res

ray_results/test_experiment/train_model_fbff5866_7_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=512,decoder_rt_num_layer=2,drop_ra_2024-06-27_14-30-30/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
+393.9815991018701,1719491580,checkpoint_000000,True,False,1,fbff5866,2024-06-27_14-33-00,64.76905679702759,64.76905679702759,35196,r8i6n8,10.159.28.66,64.76905679702759,1
+259.43172593004124,1719494192,checkpoint_000001,True,False,2,fbff5866,2024-06-27_15-16-32,65.51874613761902,130.2878029346466,69992,r8i6n8,10.159.28.66,65.51874613761902,1
+213.22481638990988,1719495267,checkpoint_000002,True,False,3,fbff5866,2024-06-27_15-34-27,64.90303182601929,195.1908347606659,69992,r8i6n8,10.159.28.66,64.90303182601929,1
+206.50029820720042,1719495319,checkpoint_000003,True,False,4,fbff5866,2024-06-27_15-35-19,51.688610315322876,246.87944507598877,69992,r8i6n8,10.159.28.66,116.59164214134216,2
+184.2231689212829,1719495987,checkpoint_000004,True,False,5,fbff5866,2024-06-27_15-46-27,65.34643507003784,312.2258801460266,69992,r8i6n8,10.159.28.66,65.34643507003784,1
+178.49155504121555,1719496039,checkpoint_000005,True,False,6,fbff5866,2024-06-27_15-47-19,51.88559675216675,364.11147689819336,69992,r8i6n8,10.159.28.66,117.23203182220459,2
+168.78981534702572,1719496091,checkpoint_000006,True,False,7,fbff5866,2024-06-27_15-48-11,51.525290727615356,415.6367676258087,69992,r8i6n8,10.159.28.66,168.75732254981995,3
+162.7404464361236,1719496142,checkpoint_000007,True,False,8,fbff5866,2024-06-27_15-49-02,51.523138761520386,467.1599063873291,69992,r8i6n8,10.159.28.66,220.28046131134033,4

ray_results/test_experiment/train_model_fbff5866_7_batch_size=2048,decoder_int_ff=512,decoder_int_num_layer=1,decoder_rt_ff=512,decoder_rt_num_layer=2,drop_ra_2024-06-27_14-30-30/result.json

+{"loss": 393.9815991018701, "timestamp": 1719491580, "checkpoint_dir_name": "checkpoint_000000", "should_checkpoint": true, "done": false, "training_iteration": 1, "trial_id": "fbff5866", "date": "2024-06-27_14-33-00", "time_this_iter_s": 64.76905679702759, "time_total_s": 64.76905679702759, "pid": 35196, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 64.76905679702759, "iterations_since_restore": 1}
+{"loss": 259.43172593004124, "timestamp": 1719494192, "checkpoint_dir_name": "checkpoint_000001", "should_checkpoint": true, "done": false, "training_iteration": 2, "trial_id": "fbff5866", "date": "2024-06-27_15-16-32", "time_this_iter_s": 65.51874613761902, "time_total_s": 130.2878029346466, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 65.51874613761902, "iterations_since_restore": 1}
+{"loss": 213.22481638990988, "timestamp": 1719495267, "checkpoint_dir_name": "checkpoint_000002", "should_checkpoint": true, "done": false, "training_iteration": 3, "trial_id": "fbff5866", "date": "2024-06-27_15-34-27", "time_this_iter_s": 64.90303182601929, "time_total_s": 195.1908347606659, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 64.90303182601929, "iterations_since_restore": 1}
+{"loss": 206.50029820720042, "timestamp": 1719495319, "checkpoint_dir_name": "checkpoint_000003", "should_checkpoint": true, "done": false, "training_iteration": 4, "trial_id": "fbff5866", "date": "2024-06-27_15-35-19", "time_this_iter_s": 51.688610315322876, "time_total_s": 246.87944507598877, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 116.59164214134216, "iterations_since_restore": 2}
+{"loss": 184.2231689212829, "timestamp": 1719495987, "checkpoint_dir_name": "checkpoint_000004", "should_checkpoint": true, "done": false, "training_iteration": 5, "trial_id": "fbff5866", "date": "2024-06-27_15-46-27", "time_this_iter_s": 65.34643507003784, "time_total_s": 312.2258801460266, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 65.34643507003784, "iterations_since_restore": 1}
+{"loss": 178.49155504121555, "timestamp": 1719496039, "checkpoint_dir_name": "checkpoint_000005", "should_checkpoint": true, "done": false, "training_iteration": 6, "trial_id": "fbff5866", "date": "2024-06-27_15-47-19", "time_this_iter_s": 51.88559675216675, "time_total_s": 364.11147689819336, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 117.23203182220459, "iterations_since_restore": 2}
+{"loss": 168.78981534702572, "timestamp": 1719496091, "checkpoint_dir_name": "checkpoint_000006", "should_checkpoint": true, "done": false, "training_iteration": 7, "trial_id": "fbff5866", "date": "2024-06-27_15-48-11", "time_this_iter_s": 51.525290727615356, "time_total_s": 415.6367676258087, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 168.75732254981995, "iterations_since_restore": 3}
+{"loss": 162.7404464361236, "timestamp": 1719496142, "checkpoint_dir_name": "checkpoint_000007", "should_checkpoint": true, "done": false, "training_iteration": 8, "trial_id": "fbff5866", "date": "2024-06-27_15-49-02", "time_this_iter_s": 51.523138761520386, "time_total_s": 467.1599063873291, "pid": 69992, "hostname": "r8i6n8", "node_ip": "10.159.28.66", "config": {"encoder_num_layer": 1, "decoder_rt_num_layer": 2, "decoder_int_num_layer": 1, "embedding_dim": 16, "encoder_ff": 1024, "decoder_rt_ff": 512, "decoder_int_ff": 512, "n_head": 8, "drop_rate": 0.4230793088582453, "lr": 0.00288155269334617, "batch_size": 2048}, "time_since_restore": 220.28046131134033, "iterations_since_restore": 4}