.3424193263053894,
|
| 407 |
+
"eval_runtime": 3.3164,
|
| 408 |
+
"eval_samples_per_second": 262.935,
|
| 409 |
+
"eval_steps_per_second": 3.015,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 4.551516532897949,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2348,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 3.823364734649658,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2571,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 5.75368070602417,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2398,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 6.450377464294434,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2144,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 6.598897933959961,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2242,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 4.258050441741943,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.1803,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 4.716231822967529,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.2215,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 5.655419826507568,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2108,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 6.684243679046631,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2044,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 5.229068279266357,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2145,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 5.563201904296875,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2318,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 7.157746315002441,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2226,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 3.1262576580047607,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2153,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 3.519652843475342,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.223,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 6.058931350708008,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2184,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 9.718982696533203,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2306,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 7.412153720855713,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2215,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 4.828917026519775,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.1932,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 2.9115374088287354,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.1996,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 4.011838436126709,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2121,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 9.254751205444336,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2174,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.857285022735596,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.1869,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.562695503234863,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.1927,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 4.998191833496094,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2102,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 7.511777877807617,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.1964,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 8.692811965942383,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2119,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.6013994216918945,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2209,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 6.491044998168945,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2138,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 4.157855987548828,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1847,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 3.590541362762451,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2157,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.870674133300781,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2154,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 5.3823394775390625,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1877,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 4.364664077758789,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.1934,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 7.614866256713867,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1739,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 5.746688365936279,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2084,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 6.020204067230225,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2358,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 8.057097434997559,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1889,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 5.1826043128967285,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2059,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 3.8114278316497803,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1737,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 5.121821880340576,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.1992,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 7.687273979187012,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.1934,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 8.796460151672363,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2269,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 4.609219074249268,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1837,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 5.105912685394287,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2139,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 4.148934364318848,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.216,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.52239990234375,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.2163,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 3.170016288757324,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2074,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 5.788916110992432,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1966,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 3.807603120803833,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1993,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 7.76474666595459,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2218,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.403115749359131,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.2077,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.8140387535095215,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2039,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 5.974566459655762,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1867,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 8.720250129699707,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1781,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 5.427388668060303,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1926,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 3.859157085418701,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.193,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8761467889908257,
|
| 807 |
+
"eval_loss": 0.3834470808506012,
|
| 808 |
+
"eval_runtime": 3.1435,
|
| 809 |
+
"eval_samples_per_second": 277.398,
|
| 810 |
+
"eval_steps_per_second": 3.181,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 6.748839855194092,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1421,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 3.2398862838745117,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1362,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 5.2825188636779785,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1391,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 7.222107410430908,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1447,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 2.1408352851867676,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1642,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 6.2689361572265625,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1355,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 7.287732124328613,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1467,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 6.28416633605957,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1473,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 5.486760139465332,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1451,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 7.431427955627441,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1577,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 5.5903401374816895,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1525,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 1.7625778913497925,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.151,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 10.455538749694824,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1519,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 6.260527610778809,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1293,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 5.963783264160156,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1383,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 4.568897247314453,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1594,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 9.205401420593262,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.118,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 5.754770278930664,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1482,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 6.858963489532471,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1662,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 5.833136081695557,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1378,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 11.824575424194336,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1353,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 5.310589790344238,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1567,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 8.139506340026855,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1333,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.791632175445557,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1429,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 4.3377485275268555,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1566,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 5.0084638595581055,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1222,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 5.0423502922058105,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1326,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 3.9388930797576904,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1057,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 8.035558700561523,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.122,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 5.809456825256348,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1438,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 5.5651068687438965,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1717,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 7.344658851623535,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1335,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 10.918816566467285,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1321,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 8.628555297851562,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1308,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 6.868607521057129,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1599,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 2.1236460208892822,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1418,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.06664514541626,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.167,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 4.013212203979492,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1166,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.86183500289917,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1135,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 3.5960147380828857,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.16,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 7.336146354675293,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1298,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 9.158062934875488,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1314,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 8.906784057617188,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1377,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 6.262416362762451,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1505,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 5.94282865524292,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1249,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 6.255961894989014,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.134,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 3.256608247756958,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1421,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 3.978806257247925,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1347,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 3.143078565597534,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.118,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 5.365161418914795,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1215,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 7.030867576599121,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.131,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 8.20202350616455,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1371,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 3.876202344894409,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1325,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.693392276763916,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1422,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 3.839710235595703,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1215,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 4.007699966430664,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1366,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8692660550458715,
|
| 1208 |
+
"eval_loss": 0.4306594431400299,
|
| 1209 |
+
"eval_runtime": 3.1687,
|
| 1210 |
+
"eval_samples_per_second": 275.191,
|
| 1211 |
+
"eval_steps_per_second": 3.156,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6195598393181184.0,
|
| 1218 |
+
"train_loss": 0.23150781453618072,
|
| 1219 |
+
"train_runtime": 607.3299,
|
| 1220 |
+
"train_samples_per_second": 332.681,
|
| 1221 |
+
"train_steps_per_second": 6.935
|
| 1222 |
+
}
|
| 1223 |
+
],
|
| 1224 |
+
"logging_steps": 25,
|
| 1225 |
+
"max_steps": 4212,
|
| 1226 |
+
"num_input_tokens_seen": 0,
|
| 1227 |
+
"num_train_epochs": 3,
|
| 1228 |
+
"save_steps": 500,
|
| 1229 |
+
"stateful_callbacks": {
|
| 1230 |
+
"TrainerControl": {
|
| 1231 |
+
"args": {
|
| 1232 |
+
"should_epoch_stop": false,
|
| 1233 |
+
"should_evaluate": false,
|
| 1234 |
+
"should_log": false,
|
| 1235 |
+
"should_save": false,
|
| 1236 |
+
"should_training_stop": false
|
| 1237 |
+
},
|
| 1238 |
+
"attributes": {}
|
| 1239 |
+
}
|
| 1240 |
+
},
|
| 1241 |
+
"total_flos": 6195598393181184.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|