Upload batch 12/14: vsfc/mBERT/cls/seed_44 ... vsfc/mBERT/hf_sequence_classifier/seed_44

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44/all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.934301958307012,
+    "eval_loss": 0.21943630278110504,
+    "eval_runtime": 5.6109,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 282.132,
+    "eval_steps_per_second": 1.782,
+    "total_flos": 1151671573807104.0,
+    "train_loss": 0.232700464608786,
+    "train_runtime": 53.0425,
+    "train_samples": 11426,
+    "train_samples_per_second": 646.237,
+    "train_steps_per_second": 8.088
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44/eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.934301958307012,
+    "eval_loss": 0.21943630278110504,
+    "eval_runtime": 5.6109,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 282.132,
+    "eval_steps_per_second": 1.782
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 44 --output_dir result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy cls --n_heads 8 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44/train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "total_flos": 1151671573807104.0,
+    "train_loss": 0.232700464608786,
+    "train_runtime": 53.0425,
+    "train_samples": 11426,
+    "train_samples_per_second": 646.237,
+    "train_steps_per_second": 8.088
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/cls/seed_44/trainer_state.json

@@ -0,0 +1,125 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 429,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.34965034965034963,
+      "grad_norm": 5.308379650115967,
+      "learning_rate": 1.766899766899767e-05,
+      "loss": 0.4749,
+      "step": 50
+    },
+    {
+      "epoch": 0.6993006993006993,
+      "grad_norm": 8.536895751953125,
+      "learning_rate": 1.533799533799534e-05,
+      "loss": 0.2955,
+      "step": 100
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9260897030953885,
+      "eval_loss": 0.23385658860206604,
+      "eval_runtime": 5.8793,
+      "eval_samples_per_second": 269.25,
+      "eval_steps_per_second": 1.701,
+      "step": 143
+    },
+    {
+      "epoch": 1.048951048951049,
+      "grad_norm": 2.7078521251678467,
+      "learning_rate": 1.3006993006993008e-05,
+      "loss": 0.237,
+      "step": 150
+    },
+    {
+      "epoch": 1.3986013986013985,
+      "grad_norm": 2.5117030143737793,
+      "learning_rate": 1.0675990675990678e-05,
+      "loss": 0.2079,
+      "step": 200
+    },
+    {
+      "epoch": 1.7482517482517483,
+      "grad_norm": 3.446140766143799,
+      "learning_rate": 8.344988344988347e-06,
+      "loss": 0.1991,
+      "step": 250
+    },
+    {
+      "epoch": 2.0,
+      "eval_accuracy": 0.934301958307012,
+      "eval_loss": 0.20997200906276703,
+      "eval_runtime": 6.2227,
+      "eval_samples_per_second": 254.39,
+      "eval_steps_per_second": 1.607,
+      "step": 286
+    },
+    {
+      "epoch": 2.097902097902098,
+      "grad_norm": 4.669304847717285,
+      "learning_rate": 6.013986013986014e-06,
+      "loss": 0.1858,
+      "step": 300
+    },
+    {
+      "epoch": 2.4475524475524475,
+      "grad_norm": 3.715109348297119,
+      "learning_rate": 3.6829836829836835e-06,
+      "loss": 0.1516,
+      "step": 350
+    },
+    {
+      "epoch": 2.797202797202797,
+      "grad_norm": 3.4745700359344482,
+      "learning_rate": 1.3519813519813522e-06,
+      "loss": 0.1539,
+      "step": 400
+    },
+    {
+      "epoch": 3.0,
+      "eval_accuracy": 0.934301958307012,
+      "eval_loss": 0.21943630278110504,
+      "eval_runtime": 6.0122,
+      "eval_samples_per_second": 263.3,
+      "eval_steps_per_second": 1.663,
+      "step": 429
+    },
+    {
+      "epoch": 3.0,
+      "step": 429,
+      "total_flos": 1151671573807104.0,
+      "train_loss": 0.232700464608786,
+      "train_runtime": 53.0425,
+      "train_samples_per_second": 646.237,
+      "train_steps_per_second": 8.088
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 429,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1151671573807104.0,
+  "train_batch_size": 80,
+  "trial_name": null,
+  "trial_params": null
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_42/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 42 --output_dir result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy gated_multi_branch --n_heads 8 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_43/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 43 --output_dir result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy gated_multi_branch --n_heads 8 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44/all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.9279848389134555,
+    "eval_loss": 0.21946457028388977,
+    "eval_runtime": 5.2714,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 300.301,
+    "eval_steps_per_second": 1.897,
+    "total_flos": 1204364884372608.0,
+    "train_loss": 0.2245002004094335,
+    "train_runtime": 39.7921,
+    "train_samples": 11426,
+    "train_samples_per_second": 861.428,
+    "train_steps_per_second": 10.781
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44/eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.9279848389134555,
+    "eval_loss": 0.21946457028388977,
+    "eval_runtime": 5.2714,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 300.301,
+    "eval_steps_per_second": 1.897
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 44 --output_dir result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save --use_mha true --pooling_strategy gated_multi_branch --n_heads 8 --head_dim 32 --classifier_hidden_dim 512 --head_dropout 0.05

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44/train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "total_flos": 1204364884372608.0,
+    "train_loss": 0.2245002004094335,
+    "train_runtime": 39.7921,
+    "train_samples": 11426,
+    "train_samples_per_second": 861.428,
+    "train_steps_per_second": 10.781
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/gated_multi_branch/seed_44/trainer_state.json

@@ -0,0 +1,125 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 429,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.34965034965034963,
+      "grad_norm": 7.766966342926025,
+      "learning_rate": 1.766899766899767e-05,
+      "loss": 0.4575,
+      "step": 50
+    },
+    {
+      "epoch": 0.6993006993006993,
+      "grad_norm": 6.951184272766113,
+      "learning_rate": 1.533799533799534e-05,
+      "loss": 0.2786,
+      "step": 100
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9267214150347441,
+      "eval_loss": 0.22632087767124176,
+      "eval_runtime": 5.6378,
+      "eval_samples_per_second": 280.785,
+      "eval_steps_per_second": 1.774,
+      "step": 143
+    },
+    {
+      "epoch": 1.048951048951049,
+      "grad_norm": 3.147196054458618,
+      "learning_rate": 1.3006993006993008e-05,
+      "loss": 0.229,
+      "step": 150
+    },
+    {
+      "epoch": 1.3986013986013985,
+      "grad_norm": 2.899240493774414,
+      "learning_rate": 1.0675990675990678e-05,
+      "loss": 0.1974,
+      "step": 200
+    },
+    {
+      "epoch": 1.7482517482517483,
+      "grad_norm": 3.776470422744751,
+      "learning_rate": 8.344988344988347e-06,
+      "loss": 0.1992,
+      "step": 250
+    },
+    {
+      "epoch": 2.0,
+      "eval_accuracy": 0.9298799747315224,
+      "eval_loss": 0.2132049798965454,
+      "eval_runtime": 5.4051,
+      "eval_samples_per_second": 292.871,
+      "eval_steps_per_second": 1.85,
+      "step": 286
+    },
+    {
+      "epoch": 2.097902097902098,
+      "grad_norm": 3.9233686923980713,
+      "learning_rate": 6.013986013986014e-06,
+      "loss": 0.1868,
+      "step": 300
+    },
+    {
+      "epoch": 2.4475524475524475,
+      "grad_norm": 4.221687316894531,
+      "learning_rate": 3.6829836829836835e-06,
+      "loss": 0.1458,
+      "step": 350
+    },
+    {
+      "epoch": 2.797202797202797,
+      "grad_norm": 3.8544631004333496,
+      "learning_rate": 1.3519813519813522e-06,
+      "loss": 0.1468,
+      "step": 400
+    },
+    {
+      "epoch": 3.0,
+      "eval_accuracy": 0.9279848389134555,
+      "eval_loss": 0.21946457028388977,
+      "eval_runtime": 5.2809,
+      "eval_samples_per_second": 299.757,
+      "eval_steps_per_second": 1.894,
+      "step": 429
+    },
+    {
+      "epoch": 3.0,
+      "step": 429,
+      "total_flos": 1204364884372608.0,
+      "train_loss": 0.2245002004094335,
+      "train_runtime": 39.7921,
+      "train_samples_per_second": 861.428,
+      "train_steps_per_second": 10.781
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 429,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1204364884372608.0,
+  "train_batch_size": 80,
+  "trial_name": null,
+  "trial_params": null
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42/all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.932406822488945,
+    "eval_loss": 0.22124937176704407,
+    "eval_runtime": 6.6035,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 239.722,
+    "eval_steps_per_second": 1.514,
+    "total_flos": 1148121092925828.0,
+    "train_loss": 0.2343874844637784,
+    "train_runtime": 68.7457,
+    "train_samples": 11426,
+    "train_samples_per_second": 498.62,
+    "train_steps_per_second": 6.24
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42/eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.932406822488945,
+    "eval_loss": 0.22124937176704407,
+    "eval_runtime": 6.6035,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 239.722,
+    "eval_steps_per_second": 1.514
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 42 --output_dir result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42/train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "total_flos": 1148121092925828.0,
+    "train_loss": 0.2343874844637784,
+    "train_runtime": 68.7457,
+    "train_samples": 11426,
+    "train_samples_per_second": 498.62,
+    "train_steps_per_second": 6.24
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_42/trainer_state.json

@@ -0,0 +1,125 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 429,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.34965034965034963,
+      "grad_norm": 3.6422343254089355,
+      "learning_rate": 1.766899766899767e-05,
+      "loss": 0.4778,
+      "step": 50
+    },
+    {
+      "epoch": 0.6993006993006993,
+      "grad_norm": 2.4763453006744385,
+      "learning_rate": 1.533799533799534e-05,
+      "loss": 0.2975,
+      "step": 100
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9279848389134555,
+      "eval_loss": 0.22652655839920044,
+      "eval_runtime": 6.581,
+      "eval_samples_per_second": 240.539,
+      "eval_steps_per_second": 1.52,
+      "step": 143
+    },
+    {
+      "epoch": 1.048951048951049,
+      "grad_norm": 4.861372470855713,
+      "learning_rate": 1.3006993006993008e-05,
+      "loss": 0.2567,
+      "step": 150
+    },
+    {
+      "epoch": 1.3986013986013985,
+      "grad_norm": 4.7382988929748535,
+      "learning_rate": 1.0675990675990678e-05,
+      "loss": 0.1964,
+      "step": 200
+    },
+    {
+      "epoch": 1.7482517482517483,
+      "grad_norm": 5.947739124298096,
+      "learning_rate": 8.344988344988347e-06,
+      "loss": 0.2085,
+      "step": 250
+    },
+    {
+      "epoch": 2.0,
+      "eval_accuracy": 0.9248262792166771,
+      "eval_loss": 0.2125282734632492,
+      "eval_runtime": 5.9176,
+      "eval_samples_per_second": 267.506,
+      "eval_steps_per_second": 1.69,
+      "step": 286
+    },
+    {
+      "epoch": 2.097902097902098,
+      "grad_norm": 5.443600177764893,
+      "learning_rate": 6.013986013986014e-06,
+      "loss": 0.1788,
+      "step": 300
+    },
+    {
+      "epoch": 2.4475524475524475,
+      "grad_norm": 3.66668963432312,
+      "learning_rate": 3.6829836829836835e-06,
+      "loss": 0.1588,
+      "step": 350
+    },
+    {
+      "epoch": 2.797202797202797,
+      "grad_norm": 6.581542491912842,
+      "learning_rate": 1.3519813519813522e-06,
+      "loss": 0.1494,
+      "step": 400
+    },
+    {
+      "epoch": 3.0,
+      "eval_accuracy": 0.932406822488945,
+      "eval_loss": 0.22124937176704407,
+      "eval_runtime": 6.7007,
+      "eval_samples_per_second": 236.243,
+      "eval_steps_per_second": 1.492,
+      "step": 429
+    },
+    {
+      "epoch": 3.0,
+      "step": 429,
+      "total_flos": 1148121092925828.0,
+      "train_loss": 0.2343874844637784,
+      "train_runtime": 68.7457,
+      "train_samples_per_second": 498.62,
+      "train_steps_per_second": 6.24
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 429,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1148121092925828.0,
+  "train_batch_size": 80,
+  "trial_name": null,
+  "trial_params": null
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43/all_results.json

@@ -0,0 +1,15 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.9317751105495894,
+    "eval_loss": 0.2234826385974884,
+    "eval_runtime": 6.7212,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 235.522,
+    "eval_steps_per_second": 1.488,
+    "total_flos": 1148390373064284.0,
+    "train_loss": 0.2347770626450474,
+    "train_runtime": 58.1343,
+    "train_samples": 11426,
+    "train_samples_per_second": 589.635,
+    "train_steps_per_second": 7.379
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43/eval_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "eval_accuracy": 0.9317751105495894,
+    "eval_loss": 0.2234826385974884,
+    "eval_runtime": 6.7212,
+    "eval_samples": 1583,
+    "eval_samples_per_second": 235.522,
+    "eval_steps_per_second": 1.488
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 43 --output_dir result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43/train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "total_flos": 1148390373064284.0,
+    "train_loss": 0.2347770626450474,
+    "train_runtime": 58.1343,
+    "train_samples": 11426,
+    "train_samples_per_second": 589.635,
+    "train_steps_per_second": 7.379
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_43/trainer_state.json

@@ -0,0 +1,125 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 429,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.34965034965034963,
+      "grad_norm": 6.4440836906433105,
+      "learning_rate": 1.766899766899767e-05,
+      "loss": 0.4522,
+      "step": 50
+    },
+    {
+      "epoch": 0.6993006993006993,
+      "grad_norm": 2.3879895210266113,
+      "learning_rate": 1.533799533799534e-05,
+      "loss": 0.2934,
+      "step": 100
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9204042956411876,
+      "eval_loss": 0.24103619158267975,
+      "eval_runtime": 6.6958,
+      "eval_samples_per_second": 236.416,
+      "eval_steps_per_second": 1.493,
+      "step": 143
+    },
+    {
+      "epoch": 1.048951048951049,
+      "grad_norm": 2.3838980197906494,
+      "learning_rate": 1.3006993006993008e-05,
+      "loss": 0.251,
+      "step": 150
+    },
+    {
+      "epoch": 1.3986013986013985,
+      "grad_norm": 4.15537166595459,
+      "learning_rate": 1.0675990675990678e-05,
+      "loss": 0.2173,
+      "step": 200
+    },
+    {
+      "epoch": 1.7482517482517483,
+      "grad_norm": 5.5938310623168945,
+      "learning_rate": 8.344988344988347e-06,
+      "loss": 0.2091,
+      "step": 250
+    },
+    {
+      "epoch": 2.0,
+      "eval_accuracy": 0.930511686670878,
+      "eval_loss": 0.20971274375915527,
+      "eval_runtime": 6.3891,
+      "eval_samples_per_second": 247.767,
+      "eval_steps_per_second": 1.565,
+      "step": 286
+    },
+    {
+      "epoch": 2.097902097902098,
+      "grad_norm": 4.270980358123779,
+      "learning_rate": 6.013986013986014e-06,
+      "loss": 0.1832,
+      "step": 300
+    },
+    {
+      "epoch": 2.4475524475524475,
+      "grad_norm": 3.5465869903564453,
+      "learning_rate": 3.6829836829836835e-06,
+      "loss": 0.164,
+      "step": 350
+    },
+    {
+      "epoch": 2.797202797202797,
+      "grad_norm": 4.430685520172119,
+      "learning_rate": 1.3519813519813522e-06,
+      "loss": 0.1564,
+      "step": 400
+    },
+    {
+      "epoch": 3.0,
+      "eval_accuracy": 0.9317751105495894,
+      "eval_loss": 0.2234826385974884,
+      "eval_runtime": 5.8876,
+      "eval_samples_per_second": 268.868,
+      "eval_steps_per_second": 1.698,
+      "step": 429
+    },
+    {
+      "epoch": 3.0,
+      "step": 429,
+      "total_flos": 1148390373064284.0,
+      "train_loss": 0.2347770626450474,
+      "train_runtime": 58.1343,
+      "train_samples_per_second": 589.635,
+      "train_steps_per_second": 7.379
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 429,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": false,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1148390373064284.0,
+  "train_batch_size": 80,
+  "trial_name": null,
+  "trial_params": null
+}

mbert_paper_metrics/result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_44/run_command.txt

@@ -0,0 +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 vsfc --max_seq_length 256 --do_train --do_eval --evaluation_strategy epoch --save_strategy no --logging_strategy steps --logging_steps 50 --report_to none --num_train_epochs 3 --learning_rate 2e-05 --per_device_train_batch_size 80 --per_device_eval_batch_size 168 --gradient_accumulation_steps 1 --dataloader_num_workers 2 --seed 44 --output_dir result_ablation_mbert_paper/vsfc/mBERT/hf_sequence_classifier/seed_44 --overwrite_output_dir --trust_remote_code true --bf16 --tf32 true --skip_model_save