Upload batch 18/24: sst2/mBERT/mha_attention/seed_44 ... sst2/mBERT/multi_branch_average/seed_44
Browse files- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/all_results.json +15 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/eval_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/run_command.txt +1 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/train_results.json +9 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/trainer_state.json +1245 -0
- mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_44/run_command.txt +1 -0
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8853211009174312,
|
| 4 |
+
"eval_loss": 0.4112522304058075,
|
| 5 |
+
"eval_runtime": 3.0837,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 282.782,
|
| 8 |
+
"eval_steps_per_second": 3.243,
|
| 9 |
+
"total_flos": 6215986915810056.0,
|
| 10 |
+
"train_loss": 0.2333670342296843,
|
| 11 |
+
"train_runtime": 624.4636,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 323.553,
|
| 14 |
+
"train_steps_per_second": 6.745
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8853211009174312,
|
| 4 |
+
"eval_loss": 0.4112522304058075,
|
| 5 |
+
"eval_runtime": 3.0837,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 282.782,
|
| 8 |
+
"eval_steps_per_second": 3.243
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/run_command.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
/usr/local/bin/python /workspace/run_glue_MHA_gated.py --model_name_or_path bert-base-multilingual-cased --dataset_name tmnam20/ViGLUE --task_name sst2 --max_seq_length 128 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 25 --report_to none --num_train_epochs 3 --learning_rate 3e-05 --per_device_train_batch_size 48 --per_device_eval_batch_size 96 --gradient_accumulation_steps 1 --dataloader_num_workers 1 --seed 44 --output_dir result_ablation_mbert/sst2/mBERT/mha_attention/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy mha_attention --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6215986915810056.0,
|
| 4 |
+
"train_loss": 0.2333670342296843,
|
| 5 |
+
"train_runtime": 624.4636,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 323.553,
|
| 8 |
+
"train_steps_per_second": 6.745
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/mha_attention/seed_44/trainer_state.json
ADDED
|
@@ -0,0 +1,1245 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 3.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 4212,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.017806267806267807,
|
| 13 |
+
"grad_norm": 6.9620256423950195,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6536,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 9.323328971862793,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5379,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 4.546147346496582,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4885,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 5.7701520919799805,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4559,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 9.602144241333008,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4511,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 5.432075023651123,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4169,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 6.68865966796875,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4407,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 5.416531562805176,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.4057,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 9.961259841918945,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4121,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.873676776885986,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.3977,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 6.663614749908447,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.3895,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 8.31119155883789,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.421,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 6.334585189819336,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3667,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 3.993108034133911,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3829,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 4.278379917144775,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3665,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 4.265623092651367,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.355,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 5.623923301696777,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3552,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 6.305392265319824,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3563,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 9.093064308166504,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.366,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 6.10793924331665,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3483,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 4.681856155395508,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3693,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 6.150272369384766,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3854,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 7.261063575744629,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3258,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 7.214659214019775,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3522,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 6.303726673126221,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3554,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 7.6474289894104,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3318,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 9.655233383178711,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3192,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.9350156784057617,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3101,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.2973129749298096,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3336,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 7.458094596862793,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3233,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.951294898986816,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3408,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 4.144304275512695,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.3052,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 6.475225925445557,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3391,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 5.488035678863525,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3301,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 6.531510829925537,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3299,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.636979103088379,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3132,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 5.925167083740234,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.3318,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 4.34539270401001,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3055,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.328081130981445,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2994,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 3.1705846786499023,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2916,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 4.262259006500244,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2928,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 4.265622615814209,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3075,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 10.959186553955078,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2741,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 6.413283824920654,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.322,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 6.270665168762207,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2774,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.754330635070801,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2883,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 7.586644649505615,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2764,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.870311737060547,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2917,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 6.047013282775879,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2787,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 7.206035137176514,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.2824,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 6.349703788757324,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2767,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 3.9343101978302,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.3051,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 3.447518825531006,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2942,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 5.431354999542236,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2472,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 5.471981048583984,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2663,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 3.556948184967041,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2896,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8612385321100917,
|
| 406 |
+
"eval_loss": 0.33795487880706787,
|
| 407 |
+
"eval_runtime": 3.1827,
|
| 408 |
+
"eval_samples_per_second": 273.981,
|
| 409 |
+
"eval_steps_per_second": 3.142,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 9.849311828613281,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2176,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 7.671362400054932,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2376,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 4.772434234619141,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2212,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 10.426095008850098,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2156,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 4.757193565368652,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2133,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 6.151193618774414,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.1991,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 5.399248123168945,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1922,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 4.966075420379639,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2351,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 4.150966167449951,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.212,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 8.587267875671387,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2208,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 6.242485046386719,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2303,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 5.158636093139648,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2259,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 5.828901767730713,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2257,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 8.660086631774902,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2246,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 2.815744161605835,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2074,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 9.118500709533691,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2178,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 4.281846046447754,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2026,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 5.285156726837158,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2442,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 11.179221153259277,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2039,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 7.265372276306152,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2426,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 8.303732872009277,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2217,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 4.60360860824585,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2089,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 5.297887802124023,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2092,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 6.805469512939453,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.212,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 9.62482738494873,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.191,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 6.134627342224121,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2142,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 5.351399898529053,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.229,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 6.764716148376465,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2104,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 6.884983539581299,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1975,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 4.105142593383789,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2275,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 4.426797389984131,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2185,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 6.119833946228027,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2016,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.444716453552246,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2239,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 6.158670902252197,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.2127,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 6.910562992095947,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2009,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 6.743879795074463,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.1782,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 7.136833190917969,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1942,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 8.863118171691895,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.1709,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 7.0310540199279785,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.2352,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 6.739023208618164,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.2278,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 4.7916765213012695,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.2182,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 5.3655476570129395,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.1938,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 9.274876594543457,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.2115,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 8.410262107849121,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2247,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 4.496730327606201,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.1735,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 6.513155937194824,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1841,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 4.4717607498168945,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2143,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 4.012744426727295,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1856,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 6.495855808258057,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.2097,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 3.6901650428771973,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2047,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 5.471368789672852,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1981,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 10.946513175964355,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.1961,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 3.431741714477539,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1861,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 4.40041971206665,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.2006,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 6.790747165679932,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1935,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 6.478447914123535,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1966,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8807339449541285,
|
| 807 |
+
"eval_loss": 0.35891780257225037,
|
| 808 |
+
"eval_runtime": 3.1957,
|
| 809 |
+
"eval_samples_per_second": 272.863,
|
| 810 |
+
"eval_steps_per_second": 3.129,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 8.029913902282715,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1834,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 4.268023490905762,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1679,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 5.9172868728637695,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1332,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 8.762829780578613,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1571,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 14.087616920471191,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1469,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 8.547751426696777,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1406,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 5.620033264160156,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1334,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 8.453056335449219,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1154,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 12.349413871765137,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1498,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 4.524767875671387,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.144,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 5.222786903381348,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1503,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 7.520452499389648,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1695,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 7.294416904449463,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1424,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 1.5562810897827148,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1166,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 3.418489456176758,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1537,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 11.725346565246582,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1275,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 14.094012260437012,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1506,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 6.143211364746094,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1782,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 6.06160306930542,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.138,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 5.117889881134033,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.152,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 6.67360782623291,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1353,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 5.311966419219971,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1298,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 11.501461029052734,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1163,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 16.54857063293457,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1225,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 3.405914306640625,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1668,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 4.641874313354492,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1415,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 7.1018757820129395,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1492,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 5.46192741394043,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1419,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 0.981749951839447,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1333,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 5.787467956542969,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1198,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 3.8719797134399414,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.141,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 8.804088592529297,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1494,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 9.729549407958984,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1405,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 7.098943710327148,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1244,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 11.16362476348877,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1348,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 5.34449577331543,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1621,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.711578369140625,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1275,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 5.455073833465576,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1129,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 7.163792133331299,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1516,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 8.701996803283691,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1592,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 4.729506492614746,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1371,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 8.890836715698242,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1222,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 7.978331565856934,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1494,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 3.8629729747772217,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1374,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 7.883028507232666,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1413,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 13.028200149536133,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1236,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 7.760869979858398,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1389,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 8.260780334472656,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.173,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 5.601540565490723,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1228,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 8.256863594055176,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1439,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 10.008103370666504,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1671,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 10.075227737426758,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1289,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 7.298343181610107,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.152,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.562656402587891,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1376,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 5.946457862854004,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1222,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 3.9914608001708984,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1271,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8853211009174312,
|
| 1208 |
+
"eval_loss": 0.4112522304058075,
|
| 1209 |
+
"eval_runtime": 2.961,
|
| 1210 |
+
"eval_samples_per_second": 294.496,
|
| 1211 |
+
"eval_steps_per_second": 3.377,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6215986915810056.0,
|
| 1218 |
+
"train_loss": 0.2333670342296843,
|
| 1219 |
+
"train_runtime": 624.4636,
|
| 1220 |
+
"train_samples_per_second": 323.553,
|
| 1221 |
+
"train_steps_per_second": 6.745
|
| 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": 6215986915810056.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8853211009174312,
|
| 4 |
+
"eval_loss": 0.4263913333415985,
|
| 5 |
+
"eval_runtime": 2.9165,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 298.987,
|
| 8 |
+
"eval_steps_per_second": 3.429,
|
| 9 |
+
"total_flos": 6393963493399842.0,
|
| 10 |
+
"train_loss": 0.2324217217996476,
|
| 11 |
+
"train_runtime": 648.2879,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 311.662,
|
| 14 |
+
"train_steps_per_second": 6.497
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8853211009174312,
|
| 4 |
+
"eval_loss": 0.4263913333415985,
|
| 5 |
+
"eval_runtime": 2.9165,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 298.987,
|
| 8 |
+
"eval_steps_per_second": 3.429
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/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/multi_branch_average/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy multi_branch_average --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_42/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6393963493399842.0,
|
| 4 |
+
"train_loss": 0.2324217217996476,
|
| 5 |
+
"train_runtime": 648.2879,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 311.662,
|
| 8 |
+
"train_steps_per_second": 6.497
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/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": 5.8924665451049805,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.647,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 8.485268592834473,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5176,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 6.4866251945495605,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4991,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 5.975020408630371,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.461,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 4.5312042236328125,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4581,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 6.622165203094482,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4591,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 5.263884544372559,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4683,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 7.362464427947998,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.415,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 3.213900327682495,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4104,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 5.839117527008057,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.4148,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 4.816735744476318,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4156,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 6.581177711486816,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4097,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 8.829036712646484,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3751,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.256892681121826,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3794,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 4.939127445220947,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.3817,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 12.047209739685059,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.361,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 6.064167022705078,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3646,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 4.789498329162598,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.366,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 5.079123497009277,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3466,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 3.71764874458313,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3484,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 7.076513767242432,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3533,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 3.876131772994995,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3715,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 9.011445999145508,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3127,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 6.27073335647583,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3261,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.556399345397949,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3553,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 7.501772403717041,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3276,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 3.875983476638794,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3591,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 4.659632205963135,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3234,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 5.709884166717529,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3153,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.813661098480225,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3173,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.98143196105957,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.3215,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 6.447616100311279,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.2935,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 4.6969218254089355,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.306,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.154044151306152,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.3014,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 7.015763282775879,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.2884,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 4.593233585357666,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3448,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 5.459078788757324,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.315,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 2.9056239128112793,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.3091,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 4.611173629760742,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.3091,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 4.450119972229004,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.3297,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 7.834102630615234,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.2797,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 4.3573503494262695,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3206,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 6.222555160522461,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.3097,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.494204998016357,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.2839,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 5.035062313079834,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.297,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 5.046907901763916,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.3041,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 5.374956130981445,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.2977,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 6.085665702819824,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.2762,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 8.097553253173828,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.2632,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 8.386218070983887,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.308,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 7.794201374053955,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.2753,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 5.241933345794678,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2822,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 6.907561779022217,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2592,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 4.91259765625,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.281,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 5.61388635635376,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2999,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 5.212361812591553,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2827,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8577981651376146,
|
| 406 |
+
"eval_loss": 0.3478904068470001,
|
| 407 |
+
"eval_runtime": 2.9411,
|
| 408 |
+
"eval_samples_per_second": 296.487,
|
| 409 |
+
"eval_steps_per_second": 3.4,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 4.56285285949707,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2337,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 6.067489147186279,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2483,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 5.497398376464844,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.235,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 7.88922119140625,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.2136,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 8.635711669921875,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2218,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 11.058414459228516,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.1906,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 6.230376720428467,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.213,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 5.289320945739746,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2012,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 7.985788345336914,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2206,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 7.029101371765137,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2202,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 3.472320079803467,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2335,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 8.023301124572754,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2132,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 5.357924461364746,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2218,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 2.5007548332214355,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2303,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 8.01485824584961,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.2101,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 6.667163372039795,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2352,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 5.184668064117432,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2193,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 4.813374042510986,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2032,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 5.640127182006836,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.1949,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 5.169174671173096,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2306,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 9.30097484588623,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2159,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 5.091939926147461,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.1991,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.185817241668701,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2007,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 3.2896721363067627,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.1966,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 6.968660831451416,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2074,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 6.882242679595947,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2123,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 6.223368167877197,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.2183,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 7.738986492156982,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2018,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 5.209794998168945,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.2012,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 3.8304970264434814,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2084,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 3.2806496620178223,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2295,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 7.997151851654053,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.1851,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 3.336121082305908,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2093,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 10.515104293823242,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1828,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 7.337524890899658,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2099,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 5.8028764724731445,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2324,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 6.693658828735352,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.1928,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 4.134847640991211,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2114,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 4.5934600830078125,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1707,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 5.656535625457764,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.208,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 7.327394485473633,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.1984,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 6.850432395935059,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2088,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 7.333581924438477,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.179,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 8.336596488952637,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2049,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 4.261532783508301,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2195,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 9.994272232055664,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.2104,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 4.472328186035156,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.2129,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 6.777381420135498,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1809,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 5.069293975830078,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.2221,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 7.438899517059326,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.2117,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.407764911651611,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1908,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.688919544219971,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2158,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 5.766843795776367,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.1918,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 7.057423114776611,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.1823,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 3.878042221069336,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1931,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 2.3518967628479004,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.1892,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8692660550458715,
|
| 807 |
+
"eval_loss": 0.40830543637275696,
|
| 808 |
+
"eval_runtime": 2.8059,
|
| 809 |
+
"eval_samples_per_second": 310.771,
|
| 810 |
+
"eval_steps_per_second": 3.564,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 6.957417964935303,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1483,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 2.0357348918914795,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1422,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 3.8646602630615234,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1517,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 4.821736812591553,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1251,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 4.82853364944458,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1574,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 6.615634441375732,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1456,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 2.6775128841400146,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1191,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 4.757756233215332,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.158,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 5.864503383636475,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1537,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 7.461996555328369,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1725,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 7.477771759033203,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1356,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 2.0639901161193848,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1625,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 9.473702430725098,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1456,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 11.831425666809082,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.1224,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 3.913426637649536,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.1583,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 6.232005596160889,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1438,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 10.521530151367188,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.129,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 10.058466911315918,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1494,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 3.1918931007385254,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.1408,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 5.781009197235107,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1221,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 14.475083351135254,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1272,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 5.994709491729736,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1585,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 7.53483772277832,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1491,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 10.771550178527832,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1502,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 2.2808046340942383,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1603,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 3.747490644454956,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1161,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 11.335894584655762,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1272,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 5.483067035675049,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1137,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 6.714562892913818,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1331,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 5.69882869720459,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1485,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 8.331250190734863,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1555,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 4.638311862945557,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1305,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 7.631982326507568,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1263,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 8.239585876464844,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1486,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 8.864460945129395,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1701,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 2.5742175579071045,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1458,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 4.997570514678955,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1592,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 4.747743606567383,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1268,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 2.2288126945495605,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1175,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 5.711939334869385,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1571,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 8.512763023376465,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1225,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 4.146716117858887,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.127,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 6.174679756164551,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1347,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 3.717965602874756,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1477,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 9.744836807250977,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1352,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 5.2534332275390625,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.129,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 3.891486644744873,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1277,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 1.6929874420166016,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1277,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 8.848504066467285,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.1435,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 10.310471534729004,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1236,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 10.908905982971191,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.118,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 3.1794395446777344,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1328,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 8.839540481567383,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1261,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 4.872939586639404,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1293,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 4.965433120727539,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.1182,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 5.680690288543701,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1518,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8853211009174312,
|
| 1208 |
+
"eval_loss": 0.4263913333415985,
|
| 1209 |
+
"eval_runtime": 3.0945,
|
| 1210 |
+
"eval_samples_per_second": 281.79,
|
| 1211 |
+
"eval_steps_per_second": 3.232,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6393963493399842.0,
|
| 1218 |
+
"train_loss": 0.2324217217996476,
|
| 1219 |
+
"train_runtime": 648.2879,
|
| 1220 |
+
"train_samples_per_second": 311.662,
|
| 1221 |
+
"train_steps_per_second": 6.497
|
| 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": 6393963493399842.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/all_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.41880160570144653,
|
| 5 |
+
"eval_runtime": 2.9902,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 291.621,
|
| 8 |
+
"eval_steps_per_second": 3.344,
|
| 9 |
+
"total_flos": 6393601201921362.0,
|
| 10 |
+
"train_loss": 0.23121720937355744,
|
| 11 |
+
"train_runtime": 648.2587,
|
| 12 |
+
"train_samples": 67349,
|
| 13 |
+
"train_samples_per_second": 311.676,
|
| 14 |
+
"train_steps_per_second": 6.497
|
| 15 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/eval_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"eval_accuracy": 0.8784403669724771,
|
| 4 |
+
"eval_loss": 0.41880160570144653,
|
| 5 |
+
"eval_runtime": 2.9902,
|
| 6 |
+
"eval_samples": 872,
|
| 7 |
+
"eval_samples_per_second": 291.621,
|
| 8 |
+
"eval_steps_per_second": 3.344
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/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/multi_branch_average/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy multi_branch_average --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/seed_43/train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 3.0,
|
| 3 |
+
"total_flos": 6393601201921362.0,
|
| 4 |
+
"train_loss": 0.23121720937355744,
|
| 5 |
+
"train_runtime": 648.2587,
|
| 6 |
+
"train_samples": 67349,
|
| 7 |
+
"train_samples_per_second": 311.676,
|
| 8 |
+
"train_steps_per_second": 6.497
|
| 9 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/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": 16.736391067504883,
|
| 14 |
+
"learning_rate": 2.9821937321937322e-05,
|
| 15 |
+
"loss": 0.6136,
|
| 16 |
+
"step": 25
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.03561253561253561,
|
| 20 |
+
"grad_norm": 6.671631813049316,
|
| 21 |
+
"learning_rate": 2.9643874643874644e-05,
|
| 22 |
+
"loss": 0.5263,
|
| 23 |
+
"step": 50
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.053418803418803416,
|
| 27 |
+
"grad_norm": 6.2195329666137695,
|
| 28 |
+
"learning_rate": 2.9465811965811966e-05,
|
| 29 |
+
"loss": 0.4776,
|
| 30 |
+
"step": 75
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.07122507122507123,
|
| 34 |
+
"grad_norm": 7.474128246307373,
|
| 35 |
+
"learning_rate": 2.9287749287749288e-05,
|
| 36 |
+
"loss": 0.4507,
|
| 37 |
+
"step": 100
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.08903133903133903,
|
| 41 |
+
"grad_norm": 5.316859245300293,
|
| 42 |
+
"learning_rate": 2.910968660968661e-05,
|
| 43 |
+
"loss": 0.4822,
|
| 44 |
+
"step": 125
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.10683760683760683,
|
| 48 |
+
"grad_norm": 15.504117965698242,
|
| 49 |
+
"learning_rate": 2.893162393162393e-05,
|
| 50 |
+
"loss": 0.4638,
|
| 51 |
+
"step": 150
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.12464387464387465,
|
| 55 |
+
"grad_norm": 6.118660926818848,
|
| 56 |
+
"learning_rate": 2.8753561253561256e-05,
|
| 57 |
+
"loss": 0.4146,
|
| 58 |
+
"step": 175
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.14245014245014245,
|
| 62 |
+
"grad_norm": 9.012158393859863,
|
| 63 |
+
"learning_rate": 2.8575498575498578e-05,
|
| 64 |
+
"loss": 0.3984,
|
| 65 |
+
"step": 200
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.16025641025641027,
|
| 69 |
+
"grad_norm": 5.647778511047363,
|
| 70 |
+
"learning_rate": 2.8397435897435896e-05,
|
| 71 |
+
"loss": 0.4359,
|
| 72 |
+
"step": 225
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.17806267806267806,
|
| 76 |
+
"grad_norm": 4.967792510986328,
|
| 77 |
+
"learning_rate": 2.821937321937322e-05,
|
| 78 |
+
"loss": 0.3888,
|
| 79 |
+
"step": 250
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.19586894586894588,
|
| 83 |
+
"grad_norm": 6.874669075012207,
|
| 84 |
+
"learning_rate": 2.8041310541310543e-05,
|
| 85 |
+
"loss": 0.4202,
|
| 86 |
+
"step": 275
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.21367521367521367,
|
| 90 |
+
"grad_norm": 5.062214374542236,
|
| 91 |
+
"learning_rate": 2.786324786324786e-05,
|
| 92 |
+
"loss": 0.4094,
|
| 93 |
+
"step": 300
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.23148148148148148,
|
| 97 |
+
"grad_norm": 5.659512042999268,
|
| 98 |
+
"learning_rate": 2.7685185185185186e-05,
|
| 99 |
+
"loss": 0.3979,
|
| 100 |
+
"step": 325
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.2492877492877493,
|
| 104 |
+
"grad_norm": 5.441919803619385,
|
| 105 |
+
"learning_rate": 2.7507122507122508e-05,
|
| 106 |
+
"loss": 0.3839,
|
| 107 |
+
"step": 350
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.2670940170940171,
|
| 111 |
+
"grad_norm": 11.517582893371582,
|
| 112 |
+
"learning_rate": 2.732905982905983e-05,
|
| 113 |
+
"loss": 0.374,
|
| 114 |
+
"step": 375
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.2849002849002849,
|
| 118 |
+
"grad_norm": 6.858891487121582,
|
| 119 |
+
"learning_rate": 2.715099715099715e-05,
|
| 120 |
+
"loss": 0.4214,
|
| 121 |
+
"step": 400
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.3027065527065527,
|
| 125 |
+
"grad_norm": 6.682157516479492,
|
| 126 |
+
"learning_rate": 2.6972934472934473e-05,
|
| 127 |
+
"loss": 0.3792,
|
| 128 |
+
"step": 425
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.32051282051282054,
|
| 132 |
+
"grad_norm": 5.94612455368042,
|
| 133 |
+
"learning_rate": 2.67948717948718e-05,
|
| 134 |
+
"loss": 0.3661,
|
| 135 |
+
"step": 450
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.3383190883190883,
|
| 139 |
+
"grad_norm": 7.416209697723389,
|
| 140 |
+
"learning_rate": 2.6616809116809117e-05,
|
| 141 |
+
"loss": 0.3608,
|
| 142 |
+
"step": 475
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.3561253561253561,
|
| 146 |
+
"grad_norm": 4.957077503204346,
|
| 147 |
+
"learning_rate": 2.643874643874644e-05,
|
| 148 |
+
"loss": 0.3341,
|
| 149 |
+
"step": 500
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.37393162393162394,
|
| 153 |
+
"grad_norm": 8.913534164428711,
|
| 154 |
+
"learning_rate": 2.6260683760683763e-05,
|
| 155 |
+
"loss": 0.3489,
|
| 156 |
+
"step": 525
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.39173789173789175,
|
| 160 |
+
"grad_norm": 5.566319465637207,
|
| 161 |
+
"learning_rate": 2.6082621082621082e-05,
|
| 162 |
+
"loss": 0.3493,
|
| 163 |
+
"step": 550
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.40954415954415957,
|
| 167 |
+
"grad_norm": 3.7891368865966797,
|
| 168 |
+
"learning_rate": 2.5904558404558404e-05,
|
| 169 |
+
"loss": 0.3285,
|
| 170 |
+
"step": 575
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.42735042735042733,
|
| 174 |
+
"grad_norm": 4.348255634307861,
|
| 175 |
+
"learning_rate": 2.572649572649573e-05,
|
| 176 |
+
"loss": 0.3333,
|
| 177 |
+
"step": 600
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.44515669515669515,
|
| 181 |
+
"grad_norm": 5.046405792236328,
|
| 182 |
+
"learning_rate": 2.5548433048433047e-05,
|
| 183 |
+
"loss": 0.3453,
|
| 184 |
+
"step": 625
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.46296296296296297,
|
| 188 |
+
"grad_norm": 5.334004878997803,
|
| 189 |
+
"learning_rate": 2.5370370370370372e-05,
|
| 190 |
+
"loss": 0.3376,
|
| 191 |
+
"step": 650
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.4807692307692308,
|
| 195 |
+
"grad_norm": 6.804669380187988,
|
| 196 |
+
"learning_rate": 2.5192307692307694e-05,
|
| 197 |
+
"loss": 0.3361,
|
| 198 |
+
"step": 675
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.4985754985754986,
|
| 202 |
+
"grad_norm": 3.1799235343933105,
|
| 203 |
+
"learning_rate": 2.5014245014245015e-05,
|
| 204 |
+
"loss": 0.3332,
|
| 205 |
+
"step": 700
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.5163817663817664,
|
| 209 |
+
"grad_norm": 3.1536426544189453,
|
| 210 |
+
"learning_rate": 2.4836182336182337e-05,
|
| 211 |
+
"loss": 0.3159,
|
| 212 |
+
"step": 725
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.5341880341880342,
|
| 216 |
+
"grad_norm": 5.099383354187012,
|
| 217 |
+
"learning_rate": 2.465811965811966e-05,
|
| 218 |
+
"loss": 0.3536,
|
| 219 |
+
"step": 750
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.551994301994302,
|
| 223 |
+
"grad_norm": 5.8598175048828125,
|
| 224 |
+
"learning_rate": 2.448005698005698e-05,
|
| 225 |
+
"loss": 0.328,
|
| 226 |
+
"step": 775
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.5698005698005698,
|
| 230 |
+
"grad_norm": 6.722732067108154,
|
| 231 |
+
"learning_rate": 2.4301994301994302e-05,
|
| 232 |
+
"loss": 0.324,
|
| 233 |
+
"step": 800
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.5876068376068376,
|
| 237 |
+
"grad_norm": 4.78679895401001,
|
| 238 |
+
"learning_rate": 2.4123931623931624e-05,
|
| 239 |
+
"loss": 0.3044,
|
| 240 |
+
"step": 825
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.6054131054131054,
|
| 244 |
+
"grad_norm": 6.189655780792236,
|
| 245 |
+
"learning_rate": 2.3945868945868946e-05,
|
| 246 |
+
"loss": 0.2789,
|
| 247 |
+
"step": 850
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.6232193732193733,
|
| 251 |
+
"grad_norm": 7.180893421173096,
|
| 252 |
+
"learning_rate": 2.3767806267806267e-05,
|
| 253 |
+
"loss": 0.3604,
|
| 254 |
+
"step": 875
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.6410256410256411,
|
| 258 |
+
"grad_norm": 5.730825901031494,
|
| 259 |
+
"learning_rate": 2.358974358974359e-05,
|
| 260 |
+
"loss": 0.3383,
|
| 261 |
+
"step": 900
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.6588319088319088,
|
| 265 |
+
"grad_norm": 4.6597981452941895,
|
| 266 |
+
"learning_rate": 2.3411680911680914e-05,
|
| 267 |
+
"loss": 0.2996,
|
| 268 |
+
"step": 925
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.6766381766381766,
|
| 272 |
+
"grad_norm": 13.028176307678223,
|
| 273 |
+
"learning_rate": 2.3233618233618236e-05,
|
| 274 |
+
"loss": 0.2826,
|
| 275 |
+
"step": 950
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.6944444444444444,
|
| 279 |
+
"grad_norm": 5.575206756591797,
|
| 280 |
+
"learning_rate": 2.3055555555555554e-05,
|
| 281 |
+
"loss": 0.2587,
|
| 282 |
+
"step": 975
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.7122507122507122,
|
| 286 |
+
"grad_norm": 6.683717250823975,
|
| 287 |
+
"learning_rate": 2.287749287749288e-05,
|
| 288 |
+
"loss": 0.2893,
|
| 289 |
+
"step": 1000
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.73005698005698,
|
| 293 |
+
"grad_norm": 6.035584926605225,
|
| 294 |
+
"learning_rate": 2.26994301994302e-05,
|
| 295 |
+
"loss": 0.302,
|
| 296 |
+
"step": 1025
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.7478632478632479,
|
| 300 |
+
"grad_norm": 3.563339948654175,
|
| 301 |
+
"learning_rate": 2.252136752136752e-05,
|
| 302 |
+
"loss": 0.3098,
|
| 303 |
+
"step": 1050
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.7656695156695157,
|
| 307 |
+
"grad_norm": 5.066797733306885,
|
| 308 |
+
"learning_rate": 2.2343304843304845e-05,
|
| 309 |
+
"loss": 0.2638,
|
| 310 |
+
"step": 1075
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.7834757834757835,
|
| 314 |
+
"grad_norm": 5.174193382263184,
|
| 315 |
+
"learning_rate": 2.2165242165242166e-05,
|
| 316 |
+
"loss": 0.3084,
|
| 317 |
+
"step": 1100
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.8012820512820513,
|
| 321 |
+
"grad_norm": 9.498007774353027,
|
| 322 |
+
"learning_rate": 2.1987179487179485e-05,
|
| 323 |
+
"loss": 0.2898,
|
| 324 |
+
"step": 1125
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.8190883190883191,
|
| 328 |
+
"grad_norm": 8.735760688781738,
|
| 329 |
+
"learning_rate": 2.180911680911681e-05,
|
| 330 |
+
"loss": 0.2895,
|
| 331 |
+
"step": 1150
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.8368945868945868,
|
| 335 |
+
"grad_norm": 7.140743732452393,
|
| 336 |
+
"learning_rate": 2.163105413105413e-05,
|
| 337 |
+
"loss": 0.313,
|
| 338 |
+
"step": 1175
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.8547008547008547,
|
| 342 |
+
"grad_norm": 3.3386270999908447,
|
| 343 |
+
"learning_rate": 2.1452991452991456e-05,
|
| 344 |
+
"loss": 0.267,
|
| 345 |
+
"step": 1200
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.8725071225071225,
|
| 349 |
+
"grad_norm": 4.888571739196777,
|
| 350 |
+
"learning_rate": 2.1274928774928775e-05,
|
| 351 |
+
"loss": 0.3097,
|
| 352 |
+
"step": 1225
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.8903133903133903,
|
| 356 |
+
"grad_norm": 5.41079568862915,
|
| 357 |
+
"learning_rate": 2.1096866096866097e-05,
|
| 358 |
+
"loss": 0.3268,
|
| 359 |
+
"step": 1250
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.9081196581196581,
|
| 363 |
+
"grad_norm": 5.11181116104126,
|
| 364 |
+
"learning_rate": 2.091880341880342e-05,
|
| 365 |
+
"loss": 0.3011,
|
| 366 |
+
"step": 1275
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.9259259259259259,
|
| 370 |
+
"grad_norm": 8.453507423400879,
|
| 371 |
+
"learning_rate": 2.074074074074074e-05,
|
| 372 |
+
"loss": 0.2868,
|
| 373 |
+
"step": 1300
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.9437321937321937,
|
| 377 |
+
"grad_norm": 5.5320539474487305,
|
| 378 |
+
"learning_rate": 2.056267806267806e-05,
|
| 379 |
+
"loss": 0.2627,
|
| 380 |
+
"step": 1325
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.9615384615384616,
|
| 384 |
+
"grad_norm": 7.705092906951904,
|
| 385 |
+
"learning_rate": 2.0384615384615387e-05,
|
| 386 |
+
"loss": 0.2535,
|
| 387 |
+
"step": 1350
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.9793447293447294,
|
| 391 |
+
"grad_norm": 5.041247367858887,
|
| 392 |
+
"learning_rate": 2.0206552706552705e-05,
|
| 393 |
+
"loss": 0.2882,
|
| 394 |
+
"step": 1375
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.9971509971509972,
|
| 398 |
+
"grad_norm": 6.706636428833008,
|
| 399 |
+
"learning_rate": 2.002849002849003e-05,
|
| 400 |
+
"loss": 0.2656,
|
| 401 |
+
"step": 1400
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 1.0,
|
| 405 |
+
"eval_accuracy": 0.8486238532110092,
|
| 406 |
+
"eval_loss": 0.38648363947868347,
|
| 407 |
+
"eval_runtime": 3.1485,
|
| 408 |
+
"eval_samples_per_second": 276.956,
|
| 409 |
+
"eval_steps_per_second": 3.176,
|
| 410 |
+
"step": 1404
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 1.014957264957265,
|
| 414 |
+
"grad_norm": 3.5597360134124756,
|
| 415 |
+
"learning_rate": 1.9850427350427352e-05,
|
| 416 |
+
"loss": 0.2069,
|
| 417 |
+
"step": 1425
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 1.0327635327635327,
|
| 421 |
+
"grad_norm": 5.004386901855469,
|
| 422 |
+
"learning_rate": 1.9672364672364674e-05,
|
| 423 |
+
"loss": 0.2326,
|
| 424 |
+
"step": 1450
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 1.0505698005698005,
|
| 428 |
+
"grad_norm": 5.548844337463379,
|
| 429 |
+
"learning_rate": 1.9494301994301995e-05,
|
| 430 |
+
"loss": 0.2152,
|
| 431 |
+
"step": 1475
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 1.0683760683760684,
|
| 435 |
+
"grad_norm": 7.057421684265137,
|
| 436 |
+
"learning_rate": 1.9316239316239317e-05,
|
| 437 |
+
"loss": 0.23,
|
| 438 |
+
"step": 1500
|
| 439 |
+
},
|
| 440 |
+
{
|
| 441 |
+
"epoch": 1.0861823361823362,
|
| 442 |
+
"grad_norm": 3.780733346939087,
|
| 443 |
+
"learning_rate": 1.913817663817664e-05,
|
| 444 |
+
"loss": 0.2181,
|
| 445 |
+
"step": 1525
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 1.103988603988604,
|
| 449 |
+
"grad_norm": 6.3818559646606445,
|
| 450 |
+
"learning_rate": 1.896011396011396e-05,
|
| 451 |
+
"loss": 0.2305,
|
| 452 |
+
"step": 1550
|
| 453 |
+
},
|
| 454 |
+
{
|
| 455 |
+
"epoch": 1.1217948717948718,
|
| 456 |
+
"grad_norm": 3.531998634338379,
|
| 457 |
+
"learning_rate": 1.8782051282051282e-05,
|
| 458 |
+
"loss": 0.1993,
|
| 459 |
+
"step": 1575
|
| 460 |
+
},
|
| 461 |
+
{
|
| 462 |
+
"epoch": 1.1396011396011396,
|
| 463 |
+
"grad_norm": 10.046457290649414,
|
| 464 |
+
"learning_rate": 1.8603988603988604e-05,
|
| 465 |
+
"loss": 0.2385,
|
| 466 |
+
"step": 1600
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 1.1574074074074074,
|
| 470 |
+
"grad_norm": 5.601193904876709,
|
| 471 |
+
"learning_rate": 1.8425925925925926e-05,
|
| 472 |
+
"loss": 0.2037,
|
| 473 |
+
"step": 1625
|
| 474 |
+
},
|
| 475 |
+
{
|
| 476 |
+
"epoch": 1.1752136752136753,
|
| 477 |
+
"grad_norm": 5.510344982147217,
|
| 478 |
+
"learning_rate": 1.8247863247863247e-05,
|
| 479 |
+
"loss": 0.2358,
|
| 480 |
+
"step": 1650
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"epoch": 1.193019943019943,
|
| 484 |
+
"grad_norm": 8.112565994262695,
|
| 485 |
+
"learning_rate": 1.8069800569800572e-05,
|
| 486 |
+
"loss": 0.2177,
|
| 487 |
+
"step": 1675
|
| 488 |
+
},
|
| 489 |
+
{
|
| 490 |
+
"epoch": 1.210826210826211,
|
| 491 |
+
"grad_norm": 5.572713375091553,
|
| 492 |
+
"learning_rate": 1.7891737891737894e-05,
|
| 493 |
+
"loss": 0.2435,
|
| 494 |
+
"step": 1700
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 1.2286324786324787,
|
| 498 |
+
"grad_norm": 7.74533224105835,
|
| 499 |
+
"learning_rate": 1.7713675213675212e-05,
|
| 500 |
+
"loss": 0.2146,
|
| 501 |
+
"step": 1725
|
| 502 |
+
},
|
| 503 |
+
{
|
| 504 |
+
"epoch": 1.2464387464387465,
|
| 505 |
+
"grad_norm": 5.939334392547607,
|
| 506 |
+
"learning_rate": 1.7535612535612538e-05,
|
| 507 |
+
"loss": 0.2167,
|
| 508 |
+
"step": 1750
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"epoch": 1.2642450142450143,
|
| 512 |
+
"grad_norm": 5.052703380584717,
|
| 513 |
+
"learning_rate": 1.735754985754986e-05,
|
| 514 |
+
"loss": 0.215,
|
| 515 |
+
"step": 1775
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 1.282051282051282,
|
| 519 |
+
"grad_norm": 4.056268215179443,
|
| 520 |
+
"learning_rate": 1.7179487179487178e-05,
|
| 521 |
+
"loss": 0.2118,
|
| 522 |
+
"step": 1800
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 1.29985754985755,
|
| 526 |
+
"grad_norm": 10.275428771972656,
|
| 527 |
+
"learning_rate": 1.7001424501424503e-05,
|
| 528 |
+
"loss": 0.2222,
|
| 529 |
+
"step": 1825
|
| 530 |
+
},
|
| 531 |
+
{
|
| 532 |
+
"epoch": 1.3176638176638176,
|
| 533 |
+
"grad_norm": 6.014178276062012,
|
| 534 |
+
"learning_rate": 1.6823361823361824e-05,
|
| 535 |
+
"loss": 0.2172,
|
| 536 |
+
"step": 1850
|
| 537 |
+
},
|
| 538 |
+
{
|
| 539 |
+
"epoch": 1.3354700854700854,
|
| 540 |
+
"grad_norm": 3.487736463546753,
|
| 541 |
+
"learning_rate": 1.6645299145299143e-05,
|
| 542 |
+
"loss": 0.2251,
|
| 543 |
+
"step": 1875
|
| 544 |
+
},
|
| 545 |
+
{
|
| 546 |
+
"epoch": 1.3532763532763532,
|
| 547 |
+
"grad_norm": 7.386679172515869,
|
| 548 |
+
"learning_rate": 1.6467236467236468e-05,
|
| 549 |
+
"loss": 0.2086,
|
| 550 |
+
"step": 1900
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 1.371082621082621,
|
| 554 |
+
"grad_norm": 4.3145222663879395,
|
| 555 |
+
"learning_rate": 1.628917378917379e-05,
|
| 556 |
+
"loss": 0.2027,
|
| 557 |
+
"step": 1925
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 1.3888888888888888,
|
| 561 |
+
"grad_norm": 5.986841201782227,
|
| 562 |
+
"learning_rate": 1.6111111111111115e-05,
|
| 563 |
+
"loss": 0.2287,
|
| 564 |
+
"step": 1950
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 1.4066951566951567,
|
| 568 |
+
"grad_norm": 4.888129234313965,
|
| 569 |
+
"learning_rate": 1.5933048433048433e-05,
|
| 570 |
+
"loss": 0.2299,
|
| 571 |
+
"step": 1975
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 1.4245014245014245,
|
| 575 |
+
"grad_norm": 3.7574594020843506,
|
| 576 |
+
"learning_rate": 1.5754985754985755e-05,
|
| 577 |
+
"loss": 0.2021,
|
| 578 |
+
"step": 2000
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 1.4423076923076923,
|
| 582 |
+
"grad_norm": 6.172779083251953,
|
| 583 |
+
"learning_rate": 1.557692307692308e-05,
|
| 584 |
+
"loss": 0.2164,
|
| 585 |
+
"step": 2025
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 1.46011396011396,
|
| 589 |
+
"grad_norm": 7.986530780792236,
|
| 590 |
+
"learning_rate": 1.5398860398860398e-05,
|
| 591 |
+
"loss": 0.2116,
|
| 592 |
+
"step": 2050
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 1.477920227920228,
|
| 596 |
+
"grad_norm": 5.937182426452637,
|
| 597 |
+
"learning_rate": 1.5220797720797721e-05,
|
| 598 |
+
"loss": 0.1824,
|
| 599 |
+
"step": 2075
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 1.4957264957264957,
|
| 603 |
+
"grad_norm": 4.384931564331055,
|
| 604 |
+
"learning_rate": 1.5042735042735045e-05,
|
| 605 |
+
"loss": 0.2037,
|
| 606 |
+
"step": 2100
|
| 607 |
+
},
|
| 608 |
+
{
|
| 609 |
+
"epoch": 1.5135327635327636,
|
| 610 |
+
"grad_norm": 8.077068328857422,
|
| 611 |
+
"learning_rate": 1.4864672364672365e-05,
|
| 612 |
+
"loss": 0.1887,
|
| 613 |
+
"step": 2125
|
| 614 |
+
},
|
| 615 |
+
{
|
| 616 |
+
"epoch": 1.5313390313390314,
|
| 617 |
+
"grad_norm": 4.132740020751953,
|
| 618 |
+
"learning_rate": 1.4686609686609687e-05,
|
| 619 |
+
"loss": 0.2175,
|
| 620 |
+
"step": 2150
|
| 621 |
+
},
|
| 622 |
+
{
|
| 623 |
+
"epoch": 1.5491452991452992,
|
| 624 |
+
"grad_norm": 9.541309356689453,
|
| 625 |
+
"learning_rate": 1.450854700854701e-05,
|
| 626 |
+
"loss": 0.2218,
|
| 627 |
+
"step": 2175
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 1.566951566951567,
|
| 631 |
+
"grad_norm": 5.1092939376831055,
|
| 632 |
+
"learning_rate": 1.433048433048433e-05,
|
| 633 |
+
"loss": 0.2048,
|
| 634 |
+
"step": 2200
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 1.5847578347578346,
|
| 638 |
+
"grad_norm": 7.195929527282715,
|
| 639 |
+
"learning_rate": 1.4152421652421652e-05,
|
| 640 |
+
"loss": 0.2083,
|
| 641 |
+
"step": 2225
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 1.6025641025641026,
|
| 645 |
+
"grad_norm": 5.269391059875488,
|
| 646 |
+
"learning_rate": 1.3974358974358975e-05,
|
| 647 |
+
"loss": 0.1961,
|
| 648 |
+
"step": 2250
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 1.6203703703703702,
|
| 652 |
+
"grad_norm": 6.70665168762207,
|
| 653 |
+
"learning_rate": 1.3796296296296297e-05,
|
| 654 |
+
"loss": 0.2357,
|
| 655 |
+
"step": 2275
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 1.6381766381766383,
|
| 659 |
+
"grad_norm": 3.78647780418396,
|
| 660 |
+
"learning_rate": 1.3618233618233619e-05,
|
| 661 |
+
"loss": 0.2007,
|
| 662 |
+
"step": 2300
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 1.6559829059829059,
|
| 666 |
+
"grad_norm": 7.052969455718994,
|
| 667 |
+
"learning_rate": 1.344017094017094e-05,
|
| 668 |
+
"loss": 0.2308,
|
| 669 |
+
"step": 2325
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 1.673789173789174,
|
| 673 |
+
"grad_norm": 4.048561096191406,
|
| 674 |
+
"learning_rate": 1.3262108262108262e-05,
|
| 675 |
+
"loss": 0.2023,
|
| 676 |
+
"step": 2350
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 1.6915954415954415,
|
| 680 |
+
"grad_norm": 10.927449226379395,
|
| 681 |
+
"learning_rate": 1.3084045584045585e-05,
|
| 682 |
+
"loss": 0.1942,
|
| 683 |
+
"step": 2375
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 1.7094017094017095,
|
| 687 |
+
"grad_norm": 4.710274696350098,
|
| 688 |
+
"learning_rate": 1.2905982905982905e-05,
|
| 689 |
+
"loss": 0.1966,
|
| 690 |
+
"step": 2400
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 1.7272079772079771,
|
| 694 |
+
"grad_norm": 6.143375396728516,
|
| 695 |
+
"learning_rate": 1.2727920227920229e-05,
|
| 696 |
+
"loss": 0.1847,
|
| 697 |
+
"step": 2425
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 1.7450142450142452,
|
| 701 |
+
"grad_norm": 8.144155502319336,
|
| 702 |
+
"learning_rate": 1.254985754985755e-05,
|
| 703 |
+
"loss": 0.2059,
|
| 704 |
+
"step": 2450
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 1.7628205128205128,
|
| 708 |
+
"grad_norm": 5.36845588684082,
|
| 709 |
+
"learning_rate": 1.2371794871794872e-05,
|
| 710 |
+
"loss": 0.1913,
|
| 711 |
+
"step": 2475
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"epoch": 1.7806267806267806,
|
| 715 |
+
"grad_norm": 9.95815372467041,
|
| 716 |
+
"learning_rate": 1.2193732193732194e-05,
|
| 717 |
+
"loss": 0.2155,
|
| 718 |
+
"step": 2500
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 1.7984330484330484,
|
| 722 |
+
"grad_norm": 6.444311618804932,
|
| 723 |
+
"learning_rate": 1.2015669515669516e-05,
|
| 724 |
+
"loss": 0.2089,
|
| 725 |
+
"step": 2525
|
| 726 |
+
},
|
| 727 |
+
{
|
| 728 |
+
"epoch": 1.8162393162393162,
|
| 729 |
+
"grad_norm": 4.8697075843811035,
|
| 730 |
+
"learning_rate": 1.1837606837606839e-05,
|
| 731 |
+
"loss": 0.1923,
|
| 732 |
+
"step": 2550
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"epoch": 1.834045584045584,
|
| 736 |
+
"grad_norm": 5.357026100158691,
|
| 737 |
+
"learning_rate": 1.1659544159544159e-05,
|
| 738 |
+
"loss": 0.1983,
|
| 739 |
+
"step": 2575
|
| 740 |
+
},
|
| 741 |
+
{
|
| 742 |
+
"epoch": 1.8518518518518519,
|
| 743 |
+
"grad_norm": 8.485574722290039,
|
| 744 |
+
"learning_rate": 1.148148148148148e-05,
|
| 745 |
+
"loss": 0.1928,
|
| 746 |
+
"step": 2600
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 1.8696581196581197,
|
| 750 |
+
"grad_norm": 3.8532121181488037,
|
| 751 |
+
"learning_rate": 1.1303418803418804e-05,
|
| 752 |
+
"loss": 0.1746,
|
| 753 |
+
"step": 2625
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 1.8874643874643875,
|
| 757 |
+
"grad_norm": 7.614182472229004,
|
| 758 |
+
"learning_rate": 1.1125356125356126e-05,
|
| 759 |
+
"loss": 0.1886,
|
| 760 |
+
"step": 2650
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 1.9052706552706553,
|
| 764 |
+
"grad_norm": 6.896753787994385,
|
| 765 |
+
"learning_rate": 1.0947293447293448e-05,
|
| 766 |
+
"loss": 0.1843,
|
| 767 |
+
"step": 2675
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 1.9230769230769231,
|
| 771 |
+
"grad_norm": 5.537007808685303,
|
| 772 |
+
"learning_rate": 1.076923076923077e-05,
|
| 773 |
+
"loss": 0.2027,
|
| 774 |
+
"step": 2700
|
| 775 |
+
},
|
| 776 |
+
{
|
| 777 |
+
"epoch": 1.940883190883191,
|
| 778 |
+
"grad_norm": 7.812449932098389,
|
| 779 |
+
"learning_rate": 1.0591168091168091e-05,
|
| 780 |
+
"loss": 0.2129,
|
| 781 |
+
"step": 2725
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"epoch": 1.9586894586894585,
|
| 785 |
+
"grad_norm": 5.226696491241455,
|
| 786 |
+
"learning_rate": 1.0413105413105414e-05,
|
| 787 |
+
"loss": 0.194,
|
| 788 |
+
"step": 2750
|
| 789 |
+
},
|
| 790 |
+
{
|
| 791 |
+
"epoch": 1.9764957264957266,
|
| 792 |
+
"grad_norm": 8.911869049072266,
|
| 793 |
+
"learning_rate": 1.0235042735042734e-05,
|
| 794 |
+
"loss": 0.1884,
|
| 795 |
+
"step": 2775
|
| 796 |
+
},
|
| 797 |
+
{
|
| 798 |
+
"epoch": 1.9943019943019942,
|
| 799 |
+
"grad_norm": 10.811223030090332,
|
| 800 |
+
"learning_rate": 1.0056980056980058e-05,
|
| 801 |
+
"loss": 0.189,
|
| 802 |
+
"step": 2800
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 2.0,
|
| 806 |
+
"eval_accuracy": 0.8715596330275229,
|
| 807 |
+
"eval_loss": 0.38384005427360535,
|
| 808 |
+
"eval_runtime": 2.8963,
|
| 809 |
+
"eval_samples_per_second": 301.074,
|
| 810 |
+
"eval_steps_per_second": 3.453,
|
| 811 |
+
"step": 2808
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 2.012108262108262,
|
| 815 |
+
"grad_norm": 9.212801933288574,
|
| 816 |
+
"learning_rate": 9.87891737891738e-06,
|
| 817 |
+
"loss": 0.1314,
|
| 818 |
+
"step": 2825
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 2.02991452991453,
|
| 822 |
+
"grad_norm": 4.87127161026001,
|
| 823 |
+
"learning_rate": 9.700854700854701e-06,
|
| 824 |
+
"loss": 0.1448,
|
| 825 |
+
"step": 2850
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 2.047720797720798,
|
| 829 |
+
"grad_norm": 1.8403747081756592,
|
| 830 |
+
"learning_rate": 9.522792022792023e-06,
|
| 831 |
+
"loss": 0.1212,
|
| 832 |
+
"step": 2875
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 2.0655270655270654,
|
| 836 |
+
"grad_norm": 6.479730606079102,
|
| 837 |
+
"learning_rate": 9.344729344729345e-06,
|
| 838 |
+
"loss": 0.1383,
|
| 839 |
+
"step": 2900
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 2.0833333333333335,
|
| 843 |
+
"grad_norm": 6.3119049072265625,
|
| 844 |
+
"learning_rate": 9.166666666666668e-06,
|
| 845 |
+
"loss": 0.1423,
|
| 846 |
+
"step": 2925
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 2.101139601139601,
|
| 850 |
+
"grad_norm": 4.698959827423096,
|
| 851 |
+
"learning_rate": 8.988603988603988e-06,
|
| 852 |
+
"loss": 0.1272,
|
| 853 |
+
"step": 2950
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 2.118945868945869,
|
| 857 |
+
"grad_norm": 9.401551246643066,
|
| 858 |
+
"learning_rate": 8.81054131054131e-06,
|
| 859 |
+
"loss": 0.1253,
|
| 860 |
+
"step": 2975
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 2.1367521367521367,
|
| 864 |
+
"grad_norm": 6.8200249671936035,
|
| 865 |
+
"learning_rate": 8.632478632478633e-06,
|
| 866 |
+
"loss": 0.1392,
|
| 867 |
+
"step": 3000
|
| 868 |
+
},
|
| 869 |
+
{
|
| 870 |
+
"epoch": 2.1545584045584047,
|
| 871 |
+
"grad_norm": 10.191108703613281,
|
| 872 |
+
"learning_rate": 8.454415954415955e-06,
|
| 873 |
+
"loss": 0.1408,
|
| 874 |
+
"step": 3025
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 2.1723646723646723,
|
| 878 |
+
"grad_norm": 5.3459367752075195,
|
| 879 |
+
"learning_rate": 8.276353276353277e-06,
|
| 880 |
+
"loss": 0.1343,
|
| 881 |
+
"step": 3050
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 2.1901709401709404,
|
| 885 |
+
"grad_norm": 8.195032119750977,
|
| 886 |
+
"learning_rate": 8.098290598290598e-06,
|
| 887 |
+
"loss": 0.1524,
|
| 888 |
+
"step": 3075
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 2.207977207977208,
|
| 892 |
+
"grad_norm": 9.373213768005371,
|
| 893 |
+
"learning_rate": 7.92022792022792e-06,
|
| 894 |
+
"loss": 0.1545,
|
| 895 |
+
"step": 3100
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 2.2257834757834756,
|
| 899 |
+
"grad_norm": 5.322947025299072,
|
| 900 |
+
"learning_rate": 7.742165242165244e-06,
|
| 901 |
+
"loss": 0.1478,
|
| 902 |
+
"step": 3125
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 2.2435897435897436,
|
| 906 |
+
"grad_norm": 4.165241241455078,
|
| 907 |
+
"learning_rate": 7.564102564102564e-06,
|
| 908 |
+
"loss": 0.123,
|
| 909 |
+
"step": 3150
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 2.261396011396011,
|
| 913 |
+
"grad_norm": 11.805953979492188,
|
| 914 |
+
"learning_rate": 7.386039886039886e-06,
|
| 915 |
+
"loss": 0.143,
|
| 916 |
+
"step": 3175
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 2.2792022792022792,
|
| 920 |
+
"grad_norm": 5.698513031005859,
|
| 921 |
+
"learning_rate": 7.207977207977209e-06,
|
| 922 |
+
"loss": 0.1321,
|
| 923 |
+
"step": 3200
|
| 924 |
+
},
|
| 925 |
+
{
|
| 926 |
+
"epoch": 2.297008547008547,
|
| 927 |
+
"grad_norm": 2.4196181297302246,
|
| 928 |
+
"learning_rate": 7.02991452991453e-06,
|
| 929 |
+
"loss": 0.1264,
|
| 930 |
+
"step": 3225
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"epoch": 2.314814814814815,
|
| 934 |
+
"grad_norm": 7.092799663543701,
|
| 935 |
+
"learning_rate": 6.851851851851852e-06,
|
| 936 |
+
"loss": 0.1601,
|
| 937 |
+
"step": 3250
|
| 938 |
+
},
|
| 939 |
+
{
|
| 940 |
+
"epoch": 2.3326210826210825,
|
| 941 |
+
"grad_norm": 13.718725204467773,
|
| 942 |
+
"learning_rate": 6.673789173789174e-06,
|
| 943 |
+
"loss": 0.147,
|
| 944 |
+
"step": 3275
|
| 945 |
+
},
|
| 946 |
+
{
|
| 947 |
+
"epoch": 2.3504273504273505,
|
| 948 |
+
"grad_norm": 6.618736267089844,
|
| 949 |
+
"learning_rate": 6.4957264957264955e-06,
|
| 950 |
+
"loss": 0.1521,
|
| 951 |
+
"step": 3300
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"epoch": 2.368233618233618,
|
| 955 |
+
"grad_norm": 13.063075065612793,
|
| 956 |
+
"learning_rate": 6.317663817663818e-06,
|
| 957 |
+
"loss": 0.1427,
|
| 958 |
+
"step": 3325
|
| 959 |
+
},
|
| 960 |
+
{
|
| 961 |
+
"epoch": 2.386039886039886,
|
| 962 |
+
"grad_norm": 8.520234107971191,
|
| 963 |
+
"learning_rate": 6.13960113960114e-06,
|
| 964 |
+
"loss": 0.1322,
|
| 965 |
+
"step": 3350
|
| 966 |
+
},
|
| 967 |
+
{
|
| 968 |
+
"epoch": 2.4038461538461537,
|
| 969 |
+
"grad_norm": 5.500666618347168,
|
| 970 |
+
"learning_rate": 5.9615384615384615e-06,
|
| 971 |
+
"loss": 0.1485,
|
| 972 |
+
"step": 3375
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 2.421652421652422,
|
| 976 |
+
"grad_norm": 6.229862689971924,
|
| 977 |
+
"learning_rate": 5.783475783475783e-06,
|
| 978 |
+
"loss": 0.1217,
|
| 979 |
+
"step": 3400
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 2.4394586894586894,
|
| 983 |
+
"grad_norm": 6.825469017028809,
|
| 984 |
+
"learning_rate": 5.605413105413106e-06,
|
| 985 |
+
"loss": 0.1237,
|
| 986 |
+
"step": 3425
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 2.4572649572649574,
|
| 990 |
+
"grad_norm": 4.714917182922363,
|
| 991 |
+
"learning_rate": 5.4273504273504275e-06,
|
| 992 |
+
"loss": 0.1445,
|
| 993 |
+
"step": 3450
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 2.475071225071225,
|
| 997 |
+
"grad_norm": 7.204947471618652,
|
| 998 |
+
"learning_rate": 5.24928774928775e-06,
|
| 999 |
+
"loss": 0.1343,
|
| 1000 |
+
"step": 3475
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"epoch": 2.492877492877493,
|
| 1004 |
+
"grad_norm": 6.491130828857422,
|
| 1005 |
+
"learning_rate": 5.071225071225071e-06,
|
| 1006 |
+
"loss": 0.1321,
|
| 1007 |
+
"step": 3500
|
| 1008 |
+
},
|
| 1009 |
+
{
|
| 1010 |
+
"epoch": 2.5106837606837606,
|
| 1011 |
+
"grad_norm": 8.82093620300293,
|
| 1012 |
+
"learning_rate": 4.8931623931623934e-06,
|
| 1013 |
+
"loss": 0.1328,
|
| 1014 |
+
"step": 3525
|
| 1015 |
+
},
|
| 1016 |
+
{
|
| 1017 |
+
"epoch": 2.5284900284900287,
|
| 1018 |
+
"grad_norm": 6.465578079223633,
|
| 1019 |
+
"learning_rate": 4.715099715099715e-06,
|
| 1020 |
+
"loss": 0.1433,
|
| 1021 |
+
"step": 3550
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"epoch": 2.5462962962962963,
|
| 1025 |
+
"grad_norm": 5.4300127029418945,
|
| 1026 |
+
"learning_rate": 4.537037037037038e-06,
|
| 1027 |
+
"loss": 0.1379,
|
| 1028 |
+
"step": 3575
|
| 1029 |
+
},
|
| 1030 |
+
{
|
| 1031 |
+
"epoch": 2.564102564102564,
|
| 1032 |
+
"grad_norm": 6.0194091796875,
|
| 1033 |
+
"learning_rate": 4.358974358974359e-06,
|
| 1034 |
+
"loss": 0.1165,
|
| 1035 |
+
"step": 3600
|
| 1036 |
+
},
|
| 1037 |
+
{
|
| 1038 |
+
"epoch": 2.581908831908832,
|
| 1039 |
+
"grad_norm": 4.892417907714844,
|
| 1040 |
+
"learning_rate": 4.18091168091168e-06,
|
| 1041 |
+
"loss": 0.1365,
|
| 1042 |
+
"step": 3625
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"epoch": 2.5997150997151,
|
| 1046 |
+
"grad_norm": 8.64039134979248,
|
| 1047 |
+
"learning_rate": 4.002849002849003e-06,
|
| 1048 |
+
"loss": 0.1512,
|
| 1049 |
+
"step": 3650
|
| 1050 |
+
},
|
| 1051 |
+
{
|
| 1052 |
+
"epoch": 2.6175213675213675,
|
| 1053 |
+
"grad_norm": 6.5000691413879395,
|
| 1054 |
+
"learning_rate": 3.8247863247863246e-06,
|
| 1055 |
+
"loss": 0.1316,
|
| 1056 |
+
"step": 3675
|
| 1057 |
+
},
|
| 1058 |
+
{
|
| 1059 |
+
"epoch": 2.635327635327635,
|
| 1060 |
+
"grad_norm": 7.008079528808594,
|
| 1061 |
+
"learning_rate": 3.646723646723647e-06,
|
| 1062 |
+
"loss": 0.1022,
|
| 1063 |
+
"step": 3700
|
| 1064 |
+
},
|
| 1065 |
+
{
|
| 1066 |
+
"epoch": 2.653133903133903,
|
| 1067 |
+
"grad_norm": 7.277550220489502,
|
| 1068 |
+
"learning_rate": 3.468660968660969e-06,
|
| 1069 |
+
"loss": 0.1092,
|
| 1070 |
+
"step": 3725
|
| 1071 |
+
},
|
| 1072 |
+
{
|
| 1073 |
+
"epoch": 2.6709401709401708,
|
| 1074 |
+
"grad_norm": 5.154464244842529,
|
| 1075 |
+
"learning_rate": 3.290598290598291e-06,
|
| 1076 |
+
"loss": 0.1212,
|
| 1077 |
+
"step": 3750
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 2.688746438746439,
|
| 1081 |
+
"grad_norm": 11.289053916931152,
|
| 1082 |
+
"learning_rate": 3.1125356125356127e-06,
|
| 1083 |
+
"loss": 0.1534,
|
| 1084 |
+
"step": 3775
|
| 1085 |
+
},
|
| 1086 |
+
{
|
| 1087 |
+
"epoch": 2.7065527065527064,
|
| 1088 |
+
"grad_norm": 4.6355791091918945,
|
| 1089 |
+
"learning_rate": 2.9344729344729344e-06,
|
| 1090 |
+
"loss": 0.1432,
|
| 1091 |
+
"step": 3800
|
| 1092 |
+
},
|
| 1093 |
+
{
|
| 1094 |
+
"epoch": 2.7243589743589745,
|
| 1095 |
+
"grad_norm": 2.6881229877471924,
|
| 1096 |
+
"learning_rate": 2.7564102564102565e-06,
|
| 1097 |
+
"loss": 0.1405,
|
| 1098 |
+
"step": 3825
|
| 1099 |
+
},
|
| 1100 |
+
{
|
| 1101 |
+
"epoch": 2.742165242165242,
|
| 1102 |
+
"grad_norm": 4.619137763977051,
|
| 1103 |
+
"learning_rate": 2.5783475783475782e-06,
|
| 1104 |
+
"loss": 0.1205,
|
| 1105 |
+
"step": 3850
|
| 1106 |
+
},
|
| 1107 |
+
{
|
| 1108 |
+
"epoch": 2.75997150997151,
|
| 1109 |
+
"grad_norm": 8.594213485717773,
|
| 1110 |
+
"learning_rate": 2.4002849002849004e-06,
|
| 1111 |
+
"loss": 0.1282,
|
| 1112 |
+
"step": 3875
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"epoch": 2.7777777777777777,
|
| 1116 |
+
"grad_norm": 9.55528736114502,
|
| 1117 |
+
"learning_rate": 2.222222222222222e-06,
|
| 1118 |
+
"loss": 0.1353,
|
| 1119 |
+
"step": 3900
|
| 1120 |
+
},
|
| 1121 |
+
{
|
| 1122 |
+
"epoch": 2.7955840455840457,
|
| 1123 |
+
"grad_norm": 12.107255935668945,
|
| 1124 |
+
"learning_rate": 2.044159544159544e-06,
|
| 1125 |
+
"loss": 0.1343,
|
| 1126 |
+
"step": 3925
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 2.8133903133903133,
|
| 1130 |
+
"grad_norm": 5.327492713928223,
|
| 1131 |
+
"learning_rate": 1.8660968660968661e-06,
|
| 1132 |
+
"loss": 0.1304,
|
| 1133 |
+
"step": 3950
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 2.8311965811965814,
|
| 1137 |
+
"grad_norm": 4.199816703796387,
|
| 1138 |
+
"learning_rate": 1.688034188034188e-06,
|
| 1139 |
+
"loss": 0.1526,
|
| 1140 |
+
"step": 3975
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 2.849002849002849,
|
| 1144 |
+
"grad_norm": 11.116687774658203,
|
| 1145 |
+
"learning_rate": 1.50997150997151e-06,
|
| 1146 |
+
"loss": 0.1175,
|
| 1147 |
+
"step": 4000
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 2.866809116809117,
|
| 1151 |
+
"grad_norm": 9.607903480529785,
|
| 1152 |
+
"learning_rate": 1.3319088319088321e-06,
|
| 1153 |
+
"loss": 0.138,
|
| 1154 |
+
"step": 4025
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 2.8846153846153846,
|
| 1158 |
+
"grad_norm": 9.188939094543457,
|
| 1159 |
+
"learning_rate": 1.153846153846154e-06,
|
| 1160 |
+
"loss": 0.1327,
|
| 1161 |
+
"step": 4050
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 2.902421652421652,
|
| 1165 |
+
"grad_norm": 4.503283500671387,
|
| 1166 |
+
"learning_rate": 9.757834757834757e-07,
|
| 1167 |
+
"loss": 0.1409,
|
| 1168 |
+
"step": 4075
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 2.92022792022792,
|
| 1172 |
+
"grad_norm": 9.340811729431152,
|
| 1173 |
+
"learning_rate": 7.977207977207978e-07,
|
| 1174 |
+
"loss": 0.1569,
|
| 1175 |
+
"step": 4100
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 2.9380341880341883,
|
| 1179 |
+
"grad_norm": 6.104117393493652,
|
| 1180 |
+
"learning_rate": 6.196581196581196e-07,
|
| 1181 |
+
"loss": 0.1222,
|
| 1182 |
+
"step": 4125
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 2.955840455840456,
|
| 1186 |
+
"grad_norm": 3.536284923553467,
|
| 1187 |
+
"learning_rate": 4.415954415954416e-07,
|
| 1188 |
+
"loss": 0.1432,
|
| 1189 |
+
"step": 4150
|
| 1190 |
+
},
|
| 1191 |
+
{
|
| 1192 |
+
"epoch": 2.9736467236467234,
|
| 1193 |
+
"grad_norm": 3.488159418106079,
|
| 1194 |
+
"learning_rate": 2.635327635327635e-07,
|
| 1195 |
+
"loss": 0.123,
|
| 1196 |
+
"step": 4175
|
| 1197 |
+
},
|
| 1198 |
+
{
|
| 1199 |
+
"epoch": 2.9914529914529915,
|
| 1200 |
+
"grad_norm": 7.122065544128418,
|
| 1201 |
+
"learning_rate": 8.547008547008547e-08,
|
| 1202 |
+
"loss": 0.1335,
|
| 1203 |
+
"step": 4200
|
| 1204 |
+
},
|
| 1205 |
+
{
|
| 1206 |
+
"epoch": 3.0,
|
| 1207 |
+
"eval_accuracy": 0.8784403669724771,
|
| 1208 |
+
"eval_loss": 0.41880160570144653,
|
| 1209 |
+
"eval_runtime": 3.0955,
|
| 1210 |
+
"eval_samples_per_second": 281.702,
|
| 1211 |
+
"eval_steps_per_second": 3.231,
|
| 1212 |
+
"step": 4212
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 3.0,
|
| 1216 |
+
"step": 4212,
|
| 1217 |
+
"total_flos": 6393601201921362.0,
|
| 1218 |
+
"train_loss": 0.23121720937355744,
|
| 1219 |
+
"train_runtime": 648.2587,
|
| 1220 |
+
"train_samples_per_second": 311.676,
|
| 1221 |
+
"train_steps_per_second": 6.497
|
| 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": 6393601201921362.0,
|
| 1242 |
+
"train_batch_size": 48,
|
| 1243 |
+
"trial_name": null,
|
| 1244 |
+
"trial_params": null
|
| 1245 |
+
}
|
mbert_rtx6000_metrics/result_ablation_mbert/sst2/mBERT/multi_branch_average/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/multi_branch_average/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy multi_branch_average --n_heads 4 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05
|