JohnnyZeppelin commited on
Commit
bed4c06
·
verified ·
1 Parent(s): 5cbd37d

Upload folder using huggingface_hub

Browse files
adv4_smoke/ckpt.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:04feb004329b6d52b11f1e012276e327a46c7015daba8082280be2b329f38db6
3
- size 3643443329
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c69ac9d45f8b6074925c85c322f233e2a21e7e1df03c5afc2a4230838a4b1a5f
3
+ size 3646624177
adv4_smoke/config.json CHANGED
@@ -2,16 +2,17 @@
2
  "train_jsonl": "data/groups_train.jsonl",
3
  "valid_jsonl": "data/groups_valid.jsonl",
4
  "run_dir": "runs/adv4_smoke",
 
5
  "backbone": "google/mt5-small",
6
  "num_latents": 16,
7
- "latent_dropout": 0.1,
8
- "latent_noise_std": 0.01,
9
  "batch_size": 4,
10
  "grad_accum": 8,
11
  "epochs": 1,
12
  "max_doc_len": 256,
13
  "max_sum_len": 64,
14
- "eval_every": 200,
15
  "max_train_examples": 2000,
16
  "max_valid_examples": 200,
17
  "lambda_align": 0.5,
@@ -25,6 +26,12 @@
25
  "adv_start_step": 300,
26
  "grl_alpha": 1.0,
27
  "grl_warmup": 200,
 
28
  "lr_model": 0.0001,
29
- "lr_lang": 0.001
 
 
 
 
 
30
  }
 
2
  "train_jsonl": "data/groups_train.jsonl",
3
  "valid_jsonl": "data/groups_valid.jsonl",
4
  "run_dir": "runs/adv4_smoke",
5
+ "resume": "",
6
  "backbone": "google/mt5-small",
7
  "num_latents": 16,
8
+ "latent_dropout": 0.0,
9
+ "latent_noise_std": 0.0,
10
  "batch_size": 4,
11
  "grad_accum": 8,
12
  "epochs": 1,
13
  "max_doc_len": 256,
14
  "max_sum_len": 64,
15
+ "eval_every": 400,
16
  "max_train_examples": 2000,
17
  "max_valid_examples": 200,
18
  "lambda_align": 0.5,
 
26
  "adv_start_step": 300,
27
  "grl_alpha": 1.0,
28
  "grl_warmup": 200,
29
+ "lambda_len": 1.0,
30
  "lr_model": 0.0001,
31
+ "lr_lang": 0.001,
32
+ "lr_len": 0.001,
33
+ "adv_clf_steps": 2,
34
+ "adv_clf_weight_decay": 0.0,
35
+ "adv_queue_size": 4096,
36
+ "adv_clf_batch": 256
37
  }
adv4_smoke/diag.jsonl CHANGED
@@ -1,2 +1,3 @@
1
  {"step": 200, "nll_en": 6.96965633392334, "nll_zh": 6.9684418773651124, "nll": 6.969049105644226}
2
  {"step": 400, "nll_en": 6.254790468215942, "nll_zh": 6.211220149993896, "nll": 6.23300530910492}
 
 
1
  {"step": 200, "nll_en": 6.96965633392334, "nll_zh": 6.9684418773651124, "nll": 6.969049105644226}
2
  {"step": 400, "nll_en": 6.254790468215942, "nll_zh": 6.211220149993896, "nll": 6.23300530910492}
3
+ {"step": 400, "nll_en": 6.16439887046814, "nll_zh": 6.124115419387818, "nll": 6.144257144927979, "lang_acc_valid_evalmode": 0.9325, "len_acc_valid_evalmode": 0.9075}
adv4_smoke/diag_full.json CHANGED
@@ -1,19 +1,19 @@
1
  {
2
  "ckpt": "runs/adv4_smoke/ckpt.pt",
3
- "nll_en": 5.918526216994884,
4
- "nll_zh": 5.921291329139887,
5
- "swap_delta_en": 0.6822249138330052,
6
- "swap_delta_zh": 0.707949943824752,
7
- "ablate_zero_delta_en": 8.685084957905097,
8
- "ablate_mean_delta_en": 0.13749601573823117,
9
- "ablate_noise_delta_en": 17.660213676113912,
10
- "ablate_zero_delta_zh": 8.682319834670858,
11
- "ablate_mean_delta_zh": 0.13413880894602478,
12
- "ablate_noise_delta_zh": 17.741045526419835,
13
- "inv_top1_full": 0.429175466299057,
14
- "inv_top5_full": 0.6818181872367859,
15
- "diag_sim_mean": 0.9796865582466125,
16
- "offdiag_sim_mean": 0.7884430289268494,
17
- "sim_margin": 0.19124352931976318,
18
  "n_valid": 946
19
  }
 
1
  {
2
  "ckpt": "runs/adv4_smoke/ckpt.pt",
3
+ "nll_en": 6.247315613462355,
4
+ "nll_zh": 6.215903989860422,
5
+ "swap_delta_en": 0.6225017795603068,
6
+ "swap_delta_zh": 0.6563727205449884,
7
+ "ablate_zero_delta_en": 11.165462739875906,
8
+ "ablate_mean_delta_en": 0.17689722839419988,
9
+ "ablate_noise_delta_en": 18.213750625513573,
10
+ "ablate_zero_delta_zh": 11.196874328193907,
11
+ "ablate_mean_delta_zh": 0.17574666065595115,
12
+ "ablate_noise_delta_zh": 18.303016081924923,
13
+ "inv_top1_full": 0.2780126929283142,
14
+ "inv_top5_full": 0.5359408259391785,
15
+ "diag_sim_mean": 0.9870116710662842,
16
+ "offdiag_sim_mean": 0.9049160480499268,
17
+ "sim_margin": 0.08209562301635742,
18
  "n_valid": 946
19
  }
adv4_smoke/logs.jsonl CHANGED
@@ -23,3 +23,28 @@
23
  {"step": 460, "epoch": 0, "loss": 7.359249591827393, "loss_render": 5.871917724609375, "loss_align": 0.4575628936290741, "loss_varcov": 0.06329986453056335, "loss_mean": 0.8516653776168823, "loss_mean_diff": 0.020374227315187454, "loss_pair": 7.266722241183743e-05, "loss_lang": 0.5383338332176208, "lang_acc_batch": 1.0, "elapsed_sec": 71.29257798194885}
24
  {"step": 480, "epoch": 0, "loss": 7.445590019226074, "loss_render": 5.938446998596191, "loss_align": 0.4598678946495056, "loss_varcov": 0.06467342376708984, "loss_mean": 0.8648350834846497, "loss_mean_diff": 0.02138454094529152, "loss_pair": 8.406119013670832e-05, "loss_lang": 0.541836142539978, "lang_acc_batch": 1.0, "elapsed_sec": 73.7085657119751}
25
  {"step": 500, "epoch": 0, "loss": 7.138613700866699, "loss_render": 5.660837173461914, "loss_align": 0.3657577335834503, "loss_varcov": 0.0611129030585289, "loss_mean": 0.8270612359046936, "loss_mean_diff": 0.015113018453121185, "loss_pair": 5.436208812170662e-05, "loss_lang": 0.599540114402771, "lang_acc_batch": 0.875, "elapsed_sec": 76.04010605812073}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  {"step": 460, "epoch": 0, "loss": 7.359249591827393, "loss_render": 5.871917724609375, "loss_align": 0.4575628936290741, "loss_varcov": 0.06329986453056335, "loss_mean": 0.8516653776168823, "loss_mean_diff": 0.020374227315187454, "loss_pair": 7.266722241183743e-05, "loss_lang": 0.5383338332176208, "lang_acc_batch": 1.0, "elapsed_sec": 71.29257798194885}
24
  {"step": 480, "epoch": 0, "loss": 7.445590019226074, "loss_render": 5.938446998596191, "loss_align": 0.4598678946495056, "loss_varcov": 0.06467342376708984, "loss_mean": 0.8648350834846497, "loss_mean_diff": 0.02138454094529152, "loss_pair": 8.406119013670832e-05, "loss_lang": 0.541836142539978, "lang_acc_batch": 1.0, "elapsed_sec": 73.7085657119751}
25
  {"step": 500, "epoch": 0, "loss": 7.138613700866699, "loss_render": 5.660837173461914, "loss_align": 0.3657577335834503, "loss_varcov": 0.0611129030585289, "loss_mean": 0.8270612359046936, "loss_mean_diff": 0.015113018453121185, "loss_pair": 5.436208812170662e-05, "loss_lang": 0.599540114402771, "lang_acc_batch": 0.875, "elapsed_sec": 76.04010605812073}
26
+ {"step": 20, "epoch": 0, "loss": 18.499496459960938, "loss_render": 17.055830001831055, "loss_align": 1.3303582668304443, "loss_varcov": 0.06684253364801407, "loss_mean": 0.8306604623794556, "loss_mean_diff": 0.26828959584236145, "mean_diff_l2": 0.5179668068885803, "loss_pair": 0.000838274834677577, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 160, "elapsed_sec": 3.099414348602295}
27
+ {"step": 40, "epoch": 0, "loss": 14.958220481872559, "loss_render": 13.618309020996094, "loss_align": 1.2517249584197998, "loss_varcov": 0.06138255447149277, "loss_mean": 0.7841033935546875, "loss_mean_diff": 0.21636219322681427, "mean_diff_l2": 0.4651474952697754, "loss_pair": 0.00088689656695351, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 320, "elapsed_sec": 5.880706310272217}
28
+ {"step": 60, "epoch": 0, "loss": 12.038142204284668, "loss_render": 10.583183288574219, "loss_align": 1.312792420387268, "loss_varcov": 0.06948152929544449, "loss_mean": 0.8867086172103882, "loss_mean_diff": 0.14969469606876373, "mean_diff_l2": 0.3869040012359619, "loss_pair": 0.0005324867088347673, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 480, "elapsed_sec": 8.573481321334839}
29
+ {"step": 80, "epoch": 0, "loss": 11.245194435119629, "loss_render": 10.069540977478027, "loss_align": 0.8200297355651855, "loss_varcov": 0.067088782787323, "loss_mean": 0.8860760927200317, "loss_mean_diff": 0.06086079776287079, "mean_diff_l2": 0.24669981002807617, "loss_pair": 0.0002803339157253504, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 640, "elapsed_sec": 11.428477764129639}
30
+ {"step": 100, "epoch": 0, "loss": 10.12475299835205, "loss_render": 8.671398162841797, "loss_align": 1.264050006866455, "loss_varcov": 0.07206511497497559, "loss_mean": 0.9273073673248291, "loss_mean_diff": 0.07884301990270615, "mean_diff_l2": 0.2807900011539459, "loss_pair": 0.0003208020643796772, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 800, "elapsed_sec": 14.237157344818115}
31
+ {"step": 120, "epoch": 0, "loss": 9.430133819580078, "loss_render": 7.8845977783203125, "loss_align": 1.409524917602539, "loss_varcov": 0.07383675873279572, "loss_mean": 0.9394289255142212, "loss_mean_diff": 0.08401330560445786, "mean_diff_l2": 0.28985050320625305, "loss_pair": 0.0003115071449428797, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 960, "elapsed_sec": 17.016093730926514}
32
+ {"step": 140, "epoch": 0, "loss": 9.195746421813965, "loss_render": 7.6231231689453125, "loss_align": 1.441493034362793, "loss_varcov": 0.07506725192070007, "loss_mean": 0.955194354057312, "loss_mean_diff": 0.05637464299798012, "mean_diff_l2": 0.23743344843387604, "loss_pair": 0.0002399538061581552, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1120, "elapsed_sec": 19.610440969467163}
33
+ {"step": 160, "epoch": 0, "loss": 9.158735275268555, "loss_render": 7.608129501342773, "loss_align": 1.3957197666168213, "loss_varcov": 0.07527558505535126, "loss_mean": 0.9616573452949524, "loss_mean_diff": 0.037870705127716064, "mean_diff_l2": 0.19460396468639374, "loss_pair": 0.00018440585699863732, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1280, "elapsed_sec": 22.252191066741943}
34
+ {"step": 180, "epoch": 0, "loss": 9.107904434204102, "loss_render": 7.565770149230957, "loss_align": 1.339573860168457, "loss_varcov": 0.07709620893001556, "loss_mean": 0.9732512831687927, "loss_mean_diff": 0.04033765569329262, "mean_diff_l2": 0.20084236562252045, "loss_pair": 0.00013015122385695577, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1440, "elapsed_sec": 24.971107482910156}
35
+ {"step": 200, "epoch": 0, "loss": 8.229147911071777, "loss_render": 6.720993995666504, "loss_align": 1.2900912761688232, "loss_varcov": 0.07626014947891235, "loss_mean": 0.9687088131904602, "loss_mean_diff": 0.03610805794596672, "mean_diff_l2": 0.19002120196819305, "loss_pair": 0.00012760673416778445, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1600, "elapsed_sec": 27.688571214675903}
36
+ {"step": 220, "epoch": 0, "loss": 8.656134605407715, "loss_render": 7.151536464691162, "loss_align": 1.2678537368774414, "loss_varcov": 0.07702940702438354, "loss_mean": 0.9759595990180969, "loss_mean_diff": 0.02763392962515354, "mean_diff_l2": 0.1662345677614212, "loss_pair": 8.684409112902358e-05, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1760, "elapsed_sec": 30.429605960845947}
37
+ {"step": 240, "epoch": 0, "loss": 8.83419418334961, "loss_render": 7.322035789489746, "loss_align": 1.3127011060714722, "loss_varcov": 0.07558764517307281, "loss_mean": 0.9647510051727295, "loss_mean_diff": 0.034272707998752594, "mean_diff_l2": 0.18512889742851257, "loss_pair": 0.00014667623327113688, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 1920, "elapsed_sec": 33.195481300354004}
38
+ {"step": 260, "epoch": 0, "loss": 8.939608573913574, "loss_render": 7.40751838684082, "loss_align": 1.3100942373275757, "loss_varcov": 0.07763171941041946, "loss_mean": 0.9797664880752563, "loss_mean_diff": 0.027312953025102615, "mean_diff_l2": 0.16526630520820618, "loss_pair": 8.35069949971512e-05, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 2080, "elapsed_sec": 35.94520688056946}
39
+ {"step": 280, "epoch": 0, "loss": 7.803938388824463, "loss_render": 6.312075614929199, "loss_align": 1.249525785446167, "loss_varcov": 0.07674624025821686, "loss_mean": 0.9759407043457031, "loss_mean_diff": 0.020285576581954956, "mean_diff_l2": 0.1424274444580078, "loss_pair": 7.365483907051384e-05, "loss_lang": 0.0, "loss_len": 0.0, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.0, "lang_acc_clf_detached": 0.0, "len_acc_batch": 0.0, "len_acc_clf_detached": 0.0, "adv_buf_fill": 2240, "elapsed_sec": 38.729328870773315}
40
+ {"step": 300, "epoch": 0, "loss": 10.61648941040039, "loss_render": 7.3151750564575195, "loss_align": 1.172516107559204, "loss_varcov": 0.07573163509368896, "loss_mean": 0.969778835773468, "loss_mean_diff": 0.01863565668463707, "mean_diff_l2": 0.13651247322559357, "loss_pair": 7.400706817861646e-05, "loss_lang": 0.6853976845741272, "loss_len": 1.1734845638275146, "grl_alpha_eff": 0.0, "lang_acc_batch": 0.5, "lang_acc_clf_detached": 0.640625, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.9140625, "adv_buf_fill": 2400, "elapsed_sec": 41.48701000213623}
41
+ {"step": 320, "epoch": 0, "loss": 8.606829643249512, "loss_render": 6.7812395095825195, "loss_align": 1.3065778017044067, "loss_varcov": 0.07692734152078629, "loss_mean": 0.9767350554466248, "loss_mean_diff": 0.021558810025453568, "mean_diff_l2": 0.14682918787002563, "loss_pair": 8.763473306316882e-05, "loss_lang": 0.20332029461860657, "loss_len": 0.09986048936843872, "grl_alpha_eff": 0.1, "lang_acc_batch": 1.0, "lang_acc_clf_detached": 1.0, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.91015625, "adv_buf_fill": 2560, "elapsed_sec": 44.30965733528137}
42
+ {"step": 340, "epoch": 0, "loss": 8.004097938537598, "loss_render": 6.535946369171143, "loss_align": 0.9883707165718079, "loss_varcov": 0.07314702868461609, "loss_mean": 0.9471300840377808, "loss_mean_diff": 0.025627586990594864, "mean_diff_l2": 0.16008618474006653, "loss_pair": 8.745014201849699e-05, "loss_lang": 0.05900183320045471, "loss_len": 0.08620031923055649, "grl_alpha_eff": 0.2, "lang_acc_batch": 1.0, "lang_acc_clf_detached": 0.99609375, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.90234375, "adv_buf_fill": 2720, "elapsed_sec": 47.20266532897949}
43
+ {"step": 360, "epoch": 0, "loss": 8.086645126342773, "loss_render": 6.68585205078125, "loss_align": 0.8841865062713623, "loss_varcov": 0.07223750650882721, "loss_mean": 0.9423986077308655, "loss_mean_diff": 0.021059028804302216, "mean_diff_l2": 0.14511729776859283, "loss_pair": 7.589106098748744e-05, "loss_lang": 0.025009140372276306, "loss_len": 0.11495532095432281, "grl_alpha_eff": 0.3, "lang_acc_batch": 1.0, "lang_acc_clf_detached": 1.0, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.9375, "adv_buf_fill": 2880, "elapsed_sec": 50.30023813247681}
44
+ {"step": 380, "epoch": 0, "loss": 7.851696968078613, "loss_render": 6.047324180603027, "loss_align": 0.8632481098175049, "loss_varcov": 0.07019291073083878, "loss_mean": 0.9198635816574097, "loss_mean_diff": 0.02198130264878273, "mean_diff_l2": 0.1482609212398529, "loss_pair": 9.579308971296996e-05, "loss_lang": 0.028939157724380493, "loss_len": 0.547677218914032, "grl_alpha_eff": 0.4, "lang_acc_batch": 1.0, "lang_acc_clf_detached": 1.0, "len_acc_batch": 0.875, "len_acc_clf_detached": 0.9375, "adv_buf_fill": 3040, "elapsed_sec": 53.08952498435974}
45
+ {"step": 400, "epoch": 0, "loss": 7.99092435836792, "loss_render": 6.512247085571289, "loss_align": 1.1077210903167725, "loss_varcov": 0.07487615942955017, "loss_mean": 0.9642704129219055, "loss_mean_diff": 0.01574179343879223, "mean_diff_l2": 0.12546630203723907, "loss_pair": 7.070218271110207e-05, "loss_lang": 0.02571921795606613, "loss_len": 0.052320364862680435, "grl_alpha_eff": 0.5, "lang_acc_batch": 1.0, "lang_acc_clf_detached": 1.0, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.95703125, "adv_buf_fill": 3200, "elapsed_sec": 55.9896285533905}
46
+ {"step": 420, "epoch": 0, "loss": 8.775760650634766, "loss_render": 6.578668594360352, "loss_align": 1.2785776853561401, "loss_varcov": 0.07677064090967178, "loss_mean": 0.9768249988555908, "loss_mean_diff": 0.01797603815793991, "mean_diff_l2": 0.13407474756240845, "loss_pair": 7.685914897592738e-05, "loss_lang": 0.5952599048614502, "loss_len": 0.09534120559692383, "grl_alpha_eff": 0.6, "lang_acc_batch": 0.625, "lang_acc_clf_detached": 0.99609375, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.921875, "adv_buf_fill": 3360, "elapsed_sec": 71.9451093673706}
47
+ {"step": 440, "epoch": 0, "loss": 10.075307846069336, "loss_render": 6.442470550537109, "loss_align": 0.7006415128707886, "loss_varcov": 0.06875553727149963, "loss_mean": 0.9116288423538208, "loss_mean_diff": 0.015436109155416489, "mean_diff_l2": 0.1242421418428421, "loss_pair": 6.74000330036506e-05, "loss_lang": 2.0320262908935547, "loss_len": 0.47021472454071045, "grl_alpha_eff": 0.7, "lang_acc_batch": 0.5, "lang_acc_clf_detached": 0.9921875, "len_acc_batch": 0.875, "len_acc_clf_detached": 0.9296875, "adv_buf_fill": 3520, "elapsed_sec": 74.90285658836365}
48
+ {"step": 460, "epoch": 0, "loss": 12.363208770751953, "loss_render": 6.626824378967285, "loss_align": 0.7880836725234985, "loss_varcov": 0.06934989988803864, "loss_mean": 0.9157642722129822, "loss_mean_diff": 0.014427954331040382, "mean_diff_l2": 0.12011642009019852, "loss_pair": 6.783143908251077e-05, "loss_lang": 4.243992805480957, "loss_len": 0.31181764602661133, "grl_alpha_eff": 0.8, "lang_acc_batch": 0.5, "lang_acc_clf_detached": 0.9609375, "len_acc_batch": 1.0, "len_acc_clf_detached": 0.94921875, "adv_buf_fill": 3680, "elapsed_sec": 77.82364559173584}
49
+ {"step": 480, "epoch": 0, "loss": 10.2700777053833, "loss_render": 6.74997615814209, "loss_align": 0.7055027484893799, "loss_varcov": 0.06759811192750931, "loss_mean": 0.8995985984802246, "loss_mean_diff": 0.010106584057211876, "mean_diff_l2": 0.10053151100873947, "loss_pair": 7.634978101123124e-05, "loss_lang": 1.7068363428115845, "loss_len": 0.6935465931892395, "grl_alpha_eff": 0.9, "lang_acc_batch": 0.5, "lang_acc_clf_detached": 0.94921875, "len_acc_batch": 0.75, "len_acc_clf_detached": 0.93359375, "adv_buf_fill": 3840, "elapsed_sec": 80.6856300830841}
50
+ {"step": 500, "epoch": 0, "loss": 10.736078262329102, "loss_render": 6.635019779205322, "loss_align": 0.8579686284065247, "loss_varcov": 0.07144085317850113, "loss_mean": 0.9368604421615601, "loss_mean_diff": 0.01698118820786476, "mean_diff_l2": 0.13031189143657684, "loss_pair": 7.457304309355095e-05, "loss_lang": 2.272097110748291, "loss_len": 0.5901700258255005, "grl_alpha_eff": 1.0, "lang_acc_batch": 0.5, "lang_acc_clf_detached": 0.97265625, "len_acc_batch": 0.75, "len_acc_clf_detached": 0.9140625, "adv_buf_fill": 4000, "elapsed_sec": 83.65679097175598}
adv4_smoke/probe_leakage.json CHANGED
@@ -8,13 +8,28 @@
8
  "baseline_majority_lang_acc": 0.5,
9
  "baseline_majority_len_acc": 0.9271249771118164,
10
  "lang_probe": {
11
- "train_acc": 0.9981875,
12
- "valid_acc": 0.9984143763213531,
13
- "best_valid_acc": 0.9984143763213531
14
  },
15
  "len_probe": {
16
- "train_acc": 0.954875,
17
- "valid_acc": 0.9344608879492601,
18
- "best_valid_acc": 0.9418604651162791
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  }
20
  }
 
8
  "baseline_majority_lang_acc": 0.5,
9
  "baseline_majority_len_acc": 0.9271249771118164,
10
  "lang_probe": {
11
+ "train_acc": 0.9965625,
12
+ "valid_acc": 0.9978858350951374,
13
+ "best_valid_acc": 0.9978858350951374
14
  },
15
  "len_probe": {
16
+ "train_acc": 0.940875,
17
+ "valid_acc": 0.919661733615222,
18
+ "best_valid_acc": 0.9233615221987315
19
+ },
20
+ "lang_probe_length_matched": {
21
+ "train_acc": 0.9962744090441932,
22
+ "valid_acc": 0.9972436604189636,
23
+ "best_valid_acc": 0.9977949283351709
24
+ },
25
+ "lang_bucketwise_valid_acc": {
26
+ "bucket_0_acc": 0.96875,
27
+ "bucket_1_acc": 1.0,
28
+ "bucket_2_acc": 0.9999999403953552,
29
+ "bucket_3_acc": 0.9982405304908752
30
+ },
31
+ "length_matched_sizes": {
32
+ "train": 15568,
33
+ "valid": 1814
34
  }
35
  }
planA_strong_adv/ckpt.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c98ea0dfd49d9e042624b8b6eee51c80f259af03260b2e86d6b3fff9eb0b693b
3
+ size 3643443329
planA_strong_adv/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "train_jsonl": "data/groups_train.jsonl",
3
+ "valid_jsonl": "data/groups_valid.jsonl",
4
+ "run_dir": "runs/planA_strong_adv",
5
+ "backbone": "google/mt5-small",
6
+ "num_latents": 16,
7
+ "latent_dropout": 0.1,
8
+ "latent_noise_std": 0.01,
9
+ "batch_size": 8,
10
+ "grad_accum": 8,
11
+ "epochs": 5,
12
+ "max_doc_len": 256,
13
+ "max_sum_len": 64,
14
+ "eval_every": 400,
15
+ "max_train_examples": 2000,
16
+ "max_valid_examples": 200,
17
+ "lambda_align": 1.0,
18
+ "tau": 0.07,
19
+ "lambda_varcov": 0.1,
20
+ "var_target_std": 0.05,
21
+ "lambda_mean": 1.0,
22
+ "lambda_mean_diff": 0.3,
23
+ "lambda_pair": 0.01,
24
+ "lambda_lang": 10.0,
25
+ "adv_start_step": 300,
26
+ "grl_alpha": 10.0,
27
+ "grl_warmup": 200,
28
+ "lr_model": 0.0003,
29
+ "lr_lang": 0.001
30
+ }
planA_strong_adv/diag.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ {"step": 400, "nll_en": 5.958963222503662, "nll_zh": 5.958595314025879, "nll": 5.95877926826477}
2
+ {"step": 800, "nll_en": 6.869254169464111, "nll_zh": 6.569053249359131, "nll": 6.719153709411621}
3
+ {"step": 1200, "nll_en": 6.827991523742676, "nll_zh": 6.201416721343994, "nll": 6.514704122543335}
planA_strong_adv/diag_full.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ckpt": "runs/planA_strong_adv/ckpt.pt",
3
+ "nll_en": 6.099432602493254,
4
+ "nll_zh": 6.407094520947898,
5
+ "swap_delta_en": -0.0003794827370512561,
6
+ "swap_delta_zh": 0.0017991015573384647,
7
+ "ablate_zero_delta_en": 0.013033882004009997,
8
+ "ablate_mean_delta_en": -0.0016182885361524263,
9
+ "ablate_noise_delta_en": 1.5645615253085565,
10
+ "ablate_zero_delta_zh": -0.29462803645063407,
11
+ "ablate_mean_delta_zh": 0.0013797187401930827,
12
+ "ablate_noise_delta_zh": 1.3044320705325105,
13
+ "inv_top1_full": 0.002114164875820279,
14
+ "inv_top5_full": 0.006342494860291481,
15
+ "diag_sim_mean": -0.5485604405403137,
16
+ "offdiag_sim_mean": -0.5489994287490845,
17
+ "sim_margin": 0.00043898820877075195,
18
+ "n_valid": 946
19
+ }
planA_strong_adv/logs.jsonl ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"step": 20, "epoch": 0, "loss": 17.44846534729004, "loss_render": 14.399291038513184, "loss_align": 2.22991681098938, "loss_varcov": 0.05787861347198486, "loss_mean": 0.7636987566947937, "loss_mean_diff": 0.1658695787191391, "loss_pair": 0.0009553167037665844, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 3.270895481109619}
2
+ {"step": 40, "epoch": 0, "loss": 13.25780963897705, "loss_render": 10.21828842163086, "loss_align": 2.121492862701416, "loss_varcov": 0.06745344400405884, "loss_mean": 0.8994826078414917, "loss_mean_diff": 0.039323680102825165, "loss_pair": 0.0004187910817563534, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 6.078152418136597}
3
+ {"step": 60, "epoch": 0, "loss": 11.189596176147461, "loss_render": 8.127042770385742, "loss_align": 2.0865025520324707, "loss_varcov": 0.07468238472938538, "loss_mean": 0.9617210626602173, "loss_mean_diff": 0.022864650934934616, "loss_pair": 0.00017129944171756506, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 8.666957139968872}
4
+ {"step": 80, "epoch": 0, "loss": 11.170843124389648, "loss_render": 8.103418350219727, "loss_align": 2.0971527099609375, "loss_varcov": 0.07409833371639252, "loss_mean": 0.9573116302490234, "loss_mean_diff": 0.01849462278187275, "loss_pair": 0.00017248486983589828, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 11.195322036743164}
5
+ {"step": 100, "epoch": 0, "loss": 10.222838401794434, "loss_render": 7.2176103591918945, "loss_align": 2.0178065299987793, "loss_varcov": 0.07645134627819061, "loss_mean": 0.9764639139175415, "loss_mean_diff": 0.011036057956516743, "loss_pair": 8.53120582178235e-05, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 13.72252869606018}
6
+ {"step": 120, "epoch": 0, "loss": 10.622635841369629, "loss_render": 7.6440229415893555, "loss_align": 1.995164394378662, "loss_varcov": 0.07581277191638947, "loss_mean": 0.9719959497451782, "loss_mean_diff": 0.012899456545710564, "loss_pair": 9.762348781805485e-05, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 16.413082122802734}
7
+ {"step": 140, "epoch": 0, "loss": 9.812697410583496, "loss_render": 7.074771881103516, "loss_align": 1.7721569538116455, "loss_varcov": 0.0734865665435791, "loss_mean": 0.9556867480278015, "loss_mean_diff": 0.009107053279876709, "loss_pair": 8.279015310108662e-05, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 18.97730851173401}
8
+ {"step": 160, "epoch": 0, "loss": 9.427239418029785, "loss_render": 7.0621490478515625, "loss_align": 1.4540939331054688, "loss_varcov": 0.06762664765119553, "loss_mean": 0.9018818736076355, "loss_mean_diff": 0.007835108786821365, "loss_pair": 9.306806896347553e-05, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 21.571370363235474}
9
+ {"step": 180, "epoch": 0, "loss": 9.209436416625977, "loss_render": 7.316653251647949, "loss_align": 1.0914573669433594, "loss_varcov": 0.058384113013744354, "loss_mean": 0.7922800779342651, "loss_mean_diff": 0.010687567293643951, "loss_pair": 0.00010606684372760355, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 24.230711936950684}
10
+ {"step": 200, "epoch": 0, "loss": 9.025256156921387, "loss_render": 7.029796600341797, "loss_align": 1.2132941484451294, "loss_varcov": 0.05777993053197861, "loss_mean": 0.7738837003707886, "loss_mean_diff": 0.008341845124959946, "loss_pair": 0.00011281467595836148, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 27.003639698028564}
11
+ {"step": 220, "epoch": 0, "loss": 8.830577850341797, "loss_render": 6.955078601837158, "loss_align": 1.0562798976898193, "loss_varcov": 0.05901762843132019, "loss_mean": 0.8042008876800537, "loss_mean_diff": 0.030382797122001648, "loss_pair": 0.00017591517826076597, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 29.733988523483276}
12
+ {"step": 240, "epoch": 0, "loss": 8.979713439941406, "loss_render": 7.460409641265869, "loss_align": 0.8178008794784546, "loss_varcov": 0.0508476197719574, "loss_mean": 0.6927801370620728, "loss_mean_diff": 0.012120133265852928, "loss_pair": 0.00011183207243448123, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 32.56232237815857}
13
+ {"step": 260, "epoch": 1, "loss": 8.385820388793945, "loss_render": 6.782420635223389, "loss_align": 0.9300903081893921, "loss_varcov": 0.05035228282213211, "loss_mean": 0.6640350818634033, "loss_mean_diff": 0.014128236100077629, "loss_pair": 0.00011439307854743674, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 35.25255846977234}
14
+ {"step": 280, "epoch": 1, "loss": 8.195206642150879, "loss_render": 6.989620685577393, "loss_align": 0.5698128938674927, "loss_varcov": 0.04689349979162216, "loss_mean": 0.6278437376022339, "loss_mean_diff": 0.010798541828989983, "loss_pair": 0.00013950970605947077, "loss_lang": 0.0, "lang_acc_batch": 0.0, "elapsed_sec": 38.01663851737976}
15
+ {"step": 300, "epoch": 1, "loss": 15.331546783447266, "loss_render": 6.95874547958374, "loss_align": 0.8953583240509033, "loss_varcov": 0.04245471581816673, "loss_mean": 0.5355871915817261, "loss_mean_diff": 0.016121020540595055, "loss_pair": 0.00014923824346624315, "loss_lang": 0.6932772397994995, "lang_acc_batch": 0.5, "elapsed_sec": 40.70964956283569}
16
+ {"step": 320, "epoch": 1, "loss": 15.220434188842773, "loss_render": 7.244317531585693, "loss_align": 0.7316961288452148, "loss_varcov": 0.034756217151880264, "loss_mean": 0.3321050703525543, "loss_mean_diff": 0.011537710204720497, "loss_pair": 0.00017562074935995042, "loss_lang": 0.6905377507209778, "lang_acc_batch": 0.5625, "elapsed_sec": 43.42813730239868}
17
+ {"step": 340, "epoch": 1, "loss": 14.676284790039062, "loss_render": 6.863646030426025, "loss_align": 0.5873464345932007, "loss_varcov": 0.03527826815843582, "loss_mean": 0.3537050485610962, "loss_mean_diff": 0.012918207794427872, "loss_pair": 0.00020811831927858293, "loss_lang": 0.6864182353019714, "lang_acc_batch": 0.5625, "elapsed_sec": 46.17159342765808}
18
+ {"step": 360, "epoch": 1, "loss": 14.379232406616211, "loss_render": 6.6727495193481445, "loss_align": 0.5381826758384705, "loss_varcov": 0.0344526544213295, "loss_mean": 0.32630687952041626, "loss_mean_diff": 0.025800246745347977, "loss_pair": 0.00019314922974444926, "loss_lang": 0.6830806732177734, "lang_acc_batch": 0.5, "elapsed_sec": 48.94628882408142}
19
+ {"step": 380, "epoch": 1, "loss": 14.412225723266602, "loss_render": 6.505192756652832, "loss_align": 0.539114236831665, "loss_varcov": 0.03541135787963867, "loss_mean": 0.3842417001724243, "loss_mean_diff": 0.03406089171767235, "loss_pair": 0.000310179078951478, "loss_lang": 0.6969915628433228, "lang_acc_batch": 0.5, "elapsed_sec": 51.65414261817932}
20
+ {"step": 400, "epoch": 1, "loss": 15.346776962280273, "loss_render": 6.667016983032227, "loss_align": 1.2550511360168457, "loss_varcov": 0.03567752242088318, "loss_mean": 0.3322405517101288, "loss_mean_diff": 0.06253795325756073, "loss_pair": 0.00036051555071026087, "loss_lang": 0.7070134878158569, "lang_acc_batch": 0.5, "elapsed_sec": 54.406588077545166}
21
+ {"step": 420, "epoch": 1, "loss": 15.20002269744873, "loss_render": 6.1708478927612305, "loss_align": 1.1804537773132324, "loss_varcov": 0.03829965740442276, "loss_mean": 0.3671625852584839, "loss_mean_diff": 0.2168521136045456, "loss_pair": 0.0005981298163533211, "loss_lang": 0.7412666082382202, "lang_acc_batch": 0.3125, "elapsed_sec": 62.89730644226074}
22
+ {"step": 440, "epoch": 1, "loss": 16.700496673583984, "loss_render": 6.395833969116211, "loss_align": 1.9209853410720825, "loss_varcov": 0.04186585918068886, "loss_mean": 0.3952062427997589, "loss_mean_diff": 0.37115272879600525, "loss_pair": 0.0009804917499423027, "loss_lang": 0.7872928380966187, "lang_acc_batch": 0.125, "elapsed_sec": 65.4900975227356}
23
+ {"step": 460, "epoch": 1, "loss": 18.013893127441406, "loss_render": 6.220047950744629, "loss_align": 2.461850166320801, "loss_varcov": 0.050187017768621445, "loss_mean": 0.46075618267059326, "loss_mean_diff": 0.828162670135498, "loss_pair": 0.0019917606841772795, "loss_lang": 0.861775279045105, "lang_acc_batch": 0.0, "elapsed_sec": 67.98564767837524}
24
+ {"step": 480, "epoch": 1, "loss": 19.312786102294922, "loss_render": 6.282978057861328, "loss_align": 3.520476818084717, "loss_varcov": 0.050538212060928345, "loss_mean": 0.3547961115837097, "loss_mean_diff": 1.3036836385726929, "loss_pair": 0.002971484325826168, "loss_lang": 0.8758347034454346, "lang_acc_batch": 0.0, "elapsed_sec": 70.5158588886261}
25
+ {"step": 500, "epoch": 1, "loss": 19.559425354003906, "loss_render": 6.605067253112793, "loss_align": 2.4705088138580322, "loss_varcov": 0.0698947161436081, "loss_mean": 0.16215939819812775, "loss_mean_diff": 3.049884796142578, "loss_pair": 0.006037846673280001, "loss_lang": 0.9399674534797668, "lang_acc_batch": 0.0, "elapsed_sec": 73.05153346061707}
26
+ {"step": 520, "epoch": 2, "loss": 21.183839797973633, "loss_render": 8.934226989746094, "loss_align": 1.9848321676254272, "loss_varcov": 0.0732588842511177, "loss_mean": 0.08993569016456604, "loss_mean_diff": 3.469738245010376, "loss_pair": 0.006872326135635376, "loss_lang": 0.9126529693603516, "lang_acc_batch": 0.0, "elapsed_sec": 75.83651423454285}
27
+ {"step": 540, "epoch": 2, "loss": 19.88075828552246, "loss_render": 8.3740816116333, "loss_align": 2.0490126609802246, "loss_varcov": 0.0764298290014267, "loss_mean": 0.0639270693063736, "loss_mean_diff": 3.6614136695861816, "loss_pair": 0.00721985986456275, "loss_lang": 0.8287598490715027, "lang_acc_batch": 0.0, "elapsed_sec": 78.56681561470032}
28
+ {"step": 560, "epoch": 2, "loss": 19.40838623046875, "loss_render": 8.462965965270996, "loss_align": 2.066204309463501, "loss_varcov": 0.07731221616268158, "loss_mean": 0.0660894438624382, "loss_mean_diff": 3.6739416122436523, "loss_pair": 0.007231426425278187, "loss_lang": 0.7703139781951904, "lang_acc_batch": 0.0, "elapsed_sec": 81.3403639793396}
29
+ {"step": 580, "epoch": 2, "loss": 18.755638122558594, "loss_render": 7.910405158996582, "loss_align": 2.080288887023926, "loss_varcov": 0.07747199386358261, "loss_mean": 0.09517866373062134, "loss_mean_diff": 3.561305046081543, "loss_pair": 0.00701095862314105, "loss_lang": 0.7593555450439453, "lang_acc_batch": 0.0, "elapsed_sec": 84.14190888404846}
30
+ {"step": 600, "epoch": 2, "loss": 18.966428756713867, "loss_render": 7.574024200439453, "loss_align": 2.0861430168151855, "loss_varcov": 0.0774904415011406, "loss_mean": 0.12419987469911575, "loss_mean_diff": 3.4454684257507324, "loss_pair": 0.006784289609640837, "loss_lang": 0.814060389995575, "lang_acc_batch": 0.0, "elapsed_sec": 86.8590316772461}
31
+ {"step": 620, "epoch": 2, "loss": 20.170150756835938, "loss_render": 7.713193893432617, "loss_align": 2.178412437438965, "loss_varcov": 0.07718919217586517, "loss_mean": 0.19056037068367004, "loss_mean_diff": 3.172530174255371, "loss_pair": 0.006258513778448105, "loss_lang": 0.9128443002700806, "lang_acc_batch": 0.0, "elapsed_sec": 89.56402516365051}
32
+ {"step": 640, "epoch": 2, "loss": 20.611522674560547, "loss_render": 7.707730770111084, "loss_align": 2.3713161945343018, "loss_varcov": 0.07692572474479675, "loss_mean": 0.25030192732810974, "loss_mean_diff": 2.9275705814361572, "loss_pair": 0.005783220753073692, "loss_lang": 0.9396150708198547, "lang_acc_batch": 0.0, "elapsed_sec": 92.36527895927429}
33
+ {"step": 660, "epoch": 2, "loss": 20.270437240600586, "loss_render": 7.323617458343506, "loss_align": 2.0843148231506348, "loss_varcov": 0.07724379003047943, "loss_mean": 0.11488687992095947, "loss_mean_diff": 3.4760355949401855, "loss_pair": 0.006850875448435545, "loss_lang": 0.9697014093399048, "lang_acc_batch": 0.0, "elapsed_sec": 95.0110559463501}
34
+ {"step": 680, "epoch": 2, "loss": 20.041240692138672, "loss_render": 7.326175689697266, "loss_align": 2.083498239517212, "loss_varcov": 0.0773894190788269, "loss_mean": 0.11391159147024155, "loss_mean_diff": 3.4835829734802246, "loss_pair": 0.0068626608699560165, "loss_lang": 0.9464774131774902, "lang_acc_batch": 0.0, "elapsed_sec": 97.71768045425415}
35
+ {"step": 700, "epoch": 2, "loss": 19.248600006103516, "loss_render": 7.304903030395508, "loss_align": 2.0885274410247803, "loss_varcov": 0.0774984136223793, "loss_mean": 0.1112559512257576, "loss_mean_diff": 3.497422695159912, "loss_pair": 0.006886047776788473, "loss_lang": 0.8686869144439697, "lang_acc_batch": 0.0, "elapsed_sec": 100.29313850402832}
36
+ {"step": 720, "epoch": 2, "loss": 19.00286102294922, "loss_render": 7.609348773956299, "loss_align": 2.0779409408569336, "loss_varcov": 0.07745957374572754, "loss_mean": 0.07925551384687424, "loss_mean_diff": 3.624250888824463, "loss_pair": 0.00713429506868124, "loss_lang": 0.814122200012207, "lang_acc_batch": 0.0, "elapsed_sec": 102.86475229263306}
37
+ {"step": 740, "epoch": 2, "loss": 18.329605102539062, "loss_render": 7.74180269241333, "loss_align": 2.0847558975219727, "loss_varcov": 0.07752522081136703, "loss_mean": 0.0647626742720604, "loss_mean_diff": 3.6839816570281982, "loss_pair": 0.007252028211951256, "loss_lang": 0.7325264811515808, "lang_acc_batch": 0.0, "elapsed_sec": 105.34986162185669}
38
+ {"step": 760, "epoch": 3, "loss": 18.20210838317871, "loss_render": 7.8491082191467285, "loss_align": 2.08491587638855, "loss_varcov": 0.07749895751476288, "loss_mean": 0.06858977675437927, "loss_mean_diff": 3.6676442623138428, "loss_pair": 0.00722117442637682, "loss_lang": 0.7091378569602966, "lang_acc_batch": 0.5, "elapsed_sec": 107.88888740539551}
39
+ {"step": 780, "epoch": 3, "loss": 19.281408309936523, "loss_render": 8.65558910369873, "loss_align": 2.207584857940674, "loss_varcov": 0.07718987762928009, "loss_mean": 0.227508544921875, "loss_mean_diff": 3.0246682167053223, "loss_pair": 0.005971693433821201, "loss_lang": 0.7275546193122864, "lang_acc_batch": 0.5, "elapsed_sec": 110.3675754070282}
40
+ {"step": 800, "epoch": 3, "loss": 18.68621253967285, "loss_render": 7.690478801727295, "loss_align": 2.4369091987609863, "loss_varcov": 0.07673110067844391, "loss_mean": 0.46585336327552795, "loss_mean_diff": 2.06011700630188, "loss_pair": 0.004098784644156694, "loss_lang": 0.7467221021652222, "lang_acc_batch": 0.5, "elapsed_sec": 112.93250679969788}
41
+ {"step": 820, "epoch": 3, "loss": 17.327693939208984, "loss_render": 6.304405212402344, "loss_align": 2.1628308296203613, "loss_varcov": 0.07973498106002808, "loss_mean": 0.8225090503692627, "loss_mean_diff": 0.7044296264648438, "loss_pair": 0.001380493980832398, "loss_lang": 0.7818634510040283, "lang_acc_batch": 0.5, "elapsed_sec": 125.70236349105835}
42
+ {"step": 840, "epoch": 3, "loss": 17.89468002319336, "loss_render": 6.167552947998047, "loss_align": 2.230198860168457, "loss_varcov": 0.07972009479999542, "loss_mean": 0.6400442123413086, "loss_mean_diff": 1.43398118019104, "loss_pair": 0.0028066961094737053, "loss_lang": 0.8418691158294678, "lang_acc_batch": 0.5, "elapsed_sec": 128.27270078659058}
43
+ {"step": 860, "epoch": 3, "loss": 18.881128311157227, "loss_render": 6.325827598571777, "loss_align": 2.0143158435821533, "loss_varcov": 0.07504500448703766, "loss_mean": 0.08536524325609207, "loss_mean_diff": 3.5183627605438232, "loss_pair": 0.006983502767980099, "loss_lang": 0.9392537474632263, "lang_acc_batch": 0.0, "elapsed_sec": 130.80610394477844}
44
+ {"step": 880, "epoch": 3, "loss": 18.85932159423828, "loss_render": 6.366743087768555, "loss_align": 2.102109909057617, "loss_varcov": 0.07923167943954468, "loss_mean": 0.1053098663687706, "loss_mean_diff": 3.5624489784240723, "loss_pair": 0.006971156224608421, "loss_lang": 0.9208430647850037, "lang_acc_batch": 0.0, "elapsed_sec": 133.34682822227478}
45
+ {"step": 900, "epoch": 3, "loss": 17.620235443115234, "loss_render": 6.022817611694336, "loss_align": 2.1106433868408203, "loss_varcov": 0.07922548055648804, "loss_mean": 0.16562314331531525, "loss_mean_diff": 3.321043014526367, "loss_pair": 0.006491813808679581, "loss_lang": 0.8316853046417236, "lang_acc_batch": 0.0, "elapsed_sec": 135.93458199501038}
46
+ {"step": 920, "epoch": 3, "loss": 17.28537368774414, "loss_render": 6.245058059692383, "loss_align": 2.141601085662842, "loss_varcov": 0.07957321405410767, "loss_mean": 0.21697665750980377, "loss_mean_diff": 3.123063564300537, "loss_pair": 0.006106463260948658, "loss_lang": 0.7736799120903015, "lang_acc_batch": 0.0, "elapsed_sec": 138.67508697509766}
47
+ {"step": 940, "epoch": 3, "loss": 16.66741180419922, "loss_render": 6.223911285400391, "loss_align": 2.083798885345459, "loss_varcov": 0.07997430860996246, "loss_mean": 0.16306647658348083, "loss_mean_diff": 3.347207546234131, "loss_pair": 0.006537998095154762, "loss_lang": 0.71844083070755, "lang_acc_batch": 0.0, "elapsed_sec": 141.16849660873413}
48
+ {"step": 960, "epoch": 3, "loss": 16.86397361755371, "loss_render": 6.468313217163086, "loss_align": 2.096010684967041, "loss_varcov": 0.07985350489616394, "loss_mean": 0.13943332433700562, "loss_mean_diff": 3.4392380714416504, "loss_pair": 0.006719791330397129, "loss_lang": 0.7120393514633179, "lang_acc_batch": 0.0, "elapsed_sec": 143.72027802467346}
49
+ {"step": 980, "epoch": 3, "loss": 17.294147491455078, "loss_render": 6.549448490142822, "loss_align": 2.093204975128174, "loss_varcov": 0.07970863580703735, "loss_mean": 0.14912377297878265, "loss_mean_diff": 3.397456169128418, "loss_pair": 0.006640184670686722, "loss_lang": 0.7475094795227051, "lang_acc_batch": 0.0, "elapsed_sec": 146.1641583442688}
50
+ {"step": 1000, "epoch": 3, "loss": 18.367496490478516, "loss_render": 6.456618309020996, "loss_align": 3.3811845779418945, "loss_varcov": 0.07767140865325928, "loss_mean": 0.6067335605621338, "loss_mean_diff": 1.5166220664978027, "loss_pair": 0.003017050214111805, "loss_lang": 0.7460176944732666, "lang_acc_batch": 0.0, "elapsed_sec": 148.64404010772705}
51
+ {"step": 1020, "epoch": 4, "loss": 17.688064575195312, "loss_render": 6.950000762939453, "loss_align": 2.1016249656677246, "loss_varcov": 0.07751985639333725, "loss_mean": 0.19278621673583984, "loss_mean_diff": 3.171698808670044, "loss_pair": 0.006250890903174877, "loss_lang": 0.7484326958656311, "lang_acc_batch": 0.0, "elapsed_sec": 151.28866815567017}
52
+ {"step": 1040, "epoch": 4, "loss": 33.370033264160156, "loss_render": 22.359899520874023, "loss_align": 2.192068099975586, "loss_varcov": 0.07601580023765564, "loss_mean": 0.16283297538757324, "loss_mean_diff": 3.252607583999634, "loss_pair": 0.006452314089983702, "loss_lang": 0.7671784162521362, "lang_acc_batch": 0.0, "elapsed_sec": 154.0615839958191}
53
+ {"step": 1060, "epoch": 4, "loss": 28.231937408447266, "loss_render": 17.259353637695312, "loss_align": 2.349491596221924, "loss_varcov": 0.07718925178050995, "loss_mean": 0.3738647997379303, "loss_mean_diff": 2.4396185874938965, "loss_pair": 0.004828662611544132, "loss_lang": 0.7509574890136719, "lang_acc_batch": 0.5, "elapsed_sec": 156.65452361106873}
54
+ {"step": 1080, "epoch": 4, "loss": 18.063575744628906, "loss_render": 7.5890960693359375, "loss_align": 2.1615092754364014, "loss_varcov": 0.07754042744636536, "loss_mean": 0.5407552719116211, "loss_mean_diff": 1.779772162437439, "loss_pair": 0.0035317479632794857, "loss_lang": 0.7230494022369385, "lang_acc_batch": 0.5, "elapsed_sec": 159.3236813545227}
55
+ {"step": 1100, "epoch": 4, "loss": 17.079742431640625, "loss_render": 6.792820453643799, "loss_align": 2.0916519165039062, "loss_varcov": 0.07748641073703766, "loss_mean": 0.9437788128852844, "loss_mean_diff": 0.16691333055496216, "loss_pair": 0.00038116308860480785, "loss_lang": 0.7193666696548462, "lang_acc_batch": 0.5, "elapsed_sec": 161.9307143688202}
56
+ {"step": 1120, "epoch": 4, "loss": 18.301483154296875, "loss_render": 6.876789093017578, "loss_align": 3.0764267444610596, "loss_varcov": 0.07474061101675034, "loss_mean": 0.8511231541633606, "loss_mean_diff": 0.46793580055236816, "loss_pair": 0.0010896055027842522, "loss_lang": 0.7349280118942261, "lang_acc_batch": 0.5, "elapsed_sec": 164.61906719207764}
57
+ {"step": 1140, "epoch": 4, "loss": 18.913043975830078, "loss_render": 6.927214622497559, "loss_align": 2.256944417953491, "loss_varcov": 0.07733011245727539, "loss_mean": 0.3451493978500366, "loss_mean_diff": 2.558126926422119, "loss_pair": 0.005057120695710182, "loss_lang": 0.8608512878417969, "lang_acc_batch": 0.0, "elapsed_sec": 167.27073168754578}
58
+ {"step": 1160, "epoch": 4, "loss": 18.83636474609375, "loss_render": 7.057194232940674, "loss_align": 2.1093475818634033, "loss_varcov": 0.07755909860134125, "loss_mean": 0.2649422585964203, "loss_mean_diff": 2.8842649459838867, "loss_pair": 0.005686442367732525, "loss_lang": 0.853178858757019, "lang_acc_batch": 0.0, "elapsed_sec": 169.93603777885437}
59
+ {"step": 1180, "epoch": 4, "loss": 19.55497932434082, "loss_render": 7.639110565185547, "loss_align": 2.8046107292175293, "loss_varcov": 0.07643450796604156, "loss_mean": 0.44670742750167847, "loss_mean_diff": 2.1288349628448486, "loss_pair": 0.004239874891936779, "loss_lang": 0.8018212914466858, "lang_acc_batch": 0.0, "elapsed_sec": 172.66157722473145}
60
+ {"step": 1200, "epoch": 4, "loss": 18.104799270629883, "loss_render": 7.345398426055908, "loss_align": 2.0975468158721924, "loss_varcov": 0.07772111892700195, "loss_mean": 0.3868744373321533, "loss_mean_diff": 2.4003124237060547, "loss_pair": 0.004739599302411079, "loss_lang": 0.7547065615653992, "lang_acc_batch": 0.0, "elapsed_sec": 175.41060853004456}
61
+ {"step": 1220, "epoch": 4, "loss": 18.109895706176758, "loss_render": 6.76279354095459, "loss_align": 2.9489850997924805, "loss_varcov": 0.07811184227466583, "loss_mean": 0.7636910676956177, "loss_mean_diff": 0.9010977149009705, "loss_pair": 0.001793498289771378, "loss_lang": 0.7356266975402832, "lang_acc_batch": 0.0, "elapsed_sec": 185.65714263916016}
62
+ {"step": 1240, "epoch": 4, "loss": 17.515338897705078, "loss_render": 6.696865558624268, "loss_align": 2.1059250831604004, "loss_varcov": 0.07993534207344055, "loss_mean": 0.5966293811798096, "loss_mean_diff": 1.6121524572372437, "loss_pair": 0.003150027245283127, "loss_lang": 0.7624248266220093, "lang_acc_batch": 0.0, "elapsed_sec": 188.2951831817627}
planA_strong_adv/probe_leakage.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ckpt": "runs/planA_strong_adv/ckpt.pt",
3
+ "max_train_groups": 8000,
4
+ "max_valid_groups": 946,
5
+ "probe_hidden": 0,
6
+ "epochs": 20,
7
+ "lr": 0.01,
8
+ "baseline_majority_lang_acc": 0.5,
9
+ "baseline_majority_len_acc": 0.9271249771118164,
10
+ "lang_probe": {
11
+ "train_acc": 0.998875,
12
+ "valid_acc": 0.9989429175475687,
13
+ "best_valid_acc": 0.9989429175475687
14
+ },
15
+ "len_probe": {
16
+ "train_acc": 0.9271875,
17
+ "valid_acc": 0.9006342494714588,
18
+ "best_valid_acc": 0.9006342494714588
19
+ }
20
+ }
planB_ramp_adv/.ipynb_checkpoints/diag_full-checkpoint.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ckpt": "runs/planB_ramp_adv/ckpt.pt",
3
+ "nll_en": 4.322499513122042,
4
+ "nll_zh": 4.326212269055163,
5
+ "swap_delta_en": 0.9772960862448049,
6
+ "swap_delta_zh": 0.9534117666409837,
7
+ "ablate_zero_delta_en": 0.3369567580757383,
8
+ "ablate_mean_delta_en": 0.3581282724537759,
9
+ "ablate_noise_delta_en": 5.115654102591581,
10
+ "ablate_zero_delta_zh": 0.33324400214261785,
11
+ "ablate_mean_delta_zh": 0.36060957475142047,
12
+ "ablate_noise_delta_zh": 5.1406748622466845,
13
+ "inv_top1_full": 0.46617335081100464,
14
+ "inv_top5_full": 0.7695560455322266,
15
+ "diag_sim_mean": 0.9375637173652649,
16
+ "offdiag_sim_mean": 0.06004802882671356,
17
+ "sim_margin": 0.8775156885385513,
18
+ "n_valid": 946
19
+ }
planB_ramp_adv/ckpt.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb6a14841a15f2a685f582b4da3901d651704ded6728c0b976b78a96280e5ae3
3
+ size 3644506605
planB_ramp_adv/config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "train_jsonl": "data/groups_train.jsonl",
3
+ "valid_jsonl": "data/groups_valid.jsonl",
4
+ "run_dir": "runs/planB_ramp_adv",
5
+ "resume": "",
6
+ "backbone": "google/mt5-small",
7
+ "num_latents": 16,
8
+ "latent_dropout": 0.0,
9
+ "latent_noise_std": 0.0,
10
+ "batch_size": 8,
11
+ "grad_accum": 8,
12
+ "epochs": 1,
13
+ "max_doc_len": 256,
14
+ "max_sum_len": 64,
15
+ "eval_every": 400,
16
+ "max_train_examples": 2000,
17
+ "max_valid_examples": 200,
18
+ "lambda_align": 1.0,
19
+ "tau": 0.07,
20
+ "lambda_varcov": 0.1,
21
+ "var_target_std": 0.05,
22
+ "lambda_mean": 0.0,
23
+ "lambda_mean_diff": 0.0,
24
+ "lambda_pair": 0.0,
25
+ "lambda_lang": 1.0,
26
+ "lambda_len": 0.0,
27
+ "adv_start_step": 0,
28
+ "grl_alpha": 0.0,
29
+ "grl_warmup": 200,
30
+ "lr_model": 0.0003,
31
+ "lr_lang": 0.005,
32
+ "lr_len": 0.001,
33
+ "adv_clf_steps": 10,
34
+ "adv_clf_weight_decay": 0.0,
35
+ "adv_queue_size": 4096,
36
+ "adv_clf_batch": 1024,
37
+ "adv_mix_current": 0.2
38
+ }
planB_ramp_adv/config_resume.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "train_jsonl": "data/groups_train.jsonl",
3
+ "valid_jsonl": "data/groups_valid.jsonl",
4
+ "run_dir": "runs/planB_ramp_adv",
5
+ "resume": "runs/planB_ramp_adv/ckpt.pt",
6
+ "backbone": "google/mt5-small",
7
+ "num_latents": 16,
8
+ "latent_dropout": 0.0,
9
+ "latent_noise_std": 0.0,
10
+ "batch_size": 8,
11
+ "grad_accum": 8,
12
+ "epochs": 5,
13
+ "max_doc_len": 256,
14
+ "max_sum_len": 64,
15
+ "eval_every": 400,
16
+ "max_train_examples": 2000,
17
+ "max_valid_examples": 200,
18
+ "lambda_align": 1.0,
19
+ "tau": 0.07,
20
+ "lambda_varcov": 0.1,
21
+ "var_target_std": 0.05,
22
+ "lambda_mean": 0.5,
23
+ "lambda_mean_diff": 0.1,
24
+ "lambda_pair": 0.005,
25
+ "lambda_lang": 2.0,
26
+ "lambda_len": 0.0,
27
+ "adv_start_step": 0,
28
+ "grl_alpha": 2.0,
29
+ "grl_warmup": 200,
30
+ "lr_model": 0.0003,
31
+ "lr_lang": 0.005,
32
+ "lr_len": 0.001,
33
+ "adv_clf_steps": 20,
34
+ "adv_clf_weight_decay": 0.0,
35
+ "adv_queue_size": 4096,
36
+ "adv_clf_batch": 1024,
37
+ "adv_mix_current": 0.2
38
+ }
planB_ramp_adv/diag.jsonl ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {"step": 400, "nll_en": 6.068062076568603, "nll_zh": 6.0595797920227055, "nll": 6.063820934295654}
2
+ {"step": 400, "nll_en": 6.0633987617492675, "nll_zh": 6.057678394317627, "nll": 6.060538578033447}
3
+ {"step": 800, "nll_en": 5.106516609191894, "nll_zh": 5.131229724884033, "nll": 5.118873167037964}
4
+ {"step": 1200, "nll_en": 4.584381484985352, "nll_zh": 4.55584114074707, "nll": 4.570111312866211}
5
+ {"step": 400, "nll_en": 5.860627880096436, "nll_zh": 5.870424633026123, "nll": 5.865526256561279}
6
+ {"step": 800, "nll_en": 4.980921134948731, "nll_zh": 4.974330730438233, "nll": 4.977625932693481}
7
+ {"step": 1200, "nll_en": 4.7775664901733395, "nll_zh": 4.75729082107544, "nll": 4.767428655624389}
8
+ {"step": 1600, "nll_en": 4.471948976516724, "nll_zh": 4.4651030158996585, "nll": 4.468525996208191}
9
+ {"step": 2000, "nll_en": 4.27439206123352, "nll_zh": 4.281768321990967, "nll": 4.2780801916122435}
10
+ {"step": 400, "nll_en": 6.256834545135498, "nll_zh": 6.361428203582764, "nll": 6.309131374359131}
11
+ {"step": 800, "nll_en": 5.419311351776123, "nll_zh": 5.404887809753418, "nll": 5.412099580764771}
12
+ {"step": 1200, "nll_en": 5.113589344024658, "nll_zh": 5.0846944999694825, "nll": 5.09914192199707}
13
+ {"step": 1600, "nll_en": 4.844666538238525, "nll_zh": 4.834702758789063, "nll": 4.839684648513794}
14
+ {"step": 2000, "nll_en": 4.669110088348389, "nll_zh": 4.691750354766846, "nll": 4.680430221557617}
15
+ {"step": 400, "nll_en": 5.6863799667358395, "nll_zh": 5.810493335723877, "nll": 5.748436651229858, "lang_acc_valid_evalmode": 0.83}
16
+ {"step": 800, "nll_en": 4.935823097229004, "nll_zh": 4.916043968200683, "nll": 4.9259335327148435, "lang_acc_valid_evalmode": 0.49}
17
+ {"step": 1200, "nll_en": 4.722777328491211, "nll_zh": 4.695003204345703, "nll": 4.708890266418457, "lang_acc_valid_evalmode": 0.525}
18
+ {"step": 1600, "nll_en": 4.497901811599731, "nll_zh": 4.503025913238526, "nll": 4.500463862419128, "lang_acc_valid_evalmode": 0.5025}
19
+ {"step": 2000, "nll_en": 4.213522682189941, "nll_zh": 4.227123193740844, "nll": 4.220322937965393, "lang_acc_valid_evalmode": 0.4475}
20
+ {"step": 400, "nll_en": 5.870532836914062, "nll_zh": 5.885317974090576, "nll": 5.877925405502319, "lang_acc_valid_evalmode": 0.83}
21
+ {"step": 800, "nll_en": 4.898431243896485, "nll_zh": 4.9147431182861325, "nll": 4.906587181091308, "lang_acc_valid_evalmode": 0.5125}
22
+ {"step": 1200, "nll_en": 4.444986953735351, "nll_zh": 4.4478084754943845, "nll": 4.446397714614868, "lang_acc_valid_evalmode": 0.5775}
23
+ {"step": 1600, "nll_en": 4.133444452285767, "nll_zh": 4.132693481445313, "nll": 4.133068966865539, "lang_acc_valid_evalmode": 0.495}
24
+ {"step": 2000, "nll_en": 3.9245723342895507, "nll_zh": 3.930893030166626, "nll": 3.9277326822280885, "lang_acc_valid_evalmode": 0.4975}
25
+ {"step": 400, "nll_en": 5.933121929168701, "nll_zh": 5.9383000564575195, "nll": 5.93571099281311, "lang_acc_valid_evalmode": 0.5025, "len_acc_valid_evalmode": 0.895}
26
+ {"step": 800, "nll_en": 5.360193195343018, "nll_zh": 5.36873342514038, "nll": 5.3644633102417, "lang_acc_valid_evalmode": 0.4775, "len_acc_valid_evalmode": 0.895}
27
+ {"step": 1200, "nll_en": 5.101873779296875, "nll_zh": 5.122570514678955, "nll": 5.112222146987915, "lang_acc_valid_evalmode": 0.505, "len_acc_valid_evalmode": 0.885}
28
+ {"step": 1600, "nll_en": 4.812216300964355, "nll_zh": 4.8202118301391605, "nll": 4.816214065551758, "lang_acc_valid_evalmode": 0.5175, "len_acc_valid_evalmode": 0.895}
29
+ {"step": 2000, "nll_en": 4.627658367156982, "nll_zh": 4.629207859039306, "nll": 4.628433113098144, "lang_acc_valid_evalmode": 0.5625, "len_acc_valid_evalmode": 0.895}
30
+ {"step": 400, "nll_en": 5.962598304748536, "nll_zh": 5.9708414649963375, "nll": 5.966719884872436, "lang_acc_valid_evalmode": 0.5, "len_acc_valid_evalmode": 0.895}
31
+ {"step": 800, "nll_en": 5.8880531692504885, "nll_zh": 5.873999671936035, "nll": 5.881026420593262, "lang_acc_valid_evalmode": 0.5, "len_acc_valid_evalmode": 0.895}
32
+ {"step": 1200, "nll_en": 5.872101154327392, "nll_zh": 5.871957015991211, "nll": 5.872029085159301, "lang_acc_valid_evalmode": 0.5, "len_acc_valid_evalmode": 0.895}
33
+ {"step": 1600, "nll_en": 5.438146858215332, "nll_zh": 5.434370174407959, "nll": 5.4362585163116455, "lang_acc_valid_evalmode": 0.48, "len_acc_valid_evalmode": 0.895}
34
+ {"step": 2000, "nll_en": 5.372418766021728, "nll_zh": 5.370240535736084, "nll": 5.371329650878907, "lang_acc_valid_evalmode": 0.505, "len_acc_valid_evalmode": 0.895}
planB_ramp_adv/diag_full.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ckpt": "runs/planB_ramp_adv/ckpt.pt",
3
+ "nll_en": 5.413528582510434,
4
+ "nll_zh": 5.412229951289945,
5
+ "swap_delta_en": 0.19359639734344805,
6
+ "swap_delta_zh": 0.19407416801371988,
7
+ "ablate_zero_delta_en": 0.6879320275707931,
8
+ "ablate_mean_delta_en": 0.053069320339985175,
9
+ "ablate_noise_delta_en": 6.542363524688978,
10
+ "ablate_zero_delta_zh": 0.689230658791282,
11
+ "ablate_mean_delta_zh": 0.05477833798269389,
12
+ "ablate_noise_delta_zh": 6.765621495801349,
13
+ "inv_top1_full": 0.08033826947212219,
14
+ "inv_top5_full": 0.2167019098997116,
15
+ "diag_sim_mean": 0.9676140546798706,
16
+ "offdiag_sim_mean": 0.4448440670967102,
17
+ "sim_margin": 0.5227699875831604,
18
+ "n_valid": 946
19
+ }
planB_ramp_adv/logs.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
planB_ramp_adv/probe_leakage.json ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "ckpt": "runs/planB_ramp_adv/ckpt.pt",
3
+ "max_train_groups": 8000,
4
+ "max_valid_groups": 946,
5
+ "probe_hidden": 0,
6
+ "epochs": 20,
7
+ "lr": 0.01,
8
+ "baseline_majority_lang_acc": 0.5,
9
+ "baseline_majority_len_acc": 0.9271249771118164,
10
+ "lang_probe": {
11
+ "train_acc": 0.5645,
12
+ "valid_acc": 0.5660676532769556,
13
+ "best_valid_acc": 0.5866807610993657
14
+ },
15
+ "len_probe": {
16
+ "train_acc": 0.92875,
17
+ "valid_acc": 0.9011627906976745,
18
+ "best_valid_acc": 0.9011627906976745
19
+ },
20
+ "lang_probe_length_matched": {
21
+ "train_acc": 0.5492677286742035,
22
+ "valid_acc": 0.5485115766262404,
23
+ "best_valid_acc": 0.5904079382579934
24
+ },
25
+ "lang_bucketwise_valid_acc": {
26
+ "bucket_0_acc": 0.46875,
27
+ "bucket_1_acc": 0.6500000357627869,
28
+ "bucket_2_acc": 0.573913037776947,
29
+ "bucket_3_acc": 0.5653958916664124
30
+ },
31
+ "length_matched_sizes": {
32
+ "train": 15568,
33
+ "valid": 1814
34
+ }
35
+ }