res

ray_results/test_experiment/train_model_481f1fff_10_batch_size=1024,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=1024,decoder_rt_num_layer=8,drop__2024-06-25_17-33-30/progress.csv

+loss,timestamp,checkpoint_dir_name,should_checkpoint,done,training_iteration,trial_id,date,time_this_iter_s,time_total_s,pid,hostname,node_ip,time_since_restore,iterations_since_restore
+1961.035730076587,1719329807,checkpoint_000000,True,False,1,481f1fff,2024-06-25_17-36-47,113.85971927642822,113.85971927642822,2195253,r8i6n2,10.159.28.60,113.85971927642822,1
+1959.9800554410679,1719330869,checkpoint_000001,True,False,2,481f1fff,2024-06-25_17-54-29,114.88935971260071,228.74907898902893,2195253,r8i6n2,10.159.28.60,114.88935971260071,1

ray_results/test_experiment/train_model_481f1fff_10_batch_size=1024,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=1024,decoder_rt_num_layer=8,drop__2024-06-25_17-33-30/result.json

+{"loss": 1961.035730076587, "timestamp": 1719329807, "checkpoint_dir_name": "checkpoint_000000", "should_checkpoint": true, "done": false, "training_iteration": 1, "trial_id": "481f1fff", "date": "2024-06-25_17-36-47", "time_this_iter_s": 113.85971927642822, "time_total_s": 113.85971927642822, "pid": 2195253, "hostname": "r8i6n2", "node_ip": "10.159.28.60", "config": {"encoder_num_layer": 2, "decoder_rt_num_layer": 8, "decoder_int_num_layer": 1, "embedding_dim": 64, "encoder_ff": 1024, "decoder_rt_ff": 1024, "decoder_int_ff": 512, "n_head": 16, "drop_rate": 0.15995138350968297, "lr": 0.004656630423530019, "batch_size": 1024}, "time_since_restore": 113.85971927642822, "iterations_since_restore": 1}
+{"loss": 1959.9800554410679, "timestamp": 1719330869, "checkpoint_dir_name": "checkpoint_000001", "should_checkpoint": true, "done": false, "training_iteration": 2, "trial_id": "481f1fff", "date": "2024-06-25_17-54-29", "time_this_iter_s": 114.88935971260071, "time_total_s": 228.74907898902893, "pid": 2195253, "hostname": "r8i6n2", "node_ip": "10.159.28.60", "config": {"encoder_num_layer": 2, "decoder_rt_num_layer": 8, "decoder_int_num_layer": 1, "embedding_dim": 64, "encoder_ff": 1024, "decoder_rt_ff": 1024, "decoder_int_ff": 512, "n_head": 16, "drop_rate": 0.15995138350968297, "lr": 0.004656630423530019, "batch_size": 1024}, "time_since_restore": 114.88935971260071, "iterations_since_restore": 1}

ray_results/test_experiment/train_model_52d8dfa1_2_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=1024,decoder_rt_num_layer=4,drop_r_2024-06-25_16-08-04/params.json

+{
+  "batch_size": 2048,
+  "decoder_int_ff": 512,
+  "decoder_int_num_layer": 1,
+  "decoder_rt_ff": 1024,
+  "decoder_rt_num_layer": 4,
+  "drop_rate": 0.982125481793317,
+  "embedding_dim": 16,
+  "encoder_ff": 1024,
+  "encoder_num_layer": 8,
+  "lr": 0.002789281371289888,
+  "n_head": 8
+}
\ No newline at end of file

ray_results/test_experiment/train_model_52d8dfa1_2_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=1024,decoder_rt_num_layer=4,drop_r_2024-06-25_16-08-04/progress.csv

+loss,timestamp,checkpoint_dir_name,should_checkpoint,done,training_iteration,trial_id,date,time_this_iter_s,time_total_s,pid,hostname,node_ip,time_since_restore,iterations_since_restore
+nan,1719324663,checkpoint_000000,True,False,1,52d8dfa1,2024-06-25_16-11-03,89.4755928516388,89.4755928516388,69318,r3i5n6,10.159.8.159,89.4755928516388,1

ray_results/test_experiment/train_model_52d8dfa1_2_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=1024,decoder_rt_num_layer=4,drop_r_2024-06-25_16-08-04/result.json

+{"loss": NaN, "timestamp": 1719324663, "checkpoint_dir_name": "checkpoint_000000", "should_checkpoint": true, "done": false, "training_iteration": 1, "trial_id": "52d8dfa1", "date": "2024-06-25_16-11-03", "time_this_iter_s": 89.4755928516388, "time_total_s": 89.4755928516388, "pid": 69318, "hostname": "r3i5n6", "node_ip": "10.159.8.159", "config": {"encoder_num_layer": 8, "decoder_rt_num_layer": 4, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 1024, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.982125481793317, "lr": 0.002789281371289888, "batch_size": 2048}, "time_since_restore": 89.4755928516388, "iterations_since_restore": 1}

ray_results/test_experiment/train_model_5763f130_1_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=8,drop_r_2024-06-25_17-10-00/params.json

+{
+  "batch_size": 2048,
+  "decoder_int_ff": 512,
+  "decoder_int_num_layer": 1,
+  "decoder_rt_ff": 2048,
+  "decoder_rt_num_layer": 8,
+  "drop_rate": 0.7224904221599877,
+  "embedding_dim": 256,
+  "encoder_ff": 512,
+  "encoder_num_layer": 4,
+  "lr": 0.04561252173503944,
+  "n_head": 2
+}
\ No newline at end of file

ray_results/test_experiment/train_model_5763f130_1_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=8,drop_r_2024-06-25_17-10-00/progress.csv

+loss,timestamp,checkpoint_dir_name,should_checkpoint,done,training_iteration,trial_id,date,time_this_iter_s,time_total_s,pid,hostname,node_ip,time_since_restore,iterations_since_restore
+4538.754975086122,1719328455,checkpoint_000000,True,False,1,5763f130,2024-06-25_17-14-16,242.80822491645813,242.80822491645813,2195253,r8i6n2,10.159.28.60,242.80822491645813,1

ray_results/test_experiment/train_model_5763f130_1_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=8,drop_r_2024-06-25_17-10-00/result.json

+{"loss": 4538.754975086122, "timestamp": 1719328455, "checkpoint_dir_name": "checkpoint_000000", "should_checkpoint": true, "done": false, "training_iteration": 1, "trial_id": "5763f130", "date": "2024-06-25_17-14-16", "time_this_iter_s": 242.80822491645813, "time_total_s": 242.80822491645813, "pid": 2195253, "hostname": "r8i6n2", "node_ip": "10.159.28.60", "config": {"encoder_num_layer": 4, "decoder_rt_num_layer": 8, "decoder_int_num_layer": 1, "embedding_dim": 256, "encoder_ff": 512, "decoder_rt_ff": 2048, "decoder_int_ff": 512, "n_head": 2, "drop_rate": 0.7224904221599877, "lr": 0.04561252173503944, "batch_size": 2048}, "time_since_restore": 242.80822491645813, "iterations_since_restore": 1}

ray_results/test_experiment/train_model_5b0a0ae2_6_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=1,drop_r_2024-06-25_17-18-52/params.json

+{
+  "batch_size": 2048,
+  "decoder_int_ff": 512,
+  "decoder_int_num_layer": 1,
+  "decoder_rt_ff": 2048,
+  "decoder_rt_num_layer": 1,
+  "drop_rate": 0.0063231604639778505,
+  "embedding_dim": 16,
+  "encoder_ff": 512,
+  "encoder_num_layer": 8,
+  "lr": 0.0021079084039053323,
+  "n_head": 16
+}
\ No newline at end of file

ray_results/test_experiment/train_model_5b0a0ae2_6_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=1,drop_r_2024-06-25_17-18-52/progress.csv

+loss,timestamp,checkpoint_dir_name,should_checkpoint,done,training_iteration,trial_id,date,time_this_iter_s,time_total_s,pid,hostname,node_ip,time_since_restore,iterations_since_restore
+1952.9751620555487,1719329461,checkpoint_000000,True,False,1,5b0a0ae2,2024-06-25_17-31-01,84.39049577713013,84.39049577713013,2195253,r8i6n2,10.159.28.60,84.39049577713013,1
+1963.9745468980684,1719330671,checkpoint_000001,True,False,2,5b0a0ae2,2024-06-25_17-51-11,84.48489356040955,168.87538933753967,2195253,r8i6n2,10.159.28.60,84.48489356040955,1

ray_results/test_experiment/train_model_5b0a0ae2_6_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=2048,decoder_rt_num_layer=1,drop_r_2024-06-25_17-18-52/result.json

+{"loss": 1952.9751620555487, "timestamp": 1719329461, "checkpoint_dir_name": "checkpoint_000000", "should_checkpoint": true, "done": false, "training_iteration": 1, "trial_id": "5b0a0ae2", "date": "2024-06-25_17-31-01", "time_this_iter_s": 84.39049577713013, "time_total_s": 84.39049577713013, "pid": 2195253, "hostname": "r8i6n2", "node_ip": "10.159.28.60", "config": {"encoder_num_layer": 8, "decoder_rt_num_layer": 1, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 512, "decoder_rt_ff": 2048, "decoder_int_ff": 512, "n_head": 16, "drop_rate": 0.0063231604639778505, "lr": 0.0021079084039053323, "batch_size": 2048}, "time_since_restore": 84.39049577713013, "iterations_since_restore": 1}
+{"loss": 1963.9745468980684, "timestamp": 1719330671, "checkpoint_dir_name": "checkpoint_000001", "should_checkpoint": true, "done": false, "training_iteration": 2, "trial_id": "5b0a0ae2", "date": "2024-06-25_17-51-11", "time_this_iter_s": 84.48489356040955, "time_total_s": 168.87538933753967, "pid": 2195253, "hostname": "r8i6n2", "node_ip": "10.159.28.60", "config": {"encoder_num_layer": 8, "decoder_rt_num_layer": 1, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 512, "decoder_rt_ff": 2048, "decoder_int_ff": 512, "n_head": 16, "drop_rate": 0.0063231604639778505, "lr": 0.0021079084039053323, "batch_size": 2048}, "time_since_restore": 84.48489356040955, "iterations_since_restore": 1}