jdorairaj commited on
Commit
4c4267d
·
1 Parent(s): e10ab79

completed runs on la

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. outputs/sst2/args.json +1 -1
  2. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/logfile_la_all.log +30 -0
  3. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  4. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  5. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/gpu_stats_la.json +130 -0
  6. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  7. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  8. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/gpu_stats_la.json +130 -0
  9. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  10. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  11. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/gpu_stats_la.json +130 -0
  12. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  13. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  14. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/gpu_stats_la.json +130 -0
  15. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  16. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  17. outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/gpu_stats_la.json +130 -0
  18. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/logfile_la_all.log +30 -0
  19. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  20. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  21. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/gpu_stats_la.json +130 -0
  22. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  23. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  24. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/gpu_stats_la.json +130 -0
  25. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  26. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  27. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/gpu_stats_la.json +130 -0
  28. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  29. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  30. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/gpu_stats_la.json +130 -0
  31. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  32. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  33. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/gpu_stats_la.json +130 -0
  34. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  35. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  36. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/gpu_stats_la.json +130 -0
  37. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/logfile_la_all.log +30 -0
  38. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  39. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  40. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/gpu_stats_la.json +130 -0
  41. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  42. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  43. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/gpu_stats_la.json +130 -0
  44. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  45. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  46. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/gpu_stats_la.json +130 -0
  47. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
  48. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json +0 -0
  49. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/gpu_stats_la.json +130 -0
  50. outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json +1 -0
outputs/sst2/args.json CHANGED
@@ -4,7 +4,7 @@
4
  "validation_file": null,
5
  "max_length": 300,
6
  "pad_to_max_length": false,
7
- "model_name_or_path": "bert-base-uncased",
8
  "use_slow_tokenizer": false,
9
  "per_device_train_batch_size": 8,
10
  "per_device_eval_batch_size": 8,
 
4
  "validation_file": null,
5
  "max_length": 300,
6
  "pad_to_max_length": false,
7
+ "model_name_or_path": "roberta-base",
8
  "use_slow_tokenizer": false,
9
  "per_device_train_batch_size": 8,
10
  "per_device_eval_batch_size": 8,
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/logfile_la_all.log CHANGED
@@ -7,3 +7,33 @@
7
  06/01/2024 00:21:08 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
8
  06/01/2024 00:21:08 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
9
  06/01/2024 00:21:08 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  06/01/2024 00:21:08 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
8
  06/01/2024 00:21:08 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
9
  06/01/2024 00:21:08 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
10
+ 06/01/2024 00:57:44 - INFO - __main__ - ***** Starting script *****
11
+ 06/01/2024 00:57:47 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
12
+ 06/01/2024 00:57:47 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
13
+ 06/01/2024 00:57:47 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
14
+ 06/01/2024 01:24:39 - INFO - __main__ - ***** Completed Script *****
15
+ 06/01/2024 01:24:42 - INFO - __main__ - ***** Starting script *****
16
+ 06/01/2024 01:24:44 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
17
+ 06/01/2024 01:24:44 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
18
+ 06/01/2024 01:24:44 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
19
+ 06/01/2024 01:52:08 - INFO - __main__ - ***** Completed Script *****
20
+ 06/01/2024 01:52:11 - INFO - __main__ - ***** Starting script *****
21
+ 06/01/2024 01:52:16 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
22
+ 06/01/2024 01:52:16 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
23
+ 06/01/2024 01:52:16 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
24
+ 06/01/2024 02:19:44 - INFO - __main__ - ***** Completed Script *****
25
+ 06/01/2024 02:19:47 - INFO - __main__ - ***** Starting script *****
26
+ 06/01/2024 02:19:53 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
27
+ 06/01/2024 02:19:53 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
28
+ 06/01/2024 02:19:53 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
29
+ 06/01/2024 02:47:48 - INFO - __main__ - ***** Completed Script *****
30
+ 06/01/2024 02:47:51 - INFO - __main__ - ***** Starting script *****
31
+ 06/01/2024 02:47:57 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
32
+ 06/01/2024 02:47:57 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
33
+ 06/01/2024 02:47:57 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
34
+ 06/01/2024 03:15:55 - INFO - __main__ - ***** Completed Script *****
35
+ 06/01/2024 03:15:58 - INFO - __main__ - ***** Starting script *****
36
+ 06/01/2024 03:16:04 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [101, 12636, 19720, 1010, 5457, 2004, 2000, 3251, 2017, 1005, 2310, 2464, 19378, 2030, 4516, 1012, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
37
+ 06/01/2024 03:16:04 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [101, 2035, 1996, 4540, 27373, 1997, 2413, 5988, 2012, 2049, 2190, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
38
+ 06/01/2024 03:16:04 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [101, 11896, 2061, 24670, 2006, 2296, 7511, 2504, 2008, 2009, 6162, 2015, 2070, 2785, 1997, 27243, 9026, 3126, 102], 'token_type_ids': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
39
+ 06/01/2024 03:44:15 - INFO - __main__ - ***** Completed Script *****
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.8876146788990825}
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 802280960,
3
+ "max_memory_allocated": 1411666432,
4
+ "memory_reserved": 2021654528,
5
+ "max_memory_reserved": 2021654528,
6
+ "memory_stats": {
7
+ "active.all.allocated": 20121290,
8
+ "active.all.current": 919,
9
+ "active.all.freed": 20120371,
10
+ "active.all.peak": 1093,
11
+ "active.large_pool.allocated": 4513198,
12
+ "active.large_pool.current": 222,
13
+ "active.large_pool.freed": 4512976,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 15608092,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 15607395,
18
+ "active.small_pool.peak": 857,
19
+ "active_bytes.all.allocated": 16484981124096,
20
+ "active_bytes.all.current": 802280960,
21
+ "active_bytes.all.freed": 16484178843136,
22
+ "active_bytes.all.peak": 1411666432,
23
+ "active_bytes.large_pool.allocated": 10721397640192,
24
+ "active_bytes.large_pool.current": 799014912,
25
+ "active_bytes.large_pool.freed": 10720598625280,
26
+ "active_bytes.large_pool.peak": 1398726656,
27
+ "active_bytes.small_pool.allocated": 5763583483904,
28
+ "active_bytes.small_pool.current": 3266048,
29
+ "active_bytes.small_pool.freed": 5763580217856,
30
+ "active_bytes.small_pool.peak": 119100928,
31
+ "allocated_bytes.all.allocated": 16484981124096,
32
+ "allocated_bytes.all.current": 802280960,
33
+ "allocated_bytes.all.freed": 16484178843136,
34
+ "allocated_bytes.all.peak": 1411666432,
35
+ "allocated_bytes.large_pool.allocated": 10721397640192,
36
+ "allocated_bytes.large_pool.current": 799014912,
37
+ "allocated_bytes.large_pool.freed": 10720598625280,
38
+ "allocated_bytes.large_pool.peak": 1398726656,
39
+ "allocated_bytes.small_pool.allocated": 5763583483904,
40
+ "allocated_bytes.small_pool.current": 3266048,
41
+ "allocated_bytes.small_pool.freed": 5763580217856,
42
+ "allocated_bytes.small_pool.peak": 119100928,
43
+ "allocation.all.allocated": 20121290,
44
+ "allocation.all.current": 919,
45
+ "allocation.all.freed": 20120371,
46
+ "allocation.all.peak": 1093,
47
+ "allocation.large_pool.allocated": 4513198,
48
+ "allocation.large_pool.current": 222,
49
+ "allocation.large_pool.freed": 4512976,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 15608092,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 15607395,
54
+ "allocation.small_pool.peak": 857,
55
+ "inactive_split.all.allocated": 9761883,
56
+ "inactive_split.all.current": 116,
57
+ "inactive_split.all.freed": 9761767,
58
+ "inactive_split.all.peak": 208,
59
+ "inactive_split.large_pool.allocated": 2398367,
60
+ "inactive_split.large_pool.current": 36,
61
+ "inactive_split.large_pool.freed": 2398331,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 7363516,
64
+ "inactive_split.small_pool.current": 80,
65
+ "inactive_split.small_pool.freed": 7363436,
66
+ "inactive_split.small_pool.peak": 135,
67
+ "inactive_split_bytes.all.allocated": 21986708836864,
68
+ "inactive_split_bytes.all.current": 128854528,
69
+ "inactive_split_bytes.all.freed": 21986579982336,
70
+ "inactive_split_bytes.all.peak": 548717056,
71
+ "inactive_split_bytes.large_pool.allocated": 15586332555264,
72
+ "inactive_split_bytes.large_pool.current": 92274688,
73
+ "inactive_split_bytes.large_pool.freed": 15586240280576,
74
+ "inactive_split_bytes.large_pool.peak": 518733824,
75
+ "inactive_split_bytes.small_pool.allocated": 6400376281600,
76
+ "inactive_split_bytes.small_pool.current": 36579840,
77
+ "inactive_split_bytes.small_pool.freed": 6400339701760,
78
+ "inactive_split_bytes.small_pool.peak": 90121728,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 392,
82
+ "num_device_free": 234,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 3,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 15954340028760,
94
+ "requested_bytes.all.current": 798501024,
95
+ "requested_bytes.all.freed": 15953541527736,
96
+ "requested_bytes.all.peak": 1396203552,
97
+ "requested_bytes.large_pool.allocated": 10192185709952,
98
+ "requested_bytes.large_pool.current": 795392000,
99
+ "requested_bytes.large_pool.freed": 10191390317952,
100
+ "requested_bytes.large_pool.peak": 1383348224,
101
+ "requested_bytes.small_pool.allocated": 5762154318808,
102
+ "requested_bytes.small_pool.current": 3109024,
103
+ "requested_bytes.small_pool.freed": 5762151209784,
104
+ "requested_bytes.small_pool.peak": 118941096,
105
+ "reserved_bytes.all.allocated": 3902799872,
106
+ "reserved_bytes.all.current": 2021654528,
107
+ "reserved_bytes.all.freed": 1881145344,
108
+ "reserved_bytes.all.peak": 2021654528,
109
+ "reserved_bytes.large_pool.allocated": 3470786560,
110
+ "reserved_bytes.large_pool.current": 1897922560,
111
+ "reserved_bytes.large_pool.freed": 1572864000,
112
+ "reserved_bytes.large_pool.peak": 1897922560,
113
+ "reserved_bytes.small_pool.allocated": 432013312,
114
+ "reserved_bytes.small_pool.current": 123731968,
115
+ "reserved_bytes.small_pool.freed": 308281344,
116
+ "reserved_bytes.small_pool.peak": 123731968,
117
+ "segment.all.allocated": 392,
118
+ "segment.all.current": 158,
119
+ "segment.all.freed": 234,
120
+ "segment.all.peak": 158,
121
+ "segment.large_pool.allocated": 186,
122
+ "segment.large_pool.current": 99,
123
+ "segment.large_pool.freed": 87,
124
+ "segment.large_pool.peak": 99,
125
+ "segment.small_pool.allocated": 206,
126
+ "segment.small_pool.current": 59,
127
+ "segment.small_pool.freed": 147,
128
+ "segment.small_pool.peak": 59
129
+ }
130
+ }
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.893348623853211}
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 802280960,
3
+ "max_memory_allocated": 1411666432,
4
+ "memory_reserved": 2021654528,
5
+ "max_memory_reserved": 2021654528,
6
+ "memory_stats": {
7
+ "active.all.allocated": 30181934,
8
+ "active.all.current": 919,
9
+ "active.all.freed": 30181015,
10
+ "active.all.peak": 1093,
11
+ "active.large_pool.allocated": 6769796,
12
+ "active.large_pool.current": 222,
13
+ "active.large_pool.freed": 6769574,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 23412138,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 23411441,
18
+ "active.small_pool.peak": 857,
19
+ "active_bytes.all.allocated": 24730647784960,
20
+ "active_bytes.all.current": 802280960,
21
+ "active_bytes.all.freed": 24729845504000,
22
+ "active_bytes.all.peak": 1411666432,
23
+ "active_bytes.large_pool.allocated": 16085272559104,
24
+ "active_bytes.large_pool.current": 799014912,
25
+ "active_bytes.large_pool.freed": 16084473544192,
26
+ "active_bytes.large_pool.peak": 1398726656,
27
+ "active_bytes.small_pool.allocated": 8645375225856,
28
+ "active_bytes.small_pool.current": 3266048,
29
+ "active_bytes.small_pool.freed": 8645371959808,
30
+ "active_bytes.small_pool.peak": 119100928,
31
+ "allocated_bytes.all.allocated": 24730647784960,
32
+ "allocated_bytes.all.current": 802280960,
33
+ "allocated_bytes.all.freed": 24729845504000,
34
+ "allocated_bytes.all.peak": 1411666432,
35
+ "allocated_bytes.large_pool.allocated": 16085272559104,
36
+ "allocated_bytes.large_pool.current": 799014912,
37
+ "allocated_bytes.large_pool.freed": 16084473544192,
38
+ "allocated_bytes.large_pool.peak": 1398726656,
39
+ "allocated_bytes.small_pool.allocated": 8645375225856,
40
+ "allocated_bytes.small_pool.current": 3266048,
41
+ "allocated_bytes.small_pool.freed": 8645371959808,
42
+ "allocated_bytes.small_pool.peak": 119100928,
43
+ "allocation.all.allocated": 30181934,
44
+ "allocation.all.current": 919,
45
+ "allocation.all.freed": 30181015,
46
+ "allocation.all.peak": 1093,
47
+ "allocation.large_pool.allocated": 6769796,
48
+ "allocation.large_pool.current": 222,
49
+ "allocation.large_pool.freed": 6769574,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 23412138,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 23411441,
54
+ "allocation.small_pool.peak": 857,
55
+ "inactive_split.all.allocated": 14643556,
56
+ "inactive_split.all.current": 116,
57
+ "inactive_split.all.freed": 14643440,
58
+ "inactive_split.all.peak": 208,
59
+ "inactive_split.large_pool.allocated": 3618186,
60
+ "inactive_split.large_pool.current": 36,
61
+ "inactive_split.large_pool.freed": 3618150,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 11025370,
64
+ "inactive_split.small_pool.current": 80,
65
+ "inactive_split.small_pool.freed": 11025290,
66
+ "inactive_split.small_pool.peak": 135,
67
+ "inactive_split_bytes.all.allocated": 33392298644992,
68
+ "inactive_split_bytes.all.current": 128854528,
69
+ "inactive_split_bytes.all.freed": 33392169790464,
70
+ "inactive_split_bytes.all.peak": 548717056,
71
+ "inactive_split_bytes.large_pool.allocated": 23790269870592,
72
+ "inactive_split_bytes.large_pool.current": 92274688,
73
+ "inactive_split_bytes.large_pool.freed": 23790177595904,
74
+ "inactive_split_bytes.large_pool.peak": 518733824,
75
+ "inactive_split_bytes.small_pool.allocated": 9602028774400,
76
+ "inactive_split_bytes.small_pool.current": 36579840,
77
+ "inactive_split_bytes.small_pool.freed": 9601992194560,
78
+ "inactive_split_bytes.small_pool.peak": 90121728,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 573,
82
+ "num_device_free": 415,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 5,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 23931501523460,
94
+ "requested_bytes.all.current": 798501024,
95
+ "requested_bytes.all.freed": 23930703022436,
96
+ "requested_bytes.all.peak": 1396203552,
97
+ "requested_bytes.large_pool.allocated": 15288270045248,
98
+ "requested_bytes.large_pool.current": 795392000,
99
+ "requested_bytes.large_pool.freed": 15287474653248,
100
+ "requested_bytes.large_pool.peak": 1383348224,
101
+ "requested_bytes.small_pool.allocated": 8643231478212,
102
+ "requested_bytes.small_pool.current": 3109024,
103
+ "requested_bytes.small_pool.freed": 8643228369188,
104
+ "requested_bytes.small_pool.peak": 118941096,
105
+ "reserved_bytes.all.allocated": 5521801216,
106
+ "reserved_bytes.all.current": 2021654528,
107
+ "reserved_bytes.all.freed": 3500146688,
108
+ "reserved_bytes.all.peak": 2021654528,
109
+ "reserved_bytes.large_pool.allocated": 4875878400,
110
+ "reserved_bytes.large_pool.current": 1897922560,
111
+ "reserved_bytes.large_pool.freed": 2977955840,
112
+ "reserved_bytes.large_pool.peak": 1897922560,
113
+ "reserved_bytes.small_pool.allocated": 645922816,
114
+ "reserved_bytes.small_pool.current": 123731968,
115
+ "reserved_bytes.small_pool.freed": 522190848,
116
+ "reserved_bytes.small_pool.peak": 123731968,
117
+ "segment.all.allocated": 573,
118
+ "segment.all.current": 158,
119
+ "segment.all.freed": 415,
120
+ "segment.all.peak": 158,
121
+ "segment.large_pool.allocated": 265,
122
+ "segment.large_pool.current": 99,
123
+ "segment.large_pool.freed": 166,
124
+ "segment.large_pool.peak": 99,
125
+ "segment.small_pool.allocated": 308,
126
+ "segment.small_pool.current": 59,
127
+ "segment.small_pool.freed": 249,
128
+ "segment.small_pool.peak": 59
129
+ }
130
+ }
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.8990825688073395}
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 802280960,
3
+ "max_memory_allocated": 1411666432,
4
+ "memory_reserved": 2021654528,
5
+ "max_memory_reserved": 2021654528,
6
+ "memory_stats": {
7
+ "active.all.allocated": 40242578,
8
+ "active.all.current": 919,
9
+ "active.all.freed": 40241659,
10
+ "active.all.peak": 1093,
11
+ "active.large_pool.allocated": 9026394,
12
+ "active.large_pool.current": 222,
13
+ "active.large_pool.freed": 9026172,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 31216184,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 31215487,
18
+ "active.small_pool.peak": 857,
19
+ "active_bytes.all.allocated": 32976274128896,
20
+ "active_bytes.all.current": 802280960,
21
+ "active_bytes.all.freed": 32975471847936,
22
+ "active_bytes.all.peak": 1411666432,
23
+ "active_bytes.large_pool.allocated": 21449107161088,
24
+ "active_bytes.large_pool.current": 799014912,
25
+ "active_bytes.large_pool.freed": 21448308146176,
26
+ "active_bytes.large_pool.peak": 1398726656,
27
+ "active_bytes.small_pool.allocated": 11527166967808,
28
+ "active_bytes.small_pool.current": 3266048,
29
+ "active_bytes.small_pool.freed": 11527163701760,
30
+ "active_bytes.small_pool.peak": 119100928,
31
+ "allocated_bytes.all.allocated": 32976274128896,
32
+ "allocated_bytes.all.current": 802280960,
33
+ "allocated_bytes.all.freed": 32975471847936,
34
+ "allocated_bytes.all.peak": 1411666432,
35
+ "allocated_bytes.large_pool.allocated": 21449107161088,
36
+ "allocated_bytes.large_pool.current": 799014912,
37
+ "allocated_bytes.large_pool.freed": 21448308146176,
38
+ "allocated_bytes.large_pool.peak": 1398726656,
39
+ "allocated_bytes.small_pool.allocated": 11527166967808,
40
+ "allocated_bytes.small_pool.current": 3266048,
41
+ "allocated_bytes.small_pool.freed": 11527163701760,
42
+ "allocated_bytes.small_pool.peak": 119100928,
43
+ "allocation.all.allocated": 40242578,
44
+ "allocation.all.current": 919,
45
+ "allocation.all.freed": 40241659,
46
+ "allocation.all.peak": 1093,
47
+ "allocation.large_pool.allocated": 9026394,
48
+ "allocation.large_pool.current": 222,
49
+ "allocation.large_pool.freed": 9026172,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 31216184,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 31215487,
54
+ "allocation.small_pool.peak": 857,
55
+ "inactive_split.all.allocated": 19525219,
56
+ "inactive_split.all.current": 116,
57
+ "inactive_split.all.freed": 19525103,
58
+ "inactive_split.all.peak": 208,
59
+ "inactive_split.large_pool.allocated": 4837962,
60
+ "inactive_split.large_pool.current": 36,
61
+ "inactive_split.large_pool.freed": 4837926,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 14687257,
64
+ "inactive_split.small_pool.current": 80,
65
+ "inactive_split.small_pool.freed": 14687177,
66
+ "inactive_split.small_pool.peak": 135,
67
+ "inactive_split_bytes.all.allocated": 44797836460032,
68
+ "inactive_split_bytes.all.current": 128854528,
69
+ "inactive_split_bytes.all.freed": 44797707605504,
70
+ "inactive_split_bytes.all.peak": 548717056,
71
+ "inactive_split_bytes.large_pool.allocated": 31994153879552,
72
+ "inactive_split_bytes.large_pool.current": 92274688,
73
+ "inactive_split_bytes.large_pool.freed": 31994061604864,
74
+ "inactive_split_bytes.large_pool.peak": 518733824,
75
+ "inactive_split_bytes.small_pool.allocated": 12803682580480,
76
+ "inactive_split_bytes.small_pool.current": 36579840,
77
+ "inactive_split_bytes.small_pool.freed": 12803646000640,
78
+ "inactive_split_bytes.small_pool.peak": 90121728,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 754,
82
+ "num_device_free": 596,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 7,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 31908663018160,
94
+ "requested_bytes.all.current": 798501024,
95
+ "requested_bytes.all.freed": 31907864517136,
96
+ "requested_bytes.all.peak": 1396203552,
97
+ "requested_bytes.large_pool.allocated": 20384354380544,
98
+ "requested_bytes.large_pool.current": 795392000,
99
+ "requested_bytes.large_pool.freed": 20383558988544,
100
+ "requested_bytes.large_pool.peak": 1383348224,
101
+ "requested_bytes.small_pool.allocated": 11524308637616,
102
+ "requested_bytes.small_pool.current": 3109024,
103
+ "requested_bytes.small_pool.freed": 11524305528592,
104
+ "requested_bytes.small_pool.peak": 118941096,
105
+ "reserved_bytes.all.allocated": 7140802560,
106
+ "reserved_bytes.all.current": 2021654528,
107
+ "reserved_bytes.all.freed": 5119148032,
108
+ "reserved_bytes.all.peak": 2021654528,
109
+ "reserved_bytes.large_pool.allocated": 6280970240,
110
+ "reserved_bytes.large_pool.current": 1897922560,
111
+ "reserved_bytes.large_pool.freed": 4383047680,
112
+ "reserved_bytes.large_pool.peak": 1897922560,
113
+ "reserved_bytes.small_pool.allocated": 859832320,
114
+ "reserved_bytes.small_pool.current": 123731968,
115
+ "reserved_bytes.small_pool.freed": 736100352,
116
+ "reserved_bytes.small_pool.peak": 123731968,
117
+ "segment.all.allocated": 754,
118
+ "segment.all.current": 158,
119
+ "segment.all.freed": 596,
120
+ "segment.all.peak": 158,
121
+ "segment.large_pool.allocated": 344,
122
+ "segment.large_pool.current": 99,
123
+ "segment.large_pool.freed": 245,
124
+ "segment.large_pool.peak": 99,
125
+ "segment.small_pool.allocated": 410,
126
+ "segment.small_pool.current": 59,
127
+ "segment.small_pool.freed": 351,
128
+ "segment.small_pool.peak": 59
129
+ }
130
+ }
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9059633027522935}
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 802280960,
3
+ "max_memory_allocated": 1411666432,
4
+ "memory_reserved": 2021654528,
5
+ "max_memory_reserved": 2021654528,
6
+ "memory_stats": {
7
+ "active.all.allocated": 50303222,
8
+ "active.all.current": 919,
9
+ "active.all.freed": 50302303,
10
+ "active.all.peak": 1093,
11
+ "active.large_pool.allocated": 11282992,
12
+ "active.large_pool.current": 222,
13
+ "active.large_pool.freed": 11282770,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 39020230,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 39019533,
18
+ "active.small_pool.peak": 857,
19
+ "active_bytes.all.allocated": 41221900472832,
20
+ "active_bytes.all.current": 802280960,
21
+ "active_bytes.all.freed": 41221098191872,
22
+ "active_bytes.all.peak": 1411666432,
23
+ "active_bytes.large_pool.allocated": 26812941763072,
24
+ "active_bytes.large_pool.current": 799014912,
25
+ "active_bytes.large_pool.freed": 26812142748160,
26
+ "active_bytes.large_pool.peak": 1398726656,
27
+ "active_bytes.small_pool.allocated": 14408958709760,
28
+ "active_bytes.small_pool.current": 3266048,
29
+ "active_bytes.small_pool.freed": 14408955443712,
30
+ "active_bytes.small_pool.peak": 119100928,
31
+ "allocated_bytes.all.allocated": 41221900472832,
32
+ "allocated_bytes.all.current": 802280960,
33
+ "allocated_bytes.all.freed": 41221098191872,
34
+ "allocated_bytes.all.peak": 1411666432,
35
+ "allocated_bytes.large_pool.allocated": 26812941763072,
36
+ "allocated_bytes.large_pool.current": 799014912,
37
+ "allocated_bytes.large_pool.freed": 26812142748160,
38
+ "allocated_bytes.large_pool.peak": 1398726656,
39
+ "allocated_bytes.small_pool.allocated": 14408958709760,
40
+ "allocated_bytes.small_pool.current": 3266048,
41
+ "allocated_bytes.small_pool.freed": 14408955443712,
42
+ "allocated_bytes.small_pool.peak": 119100928,
43
+ "allocation.all.allocated": 50303222,
44
+ "allocation.all.current": 919,
45
+ "allocation.all.freed": 50302303,
46
+ "allocation.all.peak": 1093,
47
+ "allocation.large_pool.allocated": 11282992,
48
+ "allocation.large_pool.current": 222,
49
+ "allocation.large_pool.freed": 11282770,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 39020230,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 39019533,
54
+ "allocation.small_pool.peak": 857,
55
+ "inactive_split.all.allocated": 24405842,
56
+ "inactive_split.all.current": 113,
57
+ "inactive_split.all.freed": 24405729,
58
+ "inactive_split.all.peak": 208,
59
+ "inactive_split.large_pool.allocated": 6057738,
60
+ "inactive_split.large_pool.current": 36,
61
+ "inactive_split.large_pool.freed": 6057702,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 18348104,
64
+ "inactive_split.small_pool.current": 77,
65
+ "inactive_split.small_pool.freed": 18348027,
66
+ "inactive_split.small_pool.peak": 135,
67
+ "inactive_split_bytes.all.allocated": 56203426969088,
68
+ "inactive_split_bytes.all.current": 124660224,
69
+ "inactive_split_bytes.all.freed": 56203302308864,
70
+ "inactive_split_bytes.all.peak": 548717056,
71
+ "inactive_split_bytes.large_pool.allocated": 40198037888512,
72
+ "inactive_split_bytes.large_pool.current": 92274688,
73
+ "inactive_split_bytes.large_pool.freed": 40197945613824,
74
+ "inactive_split_bytes.large_pool.peak": 518733824,
75
+ "inactive_split_bytes.small_pool.allocated": 16005389080576,
76
+ "inactive_split_bytes.small_pool.current": 32385536,
77
+ "inactive_split_bytes.small_pool.freed": 16005356695040,
78
+ "inactive_split_bytes.small_pool.peak": 90121728,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 935,
82
+ "num_device_free": 777,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 9,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 39885824512860,
94
+ "requested_bytes.all.current": 798501024,
95
+ "requested_bytes.all.freed": 39885026011836,
96
+ "requested_bytes.all.peak": 1396203552,
97
+ "requested_bytes.large_pool.allocated": 25480438715840,
98
+ "requested_bytes.large_pool.current": 795392000,
99
+ "requested_bytes.large_pool.freed": 25479643323840,
100
+ "requested_bytes.large_pool.peak": 1383348224,
101
+ "requested_bytes.small_pool.allocated": 14405385797020,
102
+ "requested_bytes.small_pool.current": 3109024,
103
+ "requested_bytes.small_pool.freed": 14405382687996,
104
+ "requested_bytes.small_pool.peak": 118941096,
105
+ "reserved_bytes.all.allocated": 8759803904,
106
+ "reserved_bytes.all.current": 2021654528,
107
+ "reserved_bytes.all.freed": 6738149376,
108
+ "reserved_bytes.all.peak": 2021654528,
109
+ "reserved_bytes.large_pool.allocated": 7686062080,
110
+ "reserved_bytes.large_pool.current": 1897922560,
111
+ "reserved_bytes.large_pool.freed": 5788139520,
112
+ "reserved_bytes.large_pool.peak": 1897922560,
113
+ "reserved_bytes.small_pool.allocated": 1073741824,
114
+ "reserved_bytes.small_pool.current": 123731968,
115
+ "reserved_bytes.small_pool.freed": 950009856,
116
+ "reserved_bytes.small_pool.peak": 123731968,
117
+ "segment.all.allocated": 935,
118
+ "segment.all.current": 158,
119
+ "segment.all.freed": 777,
120
+ "segment.all.peak": 158,
121
+ "segment.large_pool.allocated": 423,
122
+ "segment.large_pool.current": 99,
123
+ "segment.large_pool.freed": 324,
124
+ "segment.large_pool.peak": 99,
125
+ "segment.small_pool.allocated": 512,
126
+ "segment.small_pool.current": 59,
127
+ "segment.small_pool.freed": 453,
128
+ "segment.small_pool.peak": 59
129
+ }
130
+ }
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9036697247706422}
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/bert-base-uncased_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_9999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 802280960,
3
+ "max_memory_allocated": 1411666432,
4
+ "memory_reserved": 2025848832,
5
+ "max_memory_reserved": 2025848832,
6
+ "memory_stats": {
7
+ "active.all.allocated": 60363866,
8
+ "active.all.current": 919,
9
+ "active.all.freed": 60362947,
10
+ "active.all.peak": 1093,
11
+ "active.large_pool.allocated": 13539590,
12
+ "active.large_pool.current": 222,
13
+ "active.large_pool.freed": 13539368,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 46824276,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 46823579,
18
+ "active.small_pool.peak": 857,
19
+ "active_bytes.all.allocated": 49467526816768,
20
+ "active_bytes.all.current": 802280960,
21
+ "active_bytes.all.freed": 49466724535808,
22
+ "active_bytes.all.peak": 1411666432,
23
+ "active_bytes.large_pool.allocated": 32176776365056,
24
+ "active_bytes.large_pool.current": 799014912,
25
+ "active_bytes.large_pool.freed": 32175977350144,
26
+ "active_bytes.large_pool.peak": 1398726656,
27
+ "active_bytes.small_pool.allocated": 17290750451712,
28
+ "active_bytes.small_pool.current": 3266048,
29
+ "active_bytes.small_pool.freed": 17290747185664,
30
+ "active_bytes.small_pool.peak": 119100928,
31
+ "allocated_bytes.all.allocated": 49467526816768,
32
+ "allocated_bytes.all.current": 802280960,
33
+ "allocated_bytes.all.freed": 49466724535808,
34
+ "allocated_bytes.all.peak": 1411666432,
35
+ "allocated_bytes.large_pool.allocated": 32176776365056,
36
+ "allocated_bytes.large_pool.current": 799014912,
37
+ "allocated_bytes.large_pool.freed": 32175977350144,
38
+ "allocated_bytes.large_pool.peak": 1398726656,
39
+ "allocated_bytes.small_pool.allocated": 17290750451712,
40
+ "allocated_bytes.small_pool.current": 3266048,
41
+ "allocated_bytes.small_pool.freed": 17290747185664,
42
+ "allocated_bytes.small_pool.peak": 119100928,
43
+ "allocation.all.allocated": 60363866,
44
+ "allocation.all.current": 919,
45
+ "allocation.all.freed": 60362947,
46
+ "allocation.all.peak": 1093,
47
+ "allocation.large_pool.allocated": 13539590,
48
+ "allocation.large_pool.current": 222,
49
+ "allocation.large_pool.freed": 13539368,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 46824276,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 46823579,
54
+ "allocation.small_pool.peak": 857,
55
+ "inactive_split.all.allocated": 29285843,
56
+ "inactive_split.all.current": 110,
57
+ "inactive_split.all.freed": 29285733,
58
+ "inactive_split.all.peak": 208,
59
+ "inactive_split.large_pool.allocated": 7277514,
60
+ "inactive_split.large_pool.current": 36,
61
+ "inactive_split.large_pool.freed": 7277478,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 22008329,
64
+ "inactive_split.small_pool.current": 74,
65
+ "inactive_split.small_pool.freed": 22008255,
66
+ "inactive_split.small_pool.peak": 135,
67
+ "inactive_split_bytes.all.allocated": 67609281007104,
68
+ "inactive_split_bytes.all.current": 120465920,
69
+ "inactive_split_bytes.all.freed": 67609160541184,
70
+ "inactive_split_bytes.all.peak": 548717056,
71
+ "inactive_split_bytes.large_pool.allocated": 48401921897472,
72
+ "inactive_split_bytes.large_pool.current": 92274688,
73
+ "inactive_split_bytes.large_pool.freed": 48401829622784,
74
+ "inactive_split_bytes.large_pool.peak": 518733824,
75
+ "inactive_split_bytes.small_pool.allocated": 19207359109632,
76
+ "inactive_split_bytes.small_pool.current": 28191232,
77
+ "inactive_split_bytes.small_pool.freed": 19207330918400,
78
+ "inactive_split_bytes.small_pool.peak": 91187712,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 1118,
82
+ "num_device_free": 958,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 11,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 47862986007560,
94
+ "requested_bytes.all.current": 798501024,
95
+ "requested_bytes.all.freed": 47862187506536,
96
+ "requested_bytes.all.peak": 1396203552,
97
+ "requested_bytes.large_pool.allocated": 30576523051136,
98
+ "requested_bytes.large_pool.current": 795392000,
99
+ "requested_bytes.large_pool.freed": 30575727659136,
100
+ "requested_bytes.large_pool.peak": 1383348224,
101
+ "requested_bytes.small_pool.allocated": 17286462956424,
102
+ "requested_bytes.small_pool.current": 3109024,
103
+ "requested_bytes.small_pool.freed": 17286459847400,
104
+ "requested_bytes.small_pool.peak": 118941096,
105
+ "reserved_bytes.all.allocated": 10382999552,
106
+ "reserved_bytes.all.current": 2025848832,
107
+ "reserved_bytes.all.freed": 8357150720,
108
+ "reserved_bytes.all.peak": 2025848832,
109
+ "reserved_bytes.large_pool.allocated": 9091153920,
110
+ "reserved_bytes.large_pool.current": 1897922560,
111
+ "reserved_bytes.large_pool.freed": 7193231360,
112
+ "reserved_bytes.large_pool.peak": 1897922560,
113
+ "reserved_bytes.small_pool.allocated": 1291845632,
114
+ "reserved_bytes.small_pool.current": 127926272,
115
+ "reserved_bytes.small_pool.freed": 1163919360,
116
+ "reserved_bytes.small_pool.peak": 127926272,
117
+ "segment.all.allocated": 1118,
118
+ "segment.all.current": 160,
119
+ "segment.all.freed": 958,
120
+ "segment.all.peak": 160,
121
+ "segment.large_pool.allocated": 502,
122
+ "segment.large_pool.current": 99,
123
+ "segment.large_pool.freed": 403,
124
+ "segment.large_pool.peak": 99,
125
+ "segment.small_pool.allocated": 616,
126
+ "segment.small_pool.current": 61,
127
+ "segment.small_pool.freed": 555,
128
+ "segment.small_pool.peak": 61
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/logfile_la_all.log ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 06/01/2024 03:44:31 - INFO - __main__ - ***** Starting script *****
2
+ 06/01/2024 03:44:36 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
3
+ 06/01/2024 03:44:36 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
4
+ 06/01/2024 03:44:36 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
5
+ 06/01/2024 04:12:50 - INFO - __main__ - ***** Completed Script *****
6
+ 06/01/2024 04:12:53 - INFO - __main__ - ***** Starting script *****
7
+ 06/01/2024 04:12:59 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
8
+ 06/01/2024 04:12:59 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
9
+ 06/01/2024 04:12:59 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
10
+ 06/01/2024 04:42:12 - INFO - __main__ - ***** Completed Script *****
11
+ 06/01/2024 04:42:14 - INFO - __main__ - ***** Starting script *****
12
+ 06/01/2024 04:42:20 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
13
+ 06/01/2024 04:42:20 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
14
+ 06/01/2024 04:42:20 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
15
+ 06/01/2024 05:11:43 - INFO - __main__ - ***** Completed Script *****
16
+ 06/01/2024 05:11:46 - INFO - __main__ - ***** Starting script *****
17
+ 06/01/2024 05:11:51 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
18
+ 06/01/2024 05:11:51 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
19
+ 06/01/2024 05:11:51 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
20
+ 06/01/2024 05:41:13 - INFO - __main__ - ***** Completed Script *****
21
+ 06/01/2024 05:41:15 - INFO - __main__ - ***** Starting script *****
22
+ 06/01/2024 05:41:20 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
23
+ 06/01/2024 05:41:20 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
24
+ 06/01/2024 05:41:20 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
25
+ 06/01/2024 06:10:35 - INFO - __main__ - ***** Completed Script *****
26
+ 06/01/2024 06:10:47 - INFO - __main__ - ***** Starting script *****
27
+ 06/01/2024 06:10:54 - INFO - __main__ - Sample 27303 of the training set: {'input_ids': [0, 10273, 5971, 2650, 2156, 114, 3694, 2156, 29166, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
28
+ 06/01/2024 06:10:54 - INFO - __main__ - Sample 48017 of the training set: {'input_ids': [0, 2629, 2526, 9288, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1], 'labels': 0}.
29
+ 06/01/2024 06:10:54 - INFO - __main__ - Sample 666 of the training set: {'input_ids': [0, 26692, 45, 173, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
30
+ 06/01/2024 06:40:42 - INFO - __main__ - ***** Completed Script *****
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.4908256880733945}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_0/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 874962944,
3
+ "max_memory_allocated": 1567147520,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 11257151,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 11256231,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 2240254,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 2240031,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 9016897,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 9016200,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 9053779064320,
20
+ "active_bytes.all.current": 874962944,
21
+ "active_bytes.all.freed": 9052904101376,
22
+ "active_bytes.all.peak": 1567147520,
23
+ "active_bytes.large_pool.allocated": 5293802369024,
24
+ "active_bytes.large_pool.current": 871698432,
25
+ "active_bytes.large_pool.freed": 5292930670592,
26
+ "active_bytes.large_pool.peak": 1554066432,
27
+ "active_bytes.small_pool.allocated": 3759976695296,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 3759973430784,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 9053779064320,
32
+ "allocated_bytes.all.current": 874962944,
33
+ "allocated_bytes.all.freed": 9052904101376,
34
+ "allocated_bytes.all.peak": 1567147520,
35
+ "allocated_bytes.large_pool.allocated": 5293802369024,
36
+ "allocated_bytes.large_pool.current": 871698432,
37
+ "allocated_bytes.large_pool.freed": 5292930670592,
38
+ "allocated_bytes.large_pool.peak": 1554066432,
39
+ "allocated_bytes.small_pool.allocated": 3759976695296,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 3759973430784,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 11257151,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 11256231,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 2240254,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 2240031,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 9016897,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 9016200,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 5859237,
56
+ "inactive_split.all.current": 96,
57
+ "inactive_split.all.freed": 5859141,
58
+ "inactive_split.all.peak": 178,
59
+ "inactive_split.large_pool.allocated": 1170506,
60
+ "inactive_split.large_pool.current": 33,
61
+ "inactive_split.large_pool.freed": 1170473,
62
+ "inactive_split.large_pool.peak": 96,
63
+ "inactive_split.small_pool.allocated": 4688731,
64
+ "inactive_split.small_pool.current": 63,
65
+ "inactive_split.small_pool.freed": 4688668,
66
+ "inactive_split.small_pool.peak": 103,
67
+ "inactive_split_bytes.all.allocated": 11421352003584,
68
+ "inactive_split_bytes.all.current": 123281408,
69
+ "inactive_split_bytes.all.freed": 11421228722176,
70
+ "inactive_split_bytes.all.peak": 525084160,
71
+ "inactive_split_bytes.large_pool.allocated": 7348056296960,
72
+ "inactive_split_bytes.large_pool.current": 80408576,
73
+ "inactive_split_bytes.large_pool.freed": 7347975888384,
74
+ "inactive_split_bytes.large_pool.peak": 497114112,
75
+ "inactive_split_bytes.small_pool.allocated": 4073295706624,
76
+ "inactive_split_bytes.small_pool.current": 42872832,
77
+ "inactive_split_bytes.small_pool.freed": 4073252833792,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 202,
82
+ "num_device_free": 43,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 1,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 8815018579287,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 8814157064343,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 5055758041280,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 5054899633344,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 3759260538007,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 3759257430999,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 2338324480,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 222298112,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 2143289344,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 146800640,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 195035136,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 75497472,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 202,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 43,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 109,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 7,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 93,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 36,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9162844036697247}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_1999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 876204032,
3
+ "max_memory_allocated": 1568388608,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 22514300,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 22513380,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 4480506,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 4480283,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 18033794,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 18033097,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 18098342233600,
20
+ "active_bytes.all.current": 876204032,
21
+ "active_bytes.all.freed": 18097466029568,
22
+ "active_bytes.all.peak": 1568388608,
23
+ "active_bytes.large_pool.allocated": 10578388843008,
24
+ "active_bytes.large_pool.current": 872939520,
25
+ "active_bytes.large_pool.freed": 10577515903488,
26
+ "active_bytes.large_pool.peak": 1555307520,
27
+ "active_bytes.small_pool.allocated": 7519953390592,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 7519950126080,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 18098342233600,
32
+ "allocated_bytes.all.current": 876204032,
33
+ "allocated_bytes.all.freed": 18097466029568,
34
+ "allocated_bytes.all.peak": 1568388608,
35
+ "allocated_bytes.large_pool.allocated": 10578388843008,
36
+ "allocated_bytes.large_pool.current": 872939520,
37
+ "allocated_bytes.large_pool.freed": 10577515903488,
38
+ "allocated_bytes.large_pool.peak": 1555307520,
39
+ "allocated_bytes.small_pool.allocated": 7519953390592,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 7519950126080,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 22514300,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 22513380,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 4480506,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 4480283,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 18033794,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 18033097,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 11724345,
56
+ "inactive_split.all.current": 91,
57
+ "inactive_split.all.freed": 11724254,
58
+ "inactive_split.all.peak": 178,
59
+ "inactive_split.large_pool.allocated": 2337827,
60
+ "inactive_split.large_pool.current": 32,
61
+ "inactive_split.large_pool.freed": 2337795,
62
+ "inactive_split.large_pool.peak": 105,
63
+ "inactive_split.small_pool.allocated": 9386518,
64
+ "inactive_split.small_pool.current": 59,
65
+ "inactive_split.small_pool.freed": 9386459,
66
+ "inactive_split.small_pool.peak": 118,
67
+ "inactive_split_bytes.all.allocated": 23009216232960,
68
+ "inactive_split_bytes.all.current": 111554560,
69
+ "inactive_split_bytes.all.freed": 23009104678400,
70
+ "inactive_split_bytes.all.peak": 530060288,
71
+ "inactive_split_bytes.large_pool.allocated": 14858685626880,
72
+ "inactive_split_bytes.large_pool.current": 79167488,
73
+ "inactive_split_bytes.large_pool.freed": 14858606459392,
74
+ "inactive_split_bytes.large_pool.peak": 508379648,
75
+ "inactive_split_bytes.small_pool.allocated": 8150530606080,
76
+ "inactive_split_bytes.small_pool.current": 32387072,
77
+ "inactive_split_bytes.small_pool.freed": 8150498219008,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 366,
82
+ "num_device_free": 207,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 3,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 17630020119214,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 17629158604270,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 10111499043200,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 10110640635264,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 7518521076014,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 7518517969006,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 3915382784,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 1799356416,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 3544186880,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 1547698176,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 371195904,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 251658240,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 366,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 207,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 189,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 87,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 177,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 120,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9277522935779816}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_3999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 876204032,
3
+ "max_memory_allocated": 1568388608,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 33771449,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 33770529,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 6720758,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 6720535,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 27050691,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 27049994,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 27142881768960,
20
+ "active_bytes.all.current": 876204032,
21
+ "active_bytes.all.freed": 27142005564928,
22
+ "active_bytes.all.peak": 1568388608,
23
+ "active_bytes.large_pool.allocated": 15862951683072,
24
+ "active_bytes.large_pool.current": 872939520,
25
+ "active_bytes.large_pool.freed": 15862078743552,
26
+ "active_bytes.large_pool.peak": 1555307520,
27
+ "active_bytes.small_pool.allocated": 11279930085888,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 11279926821376,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 27142881768960,
32
+ "allocated_bytes.all.current": 876204032,
33
+ "allocated_bytes.all.freed": 27142005564928,
34
+ "allocated_bytes.all.peak": 1568388608,
35
+ "allocated_bytes.large_pool.allocated": 15862951683072,
36
+ "allocated_bytes.large_pool.current": 872939520,
37
+ "allocated_bytes.large_pool.freed": 15862078743552,
38
+ "allocated_bytes.large_pool.peak": 1555307520,
39
+ "allocated_bytes.small_pool.allocated": 11279930085888,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 11279926821376,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 33771449,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 33770529,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 6720758,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 6720535,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 27050691,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 27049994,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 17589387,
56
+ "inactive_split.all.current": 91,
57
+ "inactive_split.all.freed": 17589296,
58
+ "inactive_split.all.peak": 178,
59
+ "inactive_split.large_pool.allocated": 3505084,
60
+ "inactive_split.large_pool.current": 32,
61
+ "inactive_split.large_pool.freed": 3505052,
62
+ "inactive_split.large_pool.peak": 105,
63
+ "inactive_split.small_pool.allocated": 14084303,
64
+ "inactive_split.small_pool.current": 59,
65
+ "inactive_split.small_pool.freed": 14084244,
66
+ "inactive_split.small_pool.peak": 118,
67
+ "inactive_split_bytes.all.allocated": 34597024160256,
68
+ "inactive_split_bytes.all.current": 111554560,
69
+ "inactive_split_bytes.all.freed": 34596912605696,
70
+ "inactive_split_bytes.all.peak": 530060288,
71
+ "inactive_split_bytes.large_pool.allocated": 22369258933760,
72
+ "inactive_split_bytes.large_pool.current": 79167488,
73
+ "inactive_split_bytes.large_pool.freed": 22369179766272,
74
+ "inactive_split_bytes.large_pool.peak": 508379648,
75
+ "inactive_split_bytes.small_pool.allocated": 12227765226496,
76
+ "inactive_split_bytes.small_pool.current": 32387072,
77
+ "inactive_split_bytes.small_pool.freed": 12227732839424,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 533,
82
+ "num_device_free": 374,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 5,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 26445021659141,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 26444160144197,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 15167240045120,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 15166381637184,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 11277781614021,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 11277778507013,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 5498732544,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 3382706176,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 4945084416,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 2948595712,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 553648128,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 434110464,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 533,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 374,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 269,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 167,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 264,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 207,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9243119266055045}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_5999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 876204032,
3
+ "max_memory_allocated": 1568388608,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 45028598,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 45027678,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 8961010,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 8960787,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 36067588,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 36066891,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 36187428554240,
20
+ "active_bytes.all.current": 876204032,
21
+ "active_bytes.all.freed": 36186552350208,
22
+ "active_bytes.all.peak": 1568388608,
23
+ "active_bytes.large_pool.allocated": 21147521773056,
24
+ "active_bytes.large_pool.current": 872939520,
25
+ "active_bytes.large_pool.freed": 21146648833536,
26
+ "active_bytes.large_pool.peak": 1555307520,
27
+ "active_bytes.small_pool.allocated": 15039906781184,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 15039903516672,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 36187428554240,
32
+ "allocated_bytes.all.current": 876204032,
33
+ "allocated_bytes.all.freed": 36186552350208,
34
+ "allocated_bytes.all.peak": 1568388608,
35
+ "allocated_bytes.large_pool.allocated": 21147521773056,
36
+ "allocated_bytes.large_pool.current": 872939520,
37
+ "allocated_bytes.large_pool.freed": 21146648833536,
38
+ "allocated_bytes.large_pool.peak": 1555307520,
39
+ "allocated_bytes.small_pool.allocated": 15039906781184,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 15039903516672,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 45028598,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 45027678,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 8961010,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 8960787,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 36067588,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 36066891,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 23455503,
56
+ "inactive_split.all.current": 94,
57
+ "inactive_split.all.freed": 23455409,
58
+ "inactive_split.all.peak": 178,
59
+ "inactive_split.large_pool.allocated": 4672345,
60
+ "inactive_split.large_pool.current": 32,
61
+ "inactive_split.large_pool.freed": 4672313,
62
+ "inactive_split.large_pool.peak": 105,
63
+ "inactive_split.small_pool.allocated": 18783158,
64
+ "inactive_split.small_pool.current": 62,
65
+ "inactive_split.small_pool.freed": 18783096,
66
+ "inactive_split.small_pool.peak": 118,
67
+ "inactive_split_bytes.all.allocated": 46184863010816,
68
+ "inactive_split_bytes.all.current": 105263104,
69
+ "inactive_split_bytes.all.freed": 46184757747712,
70
+ "inactive_split_bytes.all.peak": 530060288,
71
+ "inactive_split_bytes.large_pool.allocated": 29879845126656,
72
+ "inactive_split_bytes.large_pool.current": 79167488,
73
+ "inactive_split_bytes.large_pool.freed": 29879765959168,
74
+ "inactive_split_bytes.large_pool.peak": 508379648,
75
+ "inactive_split_bytes.small_pool.allocated": 16305017884160,
76
+ "inactive_split_bytes.small_pool.current": 26095616,
77
+ "inactive_split_bytes.small_pool.freed": 16304991788544,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 700,
82
+ "num_device_free": 541,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 7,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 35260023199068,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 35259161684124,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 20222981047040,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 20222122639104,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 15037042152028,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 15037039045020,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 7082082304,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 4966055936,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 6345981952,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 4349493248,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 736100352,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 616562688,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 700,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 541,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 349,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 247,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 351,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 294,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9288990825688074}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_7999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 876204032,
3
+ "max_memory_allocated": 1568388608,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 56285747,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 56284827,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 11201262,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 11201039,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 45084485,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 45083788,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 45231948534272,
20
+ "active_bytes.all.current": 876204032,
21
+ "active_bytes.all.freed": 45231072330240,
22
+ "active_bytes.all.peak": 1568388608,
23
+ "active_bytes.large_pool.allocated": 26432065057792,
24
+ "active_bytes.large_pool.current": 872939520,
25
+ "active_bytes.large_pool.freed": 26431192118272,
26
+ "active_bytes.large_pool.peak": 1555307520,
27
+ "active_bytes.small_pool.allocated": 18799883476480,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 18799880211968,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 45231948534272,
32
+ "allocated_bytes.all.current": 876204032,
33
+ "allocated_bytes.all.freed": 45231072330240,
34
+ "allocated_bytes.all.peak": 1568388608,
35
+ "allocated_bytes.large_pool.allocated": 26432065057792,
36
+ "allocated_bytes.large_pool.current": 872939520,
37
+ "allocated_bytes.large_pool.freed": 26431192118272,
38
+ "allocated_bytes.large_pool.peak": 1555307520,
39
+ "allocated_bytes.small_pool.allocated": 18799883476480,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 18799880211968,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 56285747,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 56284827,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 11201262,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 11201039,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 45084485,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 45083788,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 29320192,
56
+ "inactive_split.all.current": 95,
57
+ "inactive_split.all.freed": 29320097,
58
+ "inactive_split.all.peak": 184,
59
+ "inactive_split.large_pool.allocated": 5839595,
60
+ "inactive_split.large_pool.current": 32,
61
+ "inactive_split.large_pool.freed": 5839563,
62
+ "inactive_split.large_pool.peak": 105,
63
+ "inactive_split.small_pool.allocated": 23480597,
64
+ "inactive_split.small_pool.current": 63,
65
+ "inactive_split.small_pool.freed": 23480534,
66
+ "inactive_split.small_pool.peak": 118,
67
+ "inactive_split_bytes.all.allocated": 57772618551808,
68
+ "inactive_split_bytes.all.current": 113651712,
69
+ "inactive_split_bytes.all.freed": 57772504900096,
70
+ "inactive_split_bytes.all.peak": 530060288,
71
+ "inactive_split_bytes.large_pool.allocated": 37390352505856,
72
+ "inactive_split_bytes.large_pool.current": 79167488,
73
+ "inactive_split_bytes.large_pool.freed": 37390273338368,
74
+ "inactive_split_bytes.large_pool.peak": 508379648,
75
+ "inactive_split_bytes.small_pool.allocated": 20382266045952,
76
+ "inactive_split_bytes.small_pool.current": 34484224,
77
+ "inactive_split_bytes.small_pool.freed": 20382231561728,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 866,
82
+ "num_device_free": 707,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 9,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 44075024738995,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 44074163224051,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 25278722048960,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 25277863641024,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 18796302690035,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 18796299583027,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 8663334912,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 6547308544,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 7746879488,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 5750390784,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 916455424,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 796917760,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 866,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 707,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 429,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 327,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 437,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 380,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9243119266055045}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_12345_8_10000/step_9999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 876204032,
3
+ "max_memory_allocated": 1568388608,
4
+ "memory_reserved": 2116026368,
5
+ "max_memory_reserved": 2116026368,
6
+ "memory_stats": {
7
+ "active.all.allocated": 67542896,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 67541976,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 13441514,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 13441291,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 54101382,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 54100685,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 54276468514304,
20
+ "active_bytes.all.current": 876204032,
21
+ "active_bytes.all.freed": 54275592310272,
22
+ "active_bytes.all.peak": 1568388608,
23
+ "active_bytes.large_pool.allocated": 31716608342528,
24
+ "active_bytes.large_pool.current": 872939520,
25
+ "active_bytes.large_pool.freed": 31715735403008,
26
+ "active_bytes.large_pool.peak": 1555307520,
27
+ "active_bytes.small_pool.allocated": 22559860171776,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 22559856907264,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 54276468514304,
32
+ "allocated_bytes.all.current": 876204032,
33
+ "allocated_bytes.all.freed": 54275592310272,
34
+ "allocated_bytes.all.peak": 1568388608,
35
+ "allocated_bytes.large_pool.allocated": 31716608342528,
36
+ "allocated_bytes.large_pool.current": 872939520,
37
+ "allocated_bytes.large_pool.freed": 31715735403008,
38
+ "allocated_bytes.large_pool.peak": 1555307520,
39
+ "allocated_bytes.small_pool.allocated": 22559860171776,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 22559856907264,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 67542896,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 67541976,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 13441514,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 13441291,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 54101382,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 54100685,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 35186752,
56
+ "inactive_split.all.current": 103,
57
+ "inactive_split.all.freed": 35186649,
58
+ "inactive_split.all.peak": 184,
59
+ "inactive_split.large_pool.allocated": 7006845,
60
+ "inactive_split.large_pool.current": 32,
61
+ "inactive_split.large_pool.freed": 7006813,
62
+ "inactive_split.large_pool.peak": 105,
63
+ "inactive_split.small_pool.allocated": 28179907,
64
+ "inactive_split.small_pool.current": 71,
65
+ "inactive_split.small_pool.freed": 28179836,
66
+ "inactive_split.small_pool.peak": 118,
67
+ "inactive_split_bytes.all.allocated": 69360367065600,
68
+ "inactive_split_bytes.all.current": 107360256,
69
+ "inactive_split_bytes.all.freed": 69360259705344,
70
+ "inactive_split_bytes.all.peak": 530060288,
71
+ "inactive_split_bytes.large_pool.allocated": 44900859885056,
72
+ "inactive_split_bytes.large_pool.current": 79167488,
73
+ "inactive_split_bytes.large_pool.freed": 44900780717568,
74
+ "inactive_split_bytes.large_pool.peak": 508379648,
75
+ "inactive_split_bytes.small_pool.allocated": 24459507180544,
76
+ "inactive_split_bytes.small_pool.current": 28192768,
77
+ "inactive_split_bytes.small_pool.freed": 24459478987776,
78
+ "inactive_split_bytes.small_pool.peak": 77594624,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 1032,
82
+ "num_device_free": 873,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 11,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 52890026278922,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 52889164763978,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 30334463050880,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 30333604642944,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 22555563228042,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 22555560121034,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 10244587520,
106
+ "reserved_bytes.all.current": 2116026368,
107
+ "reserved_bytes.all.freed": 8128561152,
108
+ "reserved_bytes.all.peak": 2116026368,
109
+ "reserved_bytes.large_pool.allocated": 9147777024,
110
+ "reserved_bytes.large_pool.current": 1996488704,
111
+ "reserved_bytes.large_pool.freed": 7151288320,
112
+ "reserved_bytes.large_pool.peak": 1996488704,
113
+ "reserved_bytes.small_pool.allocated": 1096810496,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 977272832,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 1032,
118
+ "segment.all.current": 159,
119
+ "segment.all.freed": 873,
120
+ "segment.all.peak": 159,
121
+ "segment.large_pool.allocated": 509,
122
+ "segment.large_pool.current": 102,
123
+ "segment.large_pool.freed": 407,
124
+ "segment.large_pool.peak": 102,
125
+ "segment.small_pool.allocated": 523,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 466,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/logfile_la_all.log ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 06/01/2024 06:40:58 - INFO - __main__ - ***** Starting script *****
2
+ 06/01/2024 06:41:04 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
3
+ 06/01/2024 06:41:04 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
4
+ 06/01/2024 06:41:04 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
5
+ 06/01/2024 07:10:31 - INFO - __main__ - ***** Completed Script *****
6
+ 06/01/2024 07:10:35 - INFO - __main__ - ***** Starting script *****
7
+ 06/01/2024 07:10:40 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
8
+ 06/01/2024 07:10:40 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
9
+ 06/01/2024 07:10:40 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
10
+ 06/01/2024 07:40:56 - INFO - __main__ - ***** Completed Script *****
11
+ 06/01/2024 07:40:58 - INFO - __main__ - ***** Starting script *****
12
+ 06/01/2024 07:41:03 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
13
+ 06/01/2024 07:41:03 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
14
+ 06/01/2024 07:41:03 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
15
+ 06/01/2024 08:12:14 - INFO - __main__ - ***** Completed Script *****
16
+ 06/01/2024 08:12:17 - INFO - __main__ - ***** Starting script *****
17
+ 06/01/2024 08:12:22 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
18
+ 06/01/2024 08:12:22 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
19
+ 06/01/2024 08:12:22 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
20
+ 06/01/2024 08:43:16 - INFO - __main__ - ***** Completed Script *****
21
+ 06/01/2024 08:43:19 - INFO - __main__ - ***** Starting script *****
22
+ 06/01/2024 08:43:25 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
23
+ 06/01/2024 08:43:25 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
24
+ 06/01/2024 08:43:25 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
25
+ 06/01/2024 09:14:25 - INFO - __main__ - ***** Completed Script *****
26
+ 06/01/2024 09:14:28 - INFO - __main__ - ***** Starting script *****
27
+ 06/01/2024 09:14:33 - INFO - __main__ - Sample 41905 of the training set: {'input_ids': [0, 17615, 1899, 385, 16314, 2156, 10985, 25, 7, 549, 47, 128, 548, 450, 15570, 50, 6717, 479, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
28
+ 06/01/2024 09:14:33 - INFO - __main__ - Sample 7296 of the training set: {'input_ids': [0, 1250, 5, 5567, 23959, 9, 41259, 11605, 23, 63, 275, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 1}.
29
+ 06/01/2024 09:14:33 - INFO - __main__ - Sample 1639 of the training set: {'input_ids': [0, 506, 11791, 98, 16894, 15, 358, 9164, 672, 14, 24, 35499, 103, 761, 9, 36302, 2821, 16235, 1437, 2], 'attention_mask': [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1], 'labels': 0}.
30
+ 06/01/2024 09:45:27 - INFO - __main__ - ***** Completed Script *****
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.4908256880733945}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_0/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 866340864,
3
+ "max_memory_allocated": 1561912320,
4
+ "memory_reserved": 2090860544,
5
+ "max_memory_reserved": 2090860544,
6
+ "memory_stats": {
7
+ "active.all.allocated": 11257151,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 11256231,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 2233224,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 2233001,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 9023927,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 9023230,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 9070472550912,
20
+ "active_bytes.all.current": 866340864,
21
+ "active_bytes.all.freed": 9069606210048,
22
+ "active_bytes.all.peak": 1561912320,
23
+ "active_bytes.large_pool.allocated": 5289135241216,
24
+ "active_bytes.large_pool.current": 863076352,
25
+ "active_bytes.large_pool.freed": 5288272164864,
26
+ "active_bytes.large_pool.peak": 1548831232,
27
+ "active_bytes.small_pool.allocated": 3781337309696,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 3781334045184,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 9070472550912,
32
+ "allocated_bytes.all.current": 866340864,
33
+ "allocated_bytes.all.freed": 9069606210048,
34
+ "allocated_bytes.all.peak": 1561912320,
35
+ "allocated_bytes.large_pool.allocated": 5289135241216,
36
+ "allocated_bytes.large_pool.current": 863076352,
37
+ "allocated_bytes.large_pool.freed": 5288272164864,
38
+ "allocated_bytes.large_pool.peak": 1548831232,
39
+ "allocated_bytes.small_pool.allocated": 3781337309696,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 3781334045184,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 11257151,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 11256231,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 2233224,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 2233001,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 9023927,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 9023230,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 5862295,
56
+ "inactive_split.all.current": 108,
57
+ "inactive_split.all.freed": 5862187,
58
+ "inactive_split.all.peak": 192,
59
+ "inactive_split.large_pool.allocated": 1187863,
60
+ "inactive_split.large_pool.current": 42,
61
+ "inactive_split.large_pool.freed": 1187821,
62
+ "inactive_split.large_pool.peak": 103,
63
+ "inactive_split.small_pool.allocated": 4674432,
64
+ "inactive_split.small_pool.current": 66,
65
+ "inactive_split.small_pool.freed": 4674366,
66
+ "inactive_split.small_pool.peak": 107,
67
+ "inactive_split_bytes.all.allocated": 11925857753088,
68
+ "inactive_split_bytes.all.current": 125612032,
69
+ "inactive_split_bytes.all.freed": 11925732141056,
70
+ "inactive_split_bytes.all.peak": 518794752,
71
+ "inactive_split_bytes.large_pool.allocated": 7829347278848,
72
+ "inactive_split_bytes.large_pool.current": 89030656,
73
+ "inactive_split_bytes.large_pool.freed": 7829258248192,
74
+ "inactive_split_bytes.large_pool.peak": 499212800,
75
+ "inactive_split_bytes.small_pool.allocated": 4096510474240,
76
+ "inactive_split_bytes.small_pool.current": 36581376,
77
+ "inactive_split_bytes.small_pool.freed": 4096473892864,
78
+ "inactive_split_bytes.small_pool.peak": 73389568,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 205,
82
+ "num_device_free": 48,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 1,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 8832782340291,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 8831920825347,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 5052159555776,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 5051301147840,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 3780622784515,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 3780619677507,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 2323644416,
106
+ "reserved_bytes.all.current": 2090860544,
107
+ "reserved_bytes.all.freed": 232783872,
108
+ "reserved_bytes.all.peak": 2090860544,
109
+ "reserved_bytes.large_pool.allocated": 2118123520,
110
+ "reserved_bytes.large_pool.current": 1971322880,
111
+ "reserved_bytes.large_pool.freed": 146800640,
112
+ "reserved_bytes.large_pool.peak": 1971322880,
113
+ "reserved_bytes.small_pool.allocated": 205520896,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 85983232,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 205,
118
+ "segment.all.current": 157,
119
+ "segment.all.freed": 48,
120
+ "segment.all.peak": 157,
121
+ "segment.large_pool.allocated": 107,
122
+ "segment.large_pool.current": 100,
123
+ "segment.large_pool.freed": 7,
124
+ "segment.large_pool.peak": 100,
125
+ "segment.small_pool.allocated": 98,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 41,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.911697247706422}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_1999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 866758656,
3
+ "max_memory_allocated": 1561953792,
4
+ "memory_reserved": 2090860544,
5
+ "max_memory_reserved": 2090860544,
6
+ "memory_stats": {
7
+ "active.all.allocated": 22514300,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 22513380,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 4466446,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 4466223,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 18047854,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 18047157,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 18132720948736,
20
+ "active_bytes.all.current": 866758656,
21
+ "active_bytes.all.freed": 18131854190080,
22
+ "active_bytes.all.peak": 1561953792,
23
+ "active_bytes.large_pool.allocated": 10570046329344,
24
+ "active_bytes.large_pool.current": 863494144,
25
+ "active_bytes.large_pool.freed": 10569182835200,
26
+ "active_bytes.large_pool.peak": 1548872704,
27
+ "active_bytes.small_pool.allocated": 7562674619392,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 7562671354880,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 18132720948736,
32
+ "allocated_bytes.all.current": 866758656,
33
+ "allocated_bytes.all.freed": 18131854190080,
34
+ "allocated_bytes.all.peak": 1561953792,
35
+ "allocated_bytes.large_pool.allocated": 10570046329344,
36
+ "allocated_bytes.large_pool.current": 863494144,
37
+ "allocated_bytes.large_pool.freed": 10569182835200,
38
+ "allocated_bytes.large_pool.peak": 1548872704,
39
+ "allocated_bytes.small_pool.allocated": 7562674619392,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 7562671354880,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 22514300,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 22513380,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 4466446,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 4466223,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 18047854,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 18047157,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 11663359,
56
+ "inactive_split.all.current": 105,
57
+ "inactive_split.all.freed": 11663254,
58
+ "inactive_split.all.peak": 192,
59
+ "inactive_split.large_pool.allocated": 2348602,
60
+ "inactive_split.large_pool.current": 45,
61
+ "inactive_split.large_pool.freed": 2348557,
62
+ "inactive_split.large_pool.peak": 106,
63
+ "inactive_split.small_pool.allocated": 9314757,
64
+ "inactive_split.small_pool.current": 60,
65
+ "inactive_split.small_pool.freed": 9314697,
66
+ "inactive_split.small_pool.peak": 121,
67
+ "inactive_split_bytes.all.allocated": 23531259524608,
68
+ "inactive_split_bytes.all.current": 137777152,
69
+ "inactive_split_bytes.all.freed": 23531121747456,
70
+ "inactive_split_bytes.all.peak": 518794752,
71
+ "inactive_split_bytes.large_pool.allocated": 15335652198400,
72
+ "inactive_split_bytes.large_pool.current": 109584384,
73
+ "inactive_split_bytes.large_pool.freed": 15335542614016,
74
+ "inactive_split_bytes.large_pool.peak": 499212800,
75
+ "inactive_split_bytes.small_pool.allocated": 8195607326208,
76
+ "inactive_split_bytes.small_pool.current": 28192768,
77
+ "inactive_split_bytes.small_pool.freed": 8195579133440,
78
+ "inactive_split_bytes.small_pool.peak": 73389568,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 373,
82
+ "num_device_free": 216,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 3,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 17665547641222,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 17664686126278,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 10104302072192,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 10103443664256,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 7561245569030,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 7561242462022,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 3869245440,
106
+ "reserved_bytes.all.current": 2090860544,
107
+ "reserved_bytes.all.freed": 1778384896,
108
+ "reserved_bytes.all.peak": 2090860544,
109
+ "reserved_bytes.large_pool.allocated": 3472883712,
110
+ "reserved_bytes.large_pool.current": 1971322880,
111
+ "reserved_bytes.large_pool.freed": 1501560832,
112
+ "reserved_bytes.large_pool.peak": 1971322880,
113
+ "reserved_bytes.small_pool.allocated": 396361728,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 276824064,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 373,
118
+ "segment.all.current": 157,
119
+ "segment.all.freed": 216,
120
+ "segment.all.peak": 157,
121
+ "segment.large_pool.allocated": 184,
122
+ "segment.large_pool.current": 100,
123
+ "segment.large_pool.freed": 84,
124
+ "segment.large_pool.peak": 100,
125
+ "segment.small_pool.allocated": 189,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 132,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9277522935779816}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_3999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 866758656,
3
+ "max_memory_allocated": 1561953792,
4
+ "memory_reserved": 2090860544,
5
+ "max_memory_reserved": 2090860544,
6
+ "memory_stats": {
7
+ "active.all.allocated": 33771449,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 33770529,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 6699668,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 6699445,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 27071781,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 27071084,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 27194969346560,
20
+ "active_bytes.all.current": 866758656,
21
+ "active_bytes.all.freed": 27194102587904,
22
+ "active_bytes.all.peak": 1561953792,
23
+ "active_bytes.large_pool.allocated": 15850957417472,
24
+ "active_bytes.large_pool.current": 863494144,
25
+ "active_bytes.large_pool.freed": 15850093923328,
26
+ "active_bytes.large_pool.peak": 1548872704,
27
+ "active_bytes.small_pool.allocated": 11344011929088,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 11344008664576,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 27194969346560,
32
+ "allocated_bytes.all.current": 866758656,
33
+ "allocated_bytes.all.freed": 27194102587904,
34
+ "allocated_bytes.all.peak": 1561953792,
35
+ "allocated_bytes.large_pool.allocated": 15850957417472,
36
+ "allocated_bytes.large_pool.current": 863494144,
37
+ "allocated_bytes.large_pool.freed": 15850093923328,
38
+ "allocated_bytes.large_pool.peak": 1548872704,
39
+ "allocated_bytes.small_pool.allocated": 11344011929088,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 11344008664576,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 33771449,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 33770529,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 6699668,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 6699445,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 27071781,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 27071084,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 17491813,
56
+ "inactive_split.all.current": 106,
57
+ "inactive_split.all.freed": 17491707,
58
+ "inactive_split.all.peak": 192,
59
+ "inactive_split.large_pool.allocated": 3509241,
60
+ "inactive_split.large_pool.current": 45,
61
+ "inactive_split.large_pool.freed": 3509196,
62
+ "inactive_split.large_pool.peak": 106,
63
+ "inactive_split.small_pool.allocated": 13982572,
64
+ "inactive_split.small_pool.current": 61,
65
+ "inactive_split.small_pool.freed": 13982511,
66
+ "inactive_split.small_pool.peak": 121,
67
+ "inactive_split_bytes.all.allocated": 35135096233472,
68
+ "inactive_split_bytes.all.current": 144068608,
69
+ "inactive_split_bytes.all.freed": 35134952164864,
70
+ "inactive_split_bytes.all.peak": 518794752,
71
+ "inactive_split_bytes.large_pool.allocated": 22841886267392,
72
+ "inactive_split_bytes.large_pool.current": 109584384,
73
+ "inactive_split_bytes.large_pool.freed": 22841776683008,
74
+ "inactive_split_bytes.large_pool.peak": 499212800,
75
+ "inactive_split_bytes.small_pool.allocated": 12293209966080,
76
+ "inactive_split_bytes.small_pool.current": 34484224,
77
+ "inactive_split_bytes.small_pool.freed": 12293175481856,
78
+ "inactive_split_bytes.small_pool.peak": 73389568,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 542,
82
+ "num_device_free": 385,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 5,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 26498312942153,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 26497451427209,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 15156444588608,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 15155586180672,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 11341868353545,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 11341865246537,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 5416943616,
106
+ "reserved_bytes.all.current": 2090860544,
107
+ "reserved_bytes.all.freed": 3326083072,
108
+ "reserved_bytes.all.peak": 2090860544,
109
+ "reserved_bytes.large_pool.allocated": 4827643904,
110
+ "reserved_bytes.large_pool.current": 1971322880,
111
+ "reserved_bytes.large_pool.freed": 2856321024,
112
+ "reserved_bytes.large_pool.peak": 1971322880,
113
+ "reserved_bytes.small_pool.allocated": 589299712,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 469762048,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 542,
118
+ "segment.all.current": 157,
119
+ "segment.all.freed": 385,
120
+ "segment.all.peak": 157,
121
+ "segment.large_pool.allocated": 261,
122
+ "segment.large_pool.current": 100,
123
+ "segment.large_pool.freed": 161,
124
+ "segment.large_pool.peak": 100,
125
+ "segment.small_pool.allocated": 281,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 224,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9323394495412844}
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/eval_res_la_kron_all_homo_mc_corr_1000.json ADDED
The diff for this file is too large to render. See raw diff
 
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_5999/gpu_stats_la.json ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "memory_allocated": 866758656,
3
+ "max_memory_allocated": 1561953792,
4
+ "memory_reserved": 2090860544,
5
+ "max_memory_reserved": 2090860544,
6
+ "memory_stats": {
7
+ "active.all.allocated": 45028598,
8
+ "active.all.current": 920,
9
+ "active.all.freed": 45027678,
10
+ "active.all.peak": 1077,
11
+ "active.large_pool.allocated": 8932890,
12
+ "active.large_pool.current": 223,
13
+ "active.large_pool.freed": 8932667,
14
+ "active.large_pool.peak": 347,
15
+ "active.small_pool.allocated": 36095708,
16
+ "active.small_pool.current": 697,
17
+ "active.small_pool.freed": 36095011,
18
+ "active.small_pool.peak": 843,
19
+ "active_bytes.all.allocated": 36257221914112,
20
+ "active_bytes.all.current": 866758656,
21
+ "active_bytes.all.freed": 36256355155456,
22
+ "active_bytes.all.peak": 1561953792,
23
+ "active_bytes.large_pool.allocated": 21131872675328,
24
+ "active_bytes.large_pool.current": 863494144,
25
+ "active_bytes.large_pool.freed": 21131009181184,
26
+ "active_bytes.large_pool.peak": 1548872704,
27
+ "active_bytes.small_pool.allocated": 15125349238784,
28
+ "active_bytes.small_pool.current": 3264512,
29
+ "active_bytes.small_pool.freed": 15125345974272,
30
+ "active_bytes.small_pool.peak": 112270336,
31
+ "allocated_bytes.all.allocated": 36257221914112,
32
+ "allocated_bytes.all.current": 866758656,
33
+ "allocated_bytes.all.freed": 36256355155456,
34
+ "allocated_bytes.all.peak": 1561953792,
35
+ "allocated_bytes.large_pool.allocated": 21131872675328,
36
+ "allocated_bytes.large_pool.current": 863494144,
37
+ "allocated_bytes.large_pool.freed": 21131009181184,
38
+ "allocated_bytes.large_pool.peak": 1548872704,
39
+ "allocated_bytes.small_pool.allocated": 15125349238784,
40
+ "allocated_bytes.small_pool.current": 3264512,
41
+ "allocated_bytes.small_pool.freed": 15125345974272,
42
+ "allocated_bytes.small_pool.peak": 112270336,
43
+ "allocation.all.allocated": 45028598,
44
+ "allocation.all.current": 920,
45
+ "allocation.all.freed": 45027678,
46
+ "allocation.all.peak": 1077,
47
+ "allocation.large_pool.allocated": 8932890,
48
+ "allocation.large_pool.current": 223,
49
+ "allocation.large_pool.freed": 8932667,
50
+ "allocation.large_pool.peak": 347,
51
+ "allocation.small_pool.allocated": 36095708,
52
+ "allocation.small_pool.current": 697,
53
+ "allocation.small_pool.freed": 36095011,
54
+ "allocation.small_pool.peak": 843,
55
+ "inactive_split.all.allocated": 23344474,
56
+ "inactive_split.all.current": 110,
57
+ "inactive_split.all.freed": 23344364,
58
+ "inactive_split.all.peak": 192,
59
+ "inactive_split.large_pool.allocated": 4669930,
60
+ "inactive_split.large_pool.current": 45,
61
+ "inactive_split.large_pool.freed": 4669885,
62
+ "inactive_split.large_pool.peak": 106,
63
+ "inactive_split.small_pool.allocated": 18674544,
64
+ "inactive_split.small_pool.current": 65,
65
+ "inactive_split.small_pool.freed": 18674479,
66
+ "inactive_split.small_pool.peak": 125,
67
+ "inactive_split_bytes.all.allocated": 46739736731136,
68
+ "inactive_split_bytes.all.current": 139874304,
69
+ "inactive_split_bytes.all.freed": 46739596856832,
70
+ "inactive_split_bytes.all.peak": 518794752,
71
+ "inactive_split_bytes.large_pool.allocated": 30348180850688,
72
+ "inactive_split_bytes.large_pool.current": 109584384,
73
+ "inactive_split_bytes.large_pool.freed": 30348071266304,
74
+ "inactive_split_bytes.large_pool.peak": 499212800,
75
+ "inactive_split_bytes.small_pool.allocated": 16391555880448,
76
+ "inactive_split_bytes.small_pool.current": 30289920,
77
+ "inactive_split_bytes.small_pool.freed": 16391525590528,
78
+ "inactive_split_bytes.small_pool.peak": 73400320,
79
+ "max_split_size": -1,
80
+ "num_alloc_retries": 0,
81
+ "num_device_alloc": 709,
82
+ "num_device_free": 552,
83
+ "num_ooms": 0,
84
+ "num_sync_all_streams": 7,
85
+ "oversize_allocations.allocated": 0,
86
+ "oversize_allocations.current": 0,
87
+ "oversize_allocations.freed": 0,
88
+ "oversize_allocations.peak": 0,
89
+ "oversize_segments.allocated": 0,
90
+ "oversize_segments.current": 0,
91
+ "oversize_segments.freed": 0,
92
+ "oversize_segments.peak": 0,
93
+ "requested_bytes.all.allocated": 35331078243084,
94
+ "requested_bytes.all.current": 861514944,
95
+ "requested_bytes.all.freed": 35330216728140,
96
+ "requested_bytes.all.peak": 1543430976,
97
+ "requested_bytes.large_pool.allocated": 20208587105024,
98
+ "requested_bytes.large_pool.current": 858407936,
99
+ "requested_bytes.large_pool.freed": 20207728697088,
100
+ "requested_bytes.large_pool.peak": 1530512384,
101
+ "requested_bytes.small_pool.allocated": 15122491138060,
102
+ "requested_bytes.small_pool.current": 3107008,
103
+ "requested_bytes.small_pool.freed": 15122488031052,
104
+ "requested_bytes.small_pool.peak": 112171136,
105
+ "reserved_bytes.all.allocated": 6960447488,
106
+ "reserved_bytes.all.current": 2090860544,
107
+ "reserved_bytes.all.freed": 4869586944,
108
+ "reserved_bytes.all.peak": 2090860544,
109
+ "reserved_bytes.large_pool.allocated": 6182404096,
110
+ "reserved_bytes.large_pool.current": 1971322880,
111
+ "reserved_bytes.large_pool.freed": 4211081216,
112
+ "reserved_bytes.large_pool.peak": 1971322880,
113
+ "reserved_bytes.small_pool.allocated": 778043392,
114
+ "reserved_bytes.small_pool.current": 119537664,
115
+ "reserved_bytes.small_pool.freed": 658505728,
116
+ "reserved_bytes.small_pool.peak": 119537664,
117
+ "segment.all.allocated": 709,
118
+ "segment.all.current": 157,
119
+ "segment.all.freed": 552,
120
+ "segment.all.peak": 157,
121
+ "segment.large_pool.allocated": 338,
122
+ "segment.large_pool.current": 100,
123
+ "segment.large_pool.freed": 238,
124
+ "segment.large_pool.peak": 100,
125
+ "segment.small_pool.allocated": 371,
126
+ "segment.small_pool.current": 57,
127
+ "segment.small_pool.freed": 314,
128
+ "segment.small_pool.peak": 57
129
+ }
130
+ }
outputs/sst2/roberta-base_loratrain_val_8_16_0.1_5e-05_42_8_10000/step_7999/all_results_la_kron_all_homo_mc_corr_1000.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"eval_accuracy": 0.9323394495412844}