Add files using upload-large-folder tool

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0011000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_02:49:22 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0011000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0011000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0011000.pt",
+    "ckpt_step": 11000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 3.4029979318658885,
+    "unique_tokens": 1979,
+    "token_count": 16384,
+    "distinct_1": 0.12078857421875,
+    "distinct_2": 0.43399360236220474,
+    "top_token_mass": 0.13055419921875,
+    "texts_preview": [
+      "[CLS] - rate is non - americans. the population is increasingly growing, and it was a larger, greater, significant population. and the u. s., that was the line for the recession. in 1934, the world reached at $ 1, 600, a global value. in 1974, for instance, however, announced that america ’ s overseas prices reached the world ’ s population. in the 1830s, the 2. 1 percentage in the global multinational markets grew up, for real reasons, and that ’ s why it ’ s impossible to pursue non - investment goods. in theory, the equal services, the essential goods, the [SEP]",
+      "[CLS] of standards. the goal is to meet the responsibilities of respect for the european constitutional laws, and the laws to achieve the rules. the us court determined to prepare for the treaty. the eu, compliance. according to the agreement, the court authorized the provisions, and that will require approval for the provisions. the rules required for the laws in implementing the rules and that it is to meet for the european court to establish and maintain the order of standards. the rules of standards are going to apply, ” the treaty said. “ the rules in the eu, and the rules are the guidelines for the standards of the laws and that the [SEP]",
+      "[CLS] employers to reduce immigration costs and treat workers as a result. ” the problem is greater. “ i think the very wide range of factors that make them a lot more sick. it ’ s in a small way to save the economy and save the number of people, which is likely to grow more beneficial. to meet the unemployment rate, the government is trying to compensate for many jobs. but, when costs just doesn ’ t really raise wages, many americans can ’ t grow hundreds of jobs. at this point, india ’ s economic growth in 2016 is a common goal for farmers, but it ’ s not necessary to adjust the [SEP]",
+      "[CLS] the puomirai of gautha jargasi - magidiлi, nedhara, miya kializina, jetz, tuloni, jertia, kadesrostii ( movasi, azi, tudi, ta. yavispoli. kustojisafian, babali, kakuтo, che, kakunii, yajabal,, saudii, darno, rajara, kessaacao, raiji haci, zaji, turgado, ta, tartimul, polita [SEP]"
+    ],
+    "gen_ppl": 57.278638140112385,
+    "gen_nll": 4.047927746882884,
+    "gen_tokens": 15459
+  }
+]
+[watch-dualline] 2026-05-26_02:56:28 done step_0011000

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0015000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_04:25:46 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0015000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0015000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0015000.pt",
+    "ckpt_step": 15000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 3.339562537470063,
+    "unique_tokens": 1658,
+    "token_count": 16384,
+    "distinct_1": 0.1011962890625,
+    "distinct_2": 0.4011441929133858,
+    "top_token_mass": 0.10565185546875,
+    "texts_preview": [
+      "[CLS] and said, \" i came from the father in pagans. \" he was sophia ' s father and father, surprised by the loss of \" father, that he came away from the father. \" i came from the younger father ' s house, but a younger child knew that he wanted to pray before him, to try and write, \" that continued. i came to him to death, emilia, \" \" i don ' t know. \" he realized that he came his father, and began remembering, praying in the tears that he saw from his house... he gave the power of jesus and his wife, a [SEP]",
+      "[CLS] growth. ” the bank is able to demonstrate the immediate opportunity for increasing capital growth and investment - growth banks to reduce investment investments. the b. c. favored this strategy. ” the u. s. bank is the federal government ’ s largest national bank. the bank will offer the financialization opportunities of the country ’ s financial infrastructure, a real estate investment fund, and is working for the national board for growth and economic growth. “ it ’ s capital growth in the domestic financial market, ” the bank said. “ the role in the central bank and the financial crisis, responsible for the growth for the growth of banks has [SEP]",
+      "[CLS] mother. “ you are the child father as a mother, the father of the mother. ” the child father is the father of the father of the child ’ s mother. it ’ s a child with the father of the father, and the father father to the father of the rest of the family. ” father in the mother, the child is born with a father the father. the child is the father of the father, and the father of the father, and a father to a wife - born, that it ’ s the father - born. a child father ’ s father, with the big picture of the child. [SEP]",
+      "[CLS]624889886399575795757889957997534578889578675666834753934885788688758983424759675397586756395757346e899867557995768756579399857868753457759342883966839939579996395775887539e88396 [SEP]"
+    ],
+    "gen_ppl": 52.33197734359297,
+    "gen_nll": 3.9576076057125307,
+    "gen_tokens": 15591
+  }
+]
+[watch-dualline] 2026-05-26_04:32:50 done step_0015000

LTA_openwebtext_dualt/logs/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_watch/infer_lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525_step_0030000.log

@@ -0,0 +1,199 @@
+[watch-dualline] 2026-05-26_10:24:51 infer runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0030000.pt -> docs/lta_samples/metrics_20260525/owt_bert_absrope_time4_len128_C1_to_1024_mask1_sameT_every1k_dualline_dirres_c1_1024_n128/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0030000
+[decode] max_len=128 generated=1/128
+[decode] max_len=128 generated=2/128
+[decode] max_len=128 generated=3/128
+[decode] max_len=128 generated=4/128
+[decode] max_len=128 generated=5/128
+[decode] max_len=128 generated=6/128
+[decode] max_len=128 generated=7/128
+[decode] max_len=128 generated=8/128
+[decode] max_len=128 generated=9/128
+[decode] max_len=128 generated=10/128
+[decode] max_len=128 generated=11/128
+[decode] max_len=128 generated=12/128
+[decode] max_len=128 generated=13/128
+[decode] max_len=128 generated=14/128
+[decode] max_len=128 generated=15/128
+[decode] max_len=128 generated=16/128
+[decode] max_len=128 generated=17/128
+[decode] max_len=128 generated=18/128
+[decode] max_len=128 generated=19/128
+[decode] max_len=128 generated=20/128
+[decode] max_len=128 generated=21/128
+[decode] max_len=128 generated=22/128
+[decode] max_len=128 generated=23/128
+[decode] max_len=128 generated=24/128
+[decode] max_len=128 generated=25/128
+[decode] max_len=128 generated=26/128
+[decode] max_len=128 generated=27/128
+[decode] max_len=128 generated=28/128
+[decode] max_len=128 generated=29/128
+[decode] max_len=128 generated=30/128
+[decode] max_len=128 generated=31/128
+[decode] max_len=128 generated=32/128
+[decode] max_len=128 generated=33/128
+[decode] max_len=128 generated=34/128
+[decode] max_len=128 generated=35/128
+[decode] max_len=128 generated=36/128
+[decode] max_len=128 generated=37/128
+[decode] max_len=128 generated=38/128
+[decode] max_len=128 generated=39/128
+[decode] max_len=128 generated=40/128
+[decode] max_len=128 generated=41/128
+[decode] max_len=128 generated=42/128
+[decode] max_len=128 generated=43/128
+[decode] max_len=128 generated=44/128
+[decode] max_len=128 generated=45/128
+[decode] max_len=128 generated=46/128
+[decode] max_len=128 generated=47/128
+[decode] max_len=128 generated=48/128
+[decode] max_len=128 generated=49/128
+[decode] max_len=128 generated=50/128
+[decode] max_len=128 generated=51/128
+[decode] max_len=128 generated=52/128
+[decode] max_len=128 generated=53/128
+[decode] max_len=128 generated=54/128
+[decode] max_len=128 generated=55/128
+[decode] max_len=128 generated=56/128
+[decode] max_len=128 generated=57/128
+[decode] max_len=128 generated=58/128
+[decode] max_len=128 generated=59/128
+[decode] max_len=128 generated=60/128
+[decode] max_len=128 generated=61/128
+[decode] max_len=128 generated=62/128
+[decode] max_len=128 generated=63/128
+[decode] max_len=128 generated=64/128
+[decode] max_len=128 generated=65/128
+[decode] max_len=128 generated=66/128
+[decode] max_len=128 generated=67/128
+[decode] max_len=128 generated=68/128
+[decode] max_len=128 generated=69/128
+[decode] max_len=128 generated=70/128
+[decode] max_len=128 generated=71/128
+[decode] max_len=128 generated=72/128
+[decode] max_len=128 generated=73/128
+[decode] max_len=128 generated=74/128
+[decode] max_len=128 generated=75/128
+[decode] max_len=128 generated=76/128
+[decode] max_len=128 generated=77/128
+[decode] max_len=128 generated=78/128
+[decode] max_len=128 generated=79/128
+[decode] max_len=128 generated=80/128
+[decode] max_len=128 generated=81/128
+[decode] max_len=128 generated=82/128
+[decode] max_len=128 generated=83/128
+[decode] max_len=128 generated=84/128
+[decode] max_len=128 generated=85/128
+[decode] max_len=128 generated=86/128
+[decode] max_len=128 generated=87/128
+[decode] max_len=128 generated=88/128
+[decode] max_len=128 generated=89/128
+[decode] max_len=128 generated=90/128
+[decode] max_len=128 generated=91/128
+[decode] max_len=128 generated=92/128
+[decode] max_len=128 generated=93/128
+[decode] max_len=128 generated=94/128
+[decode] max_len=128 generated=95/128
+[decode] max_len=128 generated=96/128
+[decode] max_len=128 generated=97/128
+[decode] max_len=128 generated=98/128
+[decode] max_len=128 generated=99/128
+[decode] max_len=128 generated=100/128
+[decode] max_len=128 generated=101/128
+[decode] max_len=128 generated=102/128
+[decode] max_len=128 generated=103/128
+[decode] max_len=128 generated=104/128
+[decode] max_len=128 generated=105/128
+[decode] max_len=128 generated=106/128
+[decode] max_len=128 generated=107/128
+[decode] max_len=128 generated=108/128
+[decode] max_len=128 generated=109/128
+[decode] max_len=128 generated=110/128
+[decode] max_len=128 generated=111/128
+[decode] max_len=128 generated=112/128
+[decode] max_len=128 generated=113/128
+[decode] max_len=128 generated=114/128
+[decode] max_len=128 generated=115/128
+[decode] max_len=128 generated=116/128
+[decode] max_len=128 generated=117/128
+[decode] max_len=128 generated=118/128
+[decode] max_len=128 generated=119/128
+[decode] max_len=128 generated=120/128
+[decode] max_len=128 generated=121/128
+[decode] max_len=128 generated=122/128
+[decode] max_len=128 generated=123/128
+[decode] max_len=128 generated=124/128
+[decode] max_len=128 generated=125/128
+[decode] max_len=128 generated=126/128
+[decode] max_len=128 generated=127/128
+[decode] max_len=128 generated=128/128
+[
+  {
+    "checkpoint": "runs/lta_owt_bert_absrope_time4_dirichlet_len128_C1_to_1024_mask1_sameT_gbs512_b32_4gpu_1m_save1k_20260525/step_0030000.pt",
+    "ckpt_step": 30000,
+    "max_len": 128,
+    "decode_rule": "dual_line_resample",
+    "support_power": 1.0,
+    "semantic_power": 1.5,
+    "steps": 128,
+    "c_min": 1.0,
+    "c_max": 1024.0,
+    "anchor_mode": "state",
+    "model_t_mode": "flow",
+    "time_schedule": "uniform",
+    "time_logit_mean": -1.5,
+    "time_logit_std": 0.8,
+    "time_power": 2.0,
+    "input_noise_scale": 0.0,
+    "input_noise_until": 1.0,
+    "input_noise_dirichlet_concentration": 1.0,
+    "endpoint_softening": "none",
+    "endpoint_soft_power": 2.0,
+    "endpoint_soft_min_conf": 0.0,
+    "endpoint_soft_max_conf": 1.0,
+    "soft_target_decode_mode": "off",
+    "soft_target_power": 1.0,
+    "soft_target_min_conf": 0.0,
+    "soft_target_max_conf": 1.0,
+    "soft_target_debias_start": 0.7,
+    "final_from": "blend",
+    "final_decode": "argmax",
+    "final_sample_temp": 1.0,
+    "final_top_k": 0,
+    "final_top_p": 1.0,
+    "commit_mode": "off",
+    "commit_conf_threshold": 0.0,
+    "commit_margin_threshold": 0.0,
+    "commit_start": 0.0,
+    "commit_min_ratio": 0.0,
+    "commit_max_ratio": 1.0,
+    "commit_power": 2.0,
+    "commit_freq_max_frac": 0.08,
+    "early_temp": 2.8,
+    "late_temp": 1.45,
+    "temp_end": 0.55,
+    "temp_power": 1.5,
+    "pos_extend": "repeat",
+    "fixed_first_token_id": null,
+    "fixed_first_token_text": "",
+    "fixed_first_initial_argmax": false,
+    "use_ema": false,
+    "n_samples": 128,
+    "sample_entropy": 3.380224297862822,
+    "unique_tokens": 1491,
+    "token_count": 16384,
+    "distinct_1": 0.09100341796875,
+    "distinct_2": 0.42704232283464566,
+    "top_token_mass": 0.1070556640625,
+    "texts_preview": [
+      "[CLS] re saying it, i never apologize, ' \" she said. \" i was like, \" yeah. it ' s just really okay, but always got that, and i thought, \" yeah, it ' t happened. it ' s ridiculous. \" that way, you ' ve always been here, ' give me ' and i always ask ' wait, so, you ' ve got me ', i never, ' \" she said. \" \" i don ' t have time, and you ’ ve gotten me wrong. that ' s okay, but it ’ s just okay, and it ’ s okay. i never [SEP]",
+      "[CLS] day. ” the problem is, we know that she cried, her head, realized …, “ it is not her fault, but it. in that case, i ’ m not talking, though. ” so believe …, she says, “ no, so, now, right, there is a terrible moment, for him, and for that moment, on the other hand, and things that never missed him and never any worse for him. i don ’ t really think he is, to so speak, because, now know, “ say, ” so ask that, everyone else, because i didn ’ t believe my [SEP]",
+      "[CLS] tests, you are ready. this is not a hard working application, but it allows you to get your own web application applications in your application application ( with a server ). now you have just a small number of specific frameworks that have a specific application and running some tests, so make a try. in this example, you still need to write some tests and / or tests. however, if you can give a google web application application, it requires you to write a tests to test, you can. when i ’ m more interested in running in a google web application, you can always test using a google web application. [SEP]",
+      "[CLS] kenney, darren kerr, steve burgughner, chris burbaberry, joel kerr, john peberry, steve burgberry, terry kerley, aaron kerr, steve hur, larry hullenberry, brad moyer, steve wallace, neil kerr, michael mullen, david p. s., aaron kerr, jeff mccown, joey wallace, gary mulls, gary mulls, joey perkins, joel mulls, dan kessler, liam berry, gary wallace, mark blemo, tony korballiff, gary mullensoff, adam murley, corey pebiak, [SEP]"
+    ],
+    "gen_ppl": 43.68256760090672,
+    "gen_nll": 3.7769491117810916,
+    "gen_tokens": 16191
+  }
+]
+[watch-dualline] 2026-05-26_10:32:19 done step_0030000

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/testing/__init__.pyi

@@ -0,0 +1,50 @@
+from numpy._pytesttester import PytestTester
+
+from unittest import (
+    TestCase as TestCase,
+)
+
+from numpy.testing._private.utils import (
+    assert_equal as assert_equal,
+    assert_almost_equal as assert_almost_equal,
+    assert_approx_equal as assert_approx_equal,
+    assert_array_equal as assert_array_equal,
+    assert_array_less as assert_array_less,
+    assert_string_equal as assert_string_equal,
+    assert_array_almost_equal as assert_array_almost_equal,
+    assert_raises as assert_raises,
+    build_err_msg as build_err_msg,
+    decorate_methods as decorate_methods,
+    jiffies as jiffies,
+    memusage as memusage,
+    print_assert_equal as print_assert_equal,
+    rundocs as rundocs,
+    runstring as runstring,
+    verbose as verbose,
+    measure as measure,
+    assert_ as assert_,
+    assert_array_almost_equal_nulp as assert_array_almost_equal_nulp,
+    assert_raises_regex as assert_raises_regex,
+    assert_array_max_ulp as assert_array_max_ulp,
+    assert_warns as assert_warns,
+    assert_no_warnings as assert_no_warnings,
+    assert_allclose as assert_allclose,
+    IgnoreException as IgnoreException,
+    clear_and_catch_warnings as clear_and_catch_warnings,
+    SkipTest as SkipTest,
+    KnownFailureException as KnownFailureException,
+    temppath as temppath,
+    tempdir as tempdir,
+    IS_PYPY as IS_PYPY,
+    IS_PYSTON as IS_PYSTON,
+    HAS_REFCOUNT as HAS_REFCOUNT,
+    suppress_warnings as suppress_warnings,
+    assert_array_compare as assert_array_compare,
+    assert_no_gc_cycles as assert_no_gc_cycles,
+    break_cycles as break_cycles,
+    HAS_LAPACK64 as HAS_LAPACK64,
+)
+
+__all__: list[str]
+__path__: list[str]
+test: PytestTester

LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/numpy/testing/overrides.py

@@ -0,0 +1,83 @@
+"""Tools for testing implementations of __array_function__ and ufunc overrides
+
+
+"""
+
+from numpy.core.overrides import ARRAY_FUNCTIONS as _array_functions
+from numpy import ufunc as _ufunc
+import numpy.core.umath as _umath
+
+def get_overridable_numpy_ufuncs():
+    """List all numpy ufuncs overridable via `__array_ufunc__`
+
+    Parameters
+    ----------
+    None
+
+    Returns
+    -------
+    set
+        A set containing all overridable ufuncs in the public numpy API.
+    """
+    ufuncs = {obj for obj in _umath.__dict__.values()
+              if isinstance(obj, _ufunc)}
+    return ufuncs
+    
+
+def allows_array_ufunc_override(func):
+    """Determine if a function can be overridden via `__array_ufunc__`
+
+    Parameters
+    ----------
+    func : callable
+        Function that may be overridable via `__array_ufunc__`
+
+    Returns
+    -------
+    bool
+        `True` if `func` is overridable via `__array_ufunc__` and
+        `False` otherwise.
+
+    Notes
+    -----
+    This function is equivalent to ``isinstance(func, np.ufunc)`` and
+    will work correctly for ufuncs defined outside of Numpy.
+
+    """
+    return isinstance(func, np.ufunc)
+
+
+def get_overridable_numpy_array_functions():
+    """List all numpy functions overridable via `__array_function__`
+
+    Parameters
+    ----------
+    None
+
+    Returns
+    -------
+    set
+        A set containing all functions in the public numpy API that are
+        overridable via `__array_function__`.
+
+    """
+    # 'import numpy' doesn't import recfunctions, so make sure it's imported
+    # so ufuncs defined there show up in the ufunc listing
+    from numpy.lib import recfunctions
+    return _array_functions.copy()
+
+def allows_array_function_override(func):
+    """Determine if a Numpy function can be overridden via `__array_function__`
+
+    Parameters
+    ----------
+    func : callable
+        Function that may be overridable via `__array_function__`
+
+    Returns
+    -------
+    bool
+        `True` if `func` is a function in the Numpy API that is
+        overridable via `__array_function__` and `False` otherwise.
+    """
+    return func in _array_functions

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm_aggressive_20260612/gpu1.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.1 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:03:44+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:03:44+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:03:51+00:00] done decode=32 chunk=0
+[2026-06-11T22:03:51+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:03:58+00:00] done decode=32 chunk=1
+[2026-06-11T22:03:58+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:04:05+00:00] done decode=32 chunk=2
+[2026-06-11T22:04:05+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:04:12+00:00] done decode=32 chunk=3
+[2026-06-11T22:04:12+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:04:19+00:00] done decode=32 chunk=4
+[2026-06-11T22:04:19+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:04:26+00:00] done decode=32 chunk=5
+[2026-06-11T22:04:26+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:04:33+00:00] done decode=32 chunk=6
+[2026-06-11T22:04:33+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:04:39+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.239170058026044	4.902588437310562	0.07085875897262485	0.4291708936641151	0.02978834596289522	62	62	61905	65059	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	25.9228719809299	4.9312184748129315	0.07309028053093135	0.4427194011862095	0.029876782061244233	0	0	57871	63059	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:05:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.1 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.2 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:05:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:05:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:05:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:05:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:05:25+00:00] done decode=32 chunk=1
+[2026-06-11T22:05:25+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:05:32+00:00] done decode=32 chunk=2
+[2026-06-11T22:05:32+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:05:38+00:00] done decode=32 chunk=3
+[2026-06-11T22:05:38+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:05:45+00:00] done decode=32 chunk=4
+[2026-06-11T22:05:45+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:05:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:05:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:05:59+00:00] done decode=32 chunk=6
+[2026-06-11T22:05:59+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:06:06+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	25.496219132224898	5.012091110584844	0.08152456757997857	0.46550536515177027	0.032603704270626054	61	63	62377	65330	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	30.464440218319435	5.046043523612475	0.08429616029881928	0.4812997578385907	0.032651704599411226	0	0	58066	63182	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:06:20+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.2 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.4 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:06:20+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:06:20+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:06:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:06:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:06:34+00:00] done decode=32 chunk=1
+[2026-06-11T22:06:34+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:06:40+00:00] done decode=32 chunk=2
+[2026-06-11T22:06:40+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:06:47+00:00] done decode=32 chunk=3
+[2026-06-11T22:06:47+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:06:54+00:00] done decode=32 chunk=4
+[2026-06-11T22:06:54+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:07:01+00:00] done decode=32 chunk=5
+[2026-06-11T22:07:01+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:07:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:07:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:07:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.752981342557536	4.9228951983362	0.06280710885238876	0.40541285246303155	0.0322684341849465	62	64	62494	65327	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	23.985858504581056	4.9334688460376785	0.06426703005811131	0.41477397324645215	0.03301849850414298	0	0	59501	63843	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:07:29+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.8 s=1.4 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.1 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:07:29+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:07:29+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:07:36+00:00] done decode=32 chunk=0
+[2026-06-11T22:07:36+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:07:43+00:00] done decode=32 chunk=1
+[2026-06-11T22:07:43+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:07:50+00:00] done decode=32 chunk=2
+[2026-06-11T22:07:50+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:07:56+00:00] done decode=32 chunk=3
+[2026-06-11T22:07:56+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:08:03+00:00] done decode=32 chunk=4
+[2026-06-11T22:08:03+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:08:10+00:00] done decode=32 chunk=5
+[2026-06-11T22:08:10+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:08:17+00:00] done decode=32 chunk=6
+[2026-06-11T22:08:17+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:08:24+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	24.75761613174975	4.948941612167286	0.07653539435472718	0.4497541487400123	0.029271215869454987	59	59	61442	65081	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	29.107271779629794	4.978997653339936	0.0789056807020879	0.46365148514851484	0.02892627443525647	0	0	57502	63126	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:08:37+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.1 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.2 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:08:37+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:08:37+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:08:44+00:00] done decode=32 chunk=0
+[2026-06-11T22:08:44+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:08:51+00:00] done decode=32 chunk=1
+[2026-06-11T22:08:51+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:08:57+00:00] done decode=32 chunk=2
+[2026-06-11T22:08:57+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:09:05+00:00] done decode=32 chunk=3
+[2026-06-11T22:09:05+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:09:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:09:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:09:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:09:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:09:25+00:00] done decode=32 chunk=6
+[2026-06-11T22:09:25+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:09:31+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	26.822656545037535	5.043757471946711	0.08566572758451739	0.474529558573594	0.03316389024987751	62	62	61748	65312	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	32.06038094766469	5.079950572057259	0.08849165546152021	0.4902078653462841	0.034264019615597566	0	0	57527	63215	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:09:45+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.2 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.4 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:09:45+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:09:45+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:09:52+00:00] done decode=32 chunk=0
+[2026-06-11T22:09:52+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:09:59+00:00] done decode=32 chunk=1
+[2026-06-11T22:09:59+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:10:06+00:00] done decode=32 chunk=2
+[2026-06-11T22:10:06+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:10:13+00:00] done decode=32 chunk=3
+[2026-06-11T22:10:13+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:10:19+00:00] done decode=32 chunk=4
+[2026-06-11T22:10:19+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:10:27+00:00] done decode=32 chunk=5
+[2026-06-11T22:10:27+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:10:33+00:00] done decode=32 chunk=6
+[2026-06-11T22:10:33+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:10:40+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.060738024824808	4.962277223248929	0.06517405305934673	0.41811713107086446	0.030746817972703573	64	66	62088	65210	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	25.541116218845463	4.975016531041534	0.06672005527290999	0.42807342618909283	0.03146787262106652	0	0	59005	63684	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:10:54+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.7 s=1.4 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.1 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:10:54+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:10:54+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:11:01+00:00] done decode=32 chunk=0
+[2026-06-11T22:11:01+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:11:08+00:00] done decode=32 chunk=1
+[2026-06-11T22:11:08+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:11:15+00:00] done decode=32 chunk=2
+[2026-06-11T22:11:15+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:11:22+00:00] done decode=32 chunk=3
+[2026-06-11T22:11:22+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:11:28+00:00] done decode=32 chunk=4
+[2026-06-11T22:11:28+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:11:35+00:00] done decode=32 chunk=5
+[2026-06-11T22:11:35+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:11:42+00:00] done decode=32 chunk=6
+[2026-06-11T22:11:42+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:11:49+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	26.08725378809399	4.920166974326847	0.08194956949569496	0.46327588062547087	0.030350553505535054	62	65	60489	65040	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	30.956383855549362	4.947767922250899	0.08455217487864462	0.4779573901042245	0.030806814937022114	0	0	56449	63038	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:12:02+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.1 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.2 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:12:02+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:12:02+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:12:09+00:00] done decode=32 chunk=0
+[2026-06-11T22:12:09+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:12:16+00:00] done decode=32 chunk=1
+[2026-06-11T22:12:16+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:12:23+00:00] done decode=32 chunk=2
+[2026-06-11T22:12:23+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:12:30+00:00] done decode=32 chunk=3
+[2026-06-11T22:12:30+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:12:37+00:00] done decode=32 chunk=4
+[2026-06-11T22:12:37+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:12:44+00:00] done decode=32 chunk=5
+[2026-06-11T22:12:44+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:12:51+00:00] done decode=32 chunk=6
+[2026-06-11T22:12:51+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:12:58+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	28.868490267462853	5.067419414924427	0.09041688395349551	0.49428654692700585	0.036488849351207095	63	65	62023	65198	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	36.10693495145991	5.115202957609105	0.09393673810851726	0.5134812123529974	0.032634849812763926	0	0	57106	62755	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:13:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.2 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.4 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:13:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:13:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:13:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:13:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:13:25+00:00] done decode=32 chunk=1
+[2026-06-11T22:13:25+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:13:32+00:00] done decode=32 chunk=2
+[2026-06-11T22:13:32+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:13:38+00:00] done decode=32 chunk=3
+[2026-06-11T22:13:39+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:13:45+00:00] done decode=32 chunk=4
+[2026-06-11T22:13:45+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:13:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:13:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:13:59+00:00] done decode=32 chunk=6
+[2026-06-11T22:13:59+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:14:06+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.116440481580412	4.91461829854095	0.06612495208892295	0.42547221881516006	0.02790341126868532	62	63	61653	65225	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	25.764466831501306	4.926878093686284	0.06772207563764292	0.4357713872877762	0.028568287473300667	0	0	58514	63672	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:14:20+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.6 s=1.4 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.1 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:14:20+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:14:20+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:14:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:14:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:14:34+00:00] done decode=32 chunk=1
+[2026-06-11T22:14:34+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:14:41+00:00] done decode=32 chunk=2
+[2026-06-11T22:14:41+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:14:48+00:00] done decode=32 chunk=3
+[2026-06-11T22:14:48+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:14:54+00:00] done decode=32 chunk=4
+[2026-06-11T22:14:54+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:15:02+00:00] done decode=32 chunk=5
+[2026-06-11T22:15:02+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:15:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:15:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:15:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	27.43856590016902	4.990166604929867	0.08700462529004103	0.4787632921507161	0.0294113127525854	61	61	60579	65077	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	32.69493895573401	5.01769384256394	0.08971189502709898	0.49368472765883264	0.028905581439574023	0	0	56592	63102	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:15:29+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.1 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.2 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:15:29+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:15:29+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:15:35+00:00] done decode=32 chunk=0
+[2026-06-11T22:15:35+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:15:42+00:00] done decode=32 chunk=1
+[2026-06-11T22:15:42+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:15:49+00:00] done decode=32 chunk=2
+[2026-06-11T22:15:49+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:15:56+00:00] done decode=32 chunk=3
+[2026-06-11T22:15:56+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:16:03+00:00] done decode=32 chunk=4
+[2026-06-11T22:16:03+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:16:10+00:00] done decode=32 chunk=5
+[2026-06-11T22:16:10+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:16:16+00:00] done decode=32 chunk=6
+[2026-06-11T22:16:16+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:16:23+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	30.479960794000636	5.121093410215607	0.09598662720260094	0.5163328528049566	0.03484288496633797	62	63	61657	65207	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	37.51358258611811	5.1640988263936665	0.09940285233300511	0.5346779957119034	0.031112028713909095	0	0	57148	62966	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:16:37+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.2 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.4 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=1
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:16:37+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:16:37+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:16:44+00:00] done decode=32 chunk=0
+[2026-06-11T22:16:44+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:16:50+00:00] done decode=32 chunk=1
+[2026-06-11T22:16:50+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:16:57+00:00] done decode=32 chunk=2
+[2026-06-11T22:16:57+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:17:04+00:00] done decode=32 chunk=3
+[2026-06-11T22:17:04+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:17:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:17:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:17:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:17:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:17:25+00:00] done decode=32 chunk=6
+[2026-06-11T22:17:25+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:17:31+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.95895366620678	4.96404394235026	0.07053529060600958	0.44007163850110215	0.027292626551761085	62	64	61624	65329	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	27.321054615698465	4.983244999984837	0.07243460764587525	0.45189027745028687	0.02802754024144869	0	0	58168	63616	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:17:44+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=1 m=-0.5 s=1.4 gamma=1.0 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm_aggressive_20260612/gpu2.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.1 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:03:44+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:03:44+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:03:51+00:00] done decode=32 chunk=0
+[2026-06-11T22:03:51+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:03:58+00:00] done decode=32 chunk=1
+[2026-06-11T22:03:58+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:04:04+00:00] done decode=32 chunk=2
+[2026-06-11T22:04:04+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:04:11+00:00] done decode=32 chunk=3
+[2026-06-11T22:04:11+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:04:18+00:00] done decode=32 chunk=4
+[2026-06-11T22:04:18+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:04:25+00:00] done decode=32 chunk=5
+[2026-06-11T22:04:25+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:04:32+00:00] done decode=32 chunk=6
+[2026-06-11T22:04:32+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:04:39+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	25.854354785385397	5.004204104607414	0.08112085351646331	0.46668199586111747	0.03344778956404439	59	59	62432	65236	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	31.116717254475276	5.041667335033597	0.08393471744198956	0.48283847227509197	0.03259369696585196	0	0	58033	63049	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:05:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.1 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.3 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:05:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:05:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:05:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:05:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:05:24+00:00] done decode=32 chunk=1
+[2026-06-11T22:05:24+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:05:31+00:00] done decode=32 chunk=2
+[2026-06-11T22:05:31+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:05:38+00:00] done decode=32 chunk=3
+[2026-06-11T22:05:38+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:05:45+00:00] done decode=32 chunk=4
+[2026-06-11T22:05:45+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:05:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:05:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:05:58+00:00] done decode=32 chunk=6
+[2026-06-11T22:05:58+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:06:05+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.845361496364216	4.885529105144743	0.06044873267478367	0.39902594419089044	0.03180948005207137	62	62	62327	65295	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	23.940522036916708	4.895879009087364	0.06180417456117	0.40794212520747863	0.03252274399887259	0	0	59434	63863	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:06:19+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.3 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.4 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:06:19+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:06:19+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:06:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:06:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:06:33+00:00] done decode=32 chunk=1
+[2026-06-11T22:06:33+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:06:40+00:00] done decode=32 chunk=2
+[2026-06-11T22:06:40+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:06:47+00:00] done decode=32 chunk=3
+[2026-06-11T22:06:47+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:06:54+00:00] done decode=32 chunk=4
+[2026-06-11T22:06:54+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:07:01+00:00] done decode=32 chunk=5
+[2026-06-11T22:07:01+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:07:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:07:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:07:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.771040404976237	4.975522915143686	0.07011233891456732	0.43142476697736354	0.03194159600844838	63	63	62267	65338	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	26.619272141943423	4.990216864380564	0.07189829710429255	0.44235175942493016	0.03275523816997567	0	0	58995	63715	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:07:29+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.8 s=1.4 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.1 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:07:29+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:07:29+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:07:36+00:00] done decode=32 chunk=0
+[2026-06-11T22:07:36+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:07:43+00:00] done decode=32 chunk=1
+[2026-06-11T22:07:43+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:07:49+00:00] done decode=32 chunk=2
+[2026-06-11T22:07:49+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:07:56+00:00] done decode=32 chunk=3
+[2026-06-11T22:07:56+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:08:03+00:00] done decode=32 chunk=4
+[2026-06-11T22:08:03+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:08:10+00:00] done decode=32 chunk=5
+[2026-06-11T22:08:10+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:08:17+00:00] done decode=32 chunk=6
+[2026-06-11T22:08:17+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:08:23+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	27.110385529940846	5.021679933185779	0.08765880495915897	0.4871574817630111	0.035170796744977244	63	65	61994	65253	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	33.42964956544722	5.06379631540308	0.09093221821188367	0.5053742924378299	0.0322452419188145	0	0	57243	62893	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:08:37+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.1 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.3 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:08:37+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:08:37+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:08:44+00:00] done decode=32 chunk=0
+[2026-06-11T22:08:44+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:08:50+00:00] done decode=32 chunk=1
+[2026-06-11T22:08:50+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:08:57+00:00] done decode=32 chunk=2
+[2026-06-11T22:08:57+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:09:04+00:00] done decode=32 chunk=3
+[2026-06-11T22:09:04+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:09:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:09:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:09:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:09:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:09:24+00:00] done decode=32 chunk=6
+[2026-06-11T22:09:24+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:09:31+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.728760662391164	4.855898853033717	0.06134790862881989	0.4030008420730307	0.029426174291138467	61	62	61995	65316	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	23.413444720021076	4.873284773460608	0.06300426214554204	0.4139221792331163	0.030228205652454272	0	0	58494	63583	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:09:45+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.3 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.4 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:09:45+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:09:45+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:09:52+00:00] done decode=32 chunk=0
+[2026-06-11T22:09:52+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:09:59+00:00] done decode=32 chunk=1
+[2026-06-11T22:09:59+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:10:05+00:00] done decode=32 chunk=2
+[2026-06-11T22:10:05+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:10:12+00:00] done decode=32 chunk=3
+[2026-06-11T22:10:12+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:10:19+00:00] done decode=32 chunk=4
+[2026-06-11T22:10:19+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:10:26+00:00] done decode=32 chunk=5
+[2026-06-11T22:10:26+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:10:33+00:00] done decode=32 chunk=6
+[2026-06-11T22:10:33+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:10:40+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	24.859095876019616	5.035434036954304	0.07354990886673099	0.45156843524077933	0.033037724578412905	63	63	62033	65289	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	28.218542679157547	5.0543044187222534	0.07548623449316835	0.46349056603773586	0.03391456109180673	0	0	58629	63601	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:10:53+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.7 s=1.4 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.1 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:10:53+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:10:53+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:11:00+00:00] done decode=32 chunk=0
+[2026-06-11T22:11:00+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:11:07+00:00] done decode=32 chunk=1
+[2026-06-11T22:11:07+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:11:14+00:00] done decode=32 chunk=2
+[2026-06-11T22:11:14+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:11:21+00:00] done decode=32 chunk=3
+[2026-06-11T22:11:21+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:11:28+00:00] done decode=32 chunk=4
+[2026-06-11T22:11:28+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:11:35+00:00] done decode=32 chunk=5
+[2026-06-11T22:11:35+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:11:42+00:00] done decode=32 chunk=6
+[2026-06-11T22:11:42+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:11:49+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	29.977092278562463	5.038236965402097	0.0950229026407206	0.5076317690640515	0.03247871130375972	63	65	62053	65058	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	36.47180376836833	5.07661551024827	0.09829834605597965	0.5251514814166892	0.033015267175572516	0	0	57661	62880	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:12:02+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.1 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.3 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:12:03+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:12:03+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:12:10+00:00] done decode=32 chunk=0
+[2026-06-11T22:12:10+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:12:16+00:00] done decode=32 chunk=1
+[2026-06-11T22:12:16+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:12:23+00:00] done decode=32 chunk=2
+[2026-06-11T22:12:23+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:12:30+00:00] done decode=32 chunk=3
+[2026-06-11T22:12:30+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:12:37+00:00] done decode=32 chunk=4
+[2026-06-11T22:12:37+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:12:44+00:00] done decode=32 chunk=5
+[2026-06-11T22:12:44+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:12:51+00:00] done decode=32 chunk=6
+[2026-06-11T22:12:51+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:12:58+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	22.857431909527506	4.928224686310635	0.07028460206286687	0.43090975968612066	0.028429554475930666	62	63	61496	65249	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	26.528950871871306	4.952697664712585	0.07241205078001642	0.44390582810945667	0.028247963115012948	0	0	57629	63332	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:13:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.3 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.4 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:13:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:13:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:13:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:13:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:13:25+00:00] done decode=32 chunk=1
+[2026-06-11T22:13:25+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:13:32+00:00] done decode=32 chunk=2
+[2026-06-11T22:13:32+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:13:39+00:00] done decode=32 chunk=3
+[2026-06-11T22:13:39+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:13:46+00:00] done decode=32 chunk=4
+[2026-06-11T22:13:46+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:13:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:13:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:13:59+00:00] done decode=32 chunk=6
+[2026-06-11T22:13:59+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:14:06+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	25.81992599814849	5.024819863688342	0.07371353353537209	0.4553594661197251	0.030489484801175498	62	64	61936	65334	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	29.479426711940192	5.046576623240096	0.07568136808022383	0.4675500227912167	0.03130992424004275	0	0	58487	63622	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:14:20+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.6 s=1.4 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.1 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:14:20+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64
+[2026-06-11T22:14:20+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:14:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:14:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:14:34+00:00] done decode=32 chunk=1
+[2026-06-11T22:14:34+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:14:41+00:00] done decode=32 chunk=2
+[2026-06-11T22:14:41+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:14:47+00:00] done decode=32 chunk=3
+[2026-06-11T22:14:47+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:14:54+00:00] done decode=32 chunk=4
+[2026-06-11T22:14:54+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:15:01+00:00] done decode=32 chunk=5
+[2026-06-11T22:15:01+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:15:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:15:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:15:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	32.51736815215705	5.0877369085662005	0.09877530924979999	0.5229171474728825	0.032478921779801835	58	58	61593	64996	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	39.72928580887329	5.12604755567189	0.10209145129224652	0.5404936857842669	0.03357455268389662	0	0	57317	62875	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p1_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:15:28+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.1 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.3 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:15:28+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:15:28+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:15:35+00:00] done decode=32 chunk=0
+[2026-06-11T22:15:35+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:15:42+00:00] done decode=32 chunk=1
+[2026-06-11T22:15:42+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:15:48+00:00] done decode=32 chunk=2
+[2026-06-11T22:15:48+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:15:55+00:00] done decode=32 chunk=3
+[2026-06-11T22:15:55+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:16:02+00:00] done decode=32 chunk=4
+[2026-06-11T22:16:02+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:16:09+00:00] done decode=32 chunk=5
+[2026-06-11T22:16:09+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:16:16+00:00] done decode=32 chunk=6
+[2026-06-11T22:16:16+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:16:23+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	24.218062449798417	4.927662398284183	0.07173455757036036	0.43938091143594155	0.028973268436487586	59	60	61041	65129	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	28.44472768892935	4.95534843972475	0.07393162393162393	0.45284034251887495	0.029186451408673632	0	0	57111	63180	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:16:36+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.3 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.4 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=2
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:16:36+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:16:36+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:16:43+00:00] done decode=32 chunk=0
+[2026-06-11T22:16:43+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:16:50+00:00] done decode=32 chunk=1
+[2026-06-11T22:16:50+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:16:57+00:00] done decode=32 chunk=2
+[2026-06-11T22:16:57+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:17:04+00:00] done decode=32 chunk=3
+[2026-06-11T22:17:04+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:17:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:17:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:17:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:17:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:17:24+00:00] done decode=32 chunk=6
+[2026-06-11T22:17:24+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:17:31+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	26.056941698908155	5.022691895425508	0.07904690443011807	0.470506263207865	0.030350826148875243	63	64	62390	65303	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	30.798753237880796	5.0568829629804135	0.08158780846388541	0.4856614392310611	0.029198349589768722	0	0	58268	63257	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:17:45+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=2 m=-0.5 s=1.4 gamma=0.75 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/logs/infer_lr3e3_170k_logitnorm_aggressive_20260612/gpu3.log

@@ -0,0 +1,372 @@
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.2 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:03:44+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:03:44+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:03:51+00:00] done decode=32 chunk=0
+[2026-06-11T22:03:51+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:03:58+00:00] done decode=32 chunk=1
+[2026-06-11T22:03:58+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:04:05+00:00] done decode=32 chunk=2
+[2026-06-11T22:04:05+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:04:12+00:00] done decode=32 chunk=3
+[2026-06-11T22:04:12+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:04:18+00:00] done decode=32 chunk=4
+[2026-06-11T22:04:18+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:04:26+00:00] done decode=32 chunk=5
+[2026-06-11T22:04:26+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:04:33+00:00] done decode=32 chunk=6
+[2026-06-11T22:04:33+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:04:39+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	20.203193769933044	4.824904380328837	0.06038877287802481	0.39426044497673773	0.031752855914506814	62	62	62149	65128	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	22.959212885809723	4.843907850954462	0.06212092494313874	0.4054903571259339	0.03264783927217589	0	0	58483	63312	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p8_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:05:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.2 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.3 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:05:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:05:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:05:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:05:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:05:25+00:00] done decode=32 chunk=1
+[2026-06-11T22:05:25+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:05:32+00:00] done decode=32 chunk=2
+[2026-06-11T22:05:32+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:05:38+00:00] done decode=32 chunk=3
+[2026-06-11T22:05:38+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:05:45+00:00] done decode=32 chunk=4
+[2026-06-11T22:05:45+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:05:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:05:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:05:59+00:00] done decode=32 chunk=6
+[2026-06-11T22:05:59+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:06:05+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.66583345660286	4.976099656092876	0.06959078657191865	0.43138065701814843	0.03168647390345503	64	64	62054	65296	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	26.912408796981207	4.994811950369471	0.07147352191560996	0.4430547033558314	0.03255089518894937	0	0	58559	63562	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p8_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:06:19+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.3 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.4 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:06:19+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:06:19+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:06:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:06:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:06:34+00:00] done decode=32 chunk=1
+[2026-06-11T22:06:34+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:06:40+00:00] done decode=32 chunk=2
+[2026-06-11T22:06:40+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:06:47+00:00] done decode=32 chunk=3
+[2026-06-11T22:06:47+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:06:54+00:00] done decode=32 chunk=4
+[2026-06-11T22:06:54+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:07:01+00:00] done decode=32 chunk=5
+[2026-06-11T22:07:01+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:07:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:07:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:07:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	27.249332553205008	5.0630319318752415	0.08294655576490051	0.4749088569590392	0.03153960449121517	61	65	62446	65283	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	31.78467459048637	5.0894845271680165	0.0854127890469731	0.4889821284918845	0.03246159184832329	0	0	58647	63398	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p8_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:07:29+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.8 s=1.4 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.2 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:07:29+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:07:29+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:07:36+00:00] done decode=32 chunk=0
+[2026-06-11T22:07:36+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:07:43+00:00] done decode=32 chunk=1
+[2026-06-11T22:07:43+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:07:50+00:00] done decode=32 chunk=2
+[2026-06-11T22:07:50+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:07:56+00:00] done decode=32 chunk=3
+[2026-06-11T22:07:56+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:08:03+00:00] done decode=32 chunk=4
+[2026-06-11T22:08:03+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:08:10+00:00] done decode=32 chunk=5
+[2026-06-11T22:08:10+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:08:17+00:00] done decode=32 chunk=6
+[2026-06-11T22:08:17+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:08:24+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	21.629873622400094	4.885402487605441	0.06577916992952232	0.41393606240211284	0.027899335145177884	60	61	61629	65127	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	24.855613286553762	4.9089082666421815	0.06770125478049244	0.42599996839293897	0.02786118398179462	0	0	57902	63278	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p7_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:08:37+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.2 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.3 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:08:37+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:08:37+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:08:44+00:00] done decode=32 chunk=0
+[2026-06-11T22:08:44+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:08:51+00:00] done decode=32 chunk=1
+[2026-06-11T22:08:51+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:08:58+00:00] done decode=32 chunk=2
+[2026-06-11T22:08:58+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:09:04+00:00] done decode=32 chunk=3
+[2026-06-11T22:09:04+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:09:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:09:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:09:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:09:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:09:25+00:00] done decode=32 chunk=6
+[2026-06-11T22:09:25+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:09:32+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	24.51188180834565	4.948562754084104	0.0701843084984143	0.437076758081814	0.031545403012057424	62	67	61729	65271	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	28.408654154205724	4.972840363806111	0.07225779993059718	0.44994242629777437	0.032445818480078235	0	0	57957	63398	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p7_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:09:45+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.3 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.4 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:09:45+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:09:45+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:09:52+00:00] done decode=32 chunk=0
+[2026-06-11T22:09:52+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:09:59+00:00] done decode=32 chunk=1
+[2026-06-11T22:09:59+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:10:06+00:00] done decode=32 chunk=2
+[2026-06-11T22:10:06+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:10:12+00:00] done decode=32 chunk=3
+[2026-06-11T22:10:12+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:10:19+00:00] done decode=32 chunk=4
+[2026-06-11T22:10:19+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:10:27+00:00] done decode=32 chunk=5
+[2026-06-11T22:10:27+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:10:33+00:00] done decode=32 chunk=6
+[2026-06-11T22:10:33+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:10:40+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	28.21707155811356	5.071760328993291	0.08285390782789645	0.47506694208553285	0.033110961503151966	63	63	62147	65356	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	33.12850540030104	5.0973585833389885	0.08536469994639084	0.4894908626480188	0.03410488474031093	0	0	58248	63422	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p7_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:10:54+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.7 s=1.4 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.2 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:10:54+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:10:54+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:11:01+00:00] done decode=32 chunk=0
+[2026-06-11T22:11:01+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:11:07+00:00] done decode=32 chunk=1
+[2026-06-11T22:11:07+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:11:15+00:00] done decode=32 chunk=2
+[2026-06-11T22:11:15+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:11:21+00:00] done decode=32 chunk=3
+[2026-06-11T22:11:21+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:11:28+00:00] done decode=32 chunk=4
+[2026-06-11T22:11:28+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:11:35+00:00] done decode=32 chunk=5
+[2026-06-11T22:11:35+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:11:42+00:00] done decode=32 chunk=6
+[2026-06-11T22:11:42+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:11:49+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.09369480969775	4.894980710058736	0.07052207848077928	0.42264730736728895	0.02716405985926313	62	63	60323	65086	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	26.756879833526835	4.9167257129621325	0.0725476246937001	0.4348183514086066	0.027491897873685875	0	0	56622	63255	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p6_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:12:02+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.2 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.3 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:12:02+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:12:02+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:12:09+00:00] done decode=32 chunk=0
+[2026-06-11T22:12:09+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:12:16+00:00] done decode=32 chunk=1
+[2026-06-11T22:12:16+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:12:23+00:00] done decode=32 chunk=2
+[2026-06-11T22:12:23+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:12:30+00:00] done decode=32 chunk=3
+[2026-06-11T22:12:30+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:12:37+00:00] done decode=32 chunk=4
+[2026-06-11T22:12:37+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:12:44+00:00] done decode=32 chunk=5
+[2026-06-11T22:12:44+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:12:51+00:00] done decode=32 chunk=6
+[2026-06-11T22:12:51+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:12:57+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	25.474951624188755	4.980683779309242	0.07681343775383531	0.4599987739087788	0.03013072997287315	60	60	61714	65249	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	30.131493102100617	5.014751104335049	0.07925913037976685	0.47467653664863496	0.02907169859070275	0	0	57637	63223	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p6_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:13:11+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.3 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.4 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:13:11+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:13:11+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:13:18+00:00] done decode=32 chunk=0
+[2026-06-11T22:13:18+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:13:25+00:00] done decode=32 chunk=1
+[2026-06-11T22:13:25+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:13:32+00:00] done decode=32 chunk=2
+[2026-06-11T22:13:32+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:13:39+00:00] done decode=32 chunk=3
+[2026-06-11T22:13:39+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:13:45+00:00] done decode=32 chunk=4
+[2026-06-11T22:13:45+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:13:52+00:00] done decode=32 chunk=5
+[2026-06-11T22:13:52+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:13:59+00:00] done decode=32 chunk=6
+[2026-06-11T22:13:59+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:14:06+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	28.73819607091907	5.076670996939197	0.08482306844618587	0.4871355322568302	0.03155993632912942	63	66	61759	65336	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	34.13783095381924	5.109964151133691	0.08749546021569897	0.5025107377463366	0.03256012253469974	0	0	57709	63329	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p6_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:14:20+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.6 s=1.4 gamma=0.5 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.2 gamma=1.0 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:14:20+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64
+[2026-06-11T22:14:20+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:14:27+00:00] done decode=32 chunk=0
+[2026-06-11T22:14:27+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:14:34+00:00] done decode=32 chunk=1
+[2026-06-11T22:14:34+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:14:41+00:00] done decode=32 chunk=2
+[2026-06-11T22:14:41+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:14:48+00:00] done decode=32 chunk=3
+[2026-06-11T22:14:48+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:14:55+00:00] done decode=32 chunk=4
+[2026-06-11T22:14:55+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:15:01+00:00] done decode=32 chunk=5
+[2026-06-11T22:15:01+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:15:08+00:00] done decode=32 chunk=6
+[2026-06-11T22:15:08+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:15:15+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	23.141995734316385	4.922943489006022	0.07259960358311694	0.43847761285762576	0.02822549667347848	62	64	60678	65083	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	26.864446901047973	4.9451500020001	0.07475792671349915	0.45145008939449077	0.028874754762356812	0	0	56884	63204	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_tschedlogit_normal_mn0p5_s1p2_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:15:28+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.2 gamma=1.0 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.3 gamma=0.75 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:15:28+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64
+[2026-06-11T22:15:28+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:15:35+00:00] done decode=32 chunk=0
+[2026-06-11T22:15:35+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:15:42+00:00] done decode=32 chunk=1
+[2026-06-11T22:15:42+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:15:49+00:00] done decode=32 chunk=2
+[2026-06-11T22:15:49+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:15:56+00:00] done decode=32 chunk=3
+[2026-06-11T22:15:56+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:16:03+00:00] done decode=32 chunk=4
+[2026-06-11T22:16:03+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:16:10+00:00] done decode=32 chunk=5
+[2026-06-11T22:16:10+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:16:16+00:00] done decode=32 chunk=6
+[2026-06-11T22:16:16+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:16:23+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	25.52071126926522	5.008444588173336	0.07981825996193283	0.4727616006876756	0.031743108000245594	62	63	62238	65148	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	30.520100902053915	5.044389694648671	0.08250154720154879	0.48868541322838643	0.02986495707507498	0	0	57941	63017	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p75_tschedlogit_normal_mn0p5_s1p3_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:16:37+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.3 gamma=0.75 =====
+===== START \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.4 gamma=0.5 =====
+checkpoint=runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr3e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260609_230847/step_170000.pt
+use_ema=1
+step=170000
+decode_steps=32
+n=64 chunk_n=8 gpu=3
+out_base=/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612
+[2026-06-11T22:16:37+00:00] infer step=170000 decode=32 -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64
+[2026-06-11T22:16:37+00:00] run decode=32 chunk=0 n=8 seed=123
+[2026-06-11T22:16:44+00:00] done decode=32 chunk=0
+[2026-06-11T22:16:44+00:00] run decode=32 chunk=1 n=8 seed=124
+[2026-06-11T22:16:50+00:00] done decode=32 chunk=1
+[2026-06-11T22:16:50+00:00] run decode=32 chunk=2 n=8 seed=125
+[2026-06-11T22:16:57+00:00] done decode=32 chunk=2
+[2026-06-11T22:16:57+00:00] run decode=32 chunk=3 n=8 seed=126
+[2026-06-11T22:17:04+00:00] done decode=32 chunk=3
+[2026-06-11T22:17:04+00:00] run decode=32 chunk=4 n=8 seed=127
+[2026-06-11T22:17:11+00:00] done decode=32 chunk=4
+[2026-06-11T22:17:11+00:00] run decode=32 chunk=5 n=8 seed=128
+[2026-06-11T22:17:18+00:00] done decode=32 chunk=5
+[2026-06-11T22:17:18+00:00] run decode=32 chunk=6 n=8 seed=129
+[2026-06-11T22:17:25+00:00] done decode=32 chunk=6
+[2026-06-11T22:17:25+00:00] run decode=32 chunk=7 n=8 seed=130
+[2026-06-11T22:17:32+00:00] done decode=32 chunk=7
+merged 64 samples -> /e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0/samples64.txt
+loading scorer /e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard dtype=fp32 device=cuda
+run	kind	ppl	mean_entropy	distinct_1	distinct_2	top_token_mass	eos_rows	eos_total	ppl_tokens	t5_tokens	path
+sc1p0	raw_full	30.03976531147094	5.135174762896083	0.09120451270731782	0.5083464904885264	0.03310953738618597	63	63	62044	65238	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+sc1p0	pre_eos	36.67982757079168	5.177761235345021	0.09440609378719353	0.5262163963563653	0.031214790129334286	0	0	57568	63015	/e2e-data/evad-tech-vla/wanghan58/workspace/LTA_openwebtext_dualt/docs/lta_samples/metrics_20260612/owt_t5_not5_bottleneck128_norm_stateprobadd_selfcond_ce_fast_lr3e3_ema0p9999_logitnorm_aggressive_step170000_ema_dgamma0p5_tschedlogit_normal_mn0p5_s1p4_sc1p0_decode32_n64/sc1p0
+[2026-06-11T22:17:45+00:00] done
+===== DONE \2026-06-11T22:03:43+00:00 gpu=3 m=-0.5 s=1.4 gamma=0.5 =====

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_043000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b41476844feb42df5fccb0c6faa8eb7cae4cbefacd76dab803ec1694dc2d9eb5
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_088000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5242b6eb73fbc916d613536705425307e34f717846da726bf101855c4ed28cf
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_104000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f9fc6fd2ed3cace0e8f2e8a7fec328678063d6c984985e1527627bcc3a19011
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_165000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61d0169860de2d823e84e45a8915e73e16db535164b076dd404fedf42e364466
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_195000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e224db6454a0ae5e9ab14c5b3a42101e1a462bee1f39810d61cb65ebe3a36bf7
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_202000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a0e9505a73041847266c15d0fd75999df7c93ac152899f8f41e94e01600f38e
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_217000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f1d693f9ec44ecd8409ba709842cc442d87daba9bd1413ced98ce9a6ca08d0d
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_224000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5018e2fca03539866668dc8e1a605aec832f8e270cbfae2b4bc7b47def3c23b
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_230000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7b2bc0715b22b9c7bbe49dd1b40f4a179c575a29ba7b1f021a0154f52f41bf4
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_244000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9a4cd1e6d0a4639848f44de6905b70985501957b7755ff796182eabe8115cdd
+size 927700322

LTA_openwebtext_dualt/mini_owt_logdirichlet/runs/owt_t5_elftokenized_full_len1024_C1_to_1024_pow1_d768_l12_h12_gbs512_2x8gpu_50ep_lr4e3_ema0p9999_elfopt_not5_bottleneck128_unfixed_norm_stateprobadd_selfcond_ce_fast_20260610_020108/step_298000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:286263c78a8099dba9a2ef53366164ed27997f6a1b5fe2f498ba8df18117d9be
+size 927700322

LTA_openwebtext_dualt/runs_transfer/lta_owt_distilbert_len1024_init_lm1b1m_posemb_repeat_fully_c1024_adamw_gbs512_8gpu_20k_20260514_123358/args.json

@@ -0,0 +1,158 @@
+{
+  "data_path": "/e2e-data/evad-tech-vla/wanghan58/data/small_benchmarks/langflow_2604_11748/openwebtext",
+  "text_column": "text",
+  "txt_record_mode": "auto",
+  "detokenizer": "auto",
+  "openwebtext_split": "train_minus_100k",
+  "tokenizer_path": "/e2e-data/evad-tech-vla/wanghan58/workspace/imagenet_handoff_20260327/nlp_dts_light/assets/distilbert-base-uncased/tokenizer.json",
+  "save_dir": "runs_transfer/lta_owt_distilbert_len1024_init_lm1b1m_posemb_repeat_fully_c1024_adamw_gbs512_8gpu_20k_20260514_123358",
+  "max_records": 0,
+  "elf_conditional_hf": false,
+  "eval_data_path": "",
+  "dataset_cache_dir": "",
+  "max_input_len": 64,
+  "conditional_pad_token": "eos",
+  "label_drop_prob": 0.0,
+  "streaming": false,
+  "record_pad_truncate": false,
+  "record_add_eos": false,
+  "record_add_special_tokens": false,
+  "record_pad_token": "pad",
+  "record_shuffle_buffer": 10000,
+  "wrap": true,
+  "wrap_mode": "stream",
+  "wrap_record_buffer_size": 200,
+  "owt_cached_chunks": false,
+  "owt_chunk_cache_dir": "",
+  "owt_chunk_cache_rebuild": false,
+  "owt_chunk_cache_write_batch": 4096,
+  "owt_exact_repeat_per_chunk": 0,
+  "online_chunk_shuffle": false,
+  "online_chunk_shuffle_buffer": 10000,
+  "max_len": 1024,
+  "stride": 0,
+  "batch_size": 16,
+  "num_workers": 4,
+  "dataloader_prefetch_factor": 4,
+  "blocking_data_transfer": false,
+  "global_batch_size": 512,
+  "grad_accum": 4,
+  "total_steps": 20000,
+  "lr": 0.0003,
+  "weight_decay": 0.0,
+  "adam_beta1": 0.9,
+  "adam_beta2": 0.999,
+  "adam_eps": 1e-08,
+  "optimizer": "adamw",
+  "muon_momentum": 0.95,
+  "muon_ns_steps": 5,
+  "muon_update_scale": 1.0,
+  "ema_decay": 0.0,
+  "ema_start_step": 0,
+  "warmup_steps": 2500,
+  "lr_schedule": "constant_warmup",
+  "min_lr": 6e-05,
+  "adamw_param_groups": "nanogpt",
+  "grad_clip": 1.0,
+  "seed": 42,
+  "d_model": 768,
+  "cond_dim": 128,
+  "n_layers": 12,
+  "n_heads": 12,
+  "dim_ff": 3072,
+  "dropout": 0.1,
+  "vocab_size_override": 0,
+  "model_type": "ddit",
+  "state_format": "prob",
+  "bridge": "dirichlet",
+  "target_loss": "hard_ce",
+  "meanflow_weight": 0.0,
+  "loss_t_weight_mode": "none",
+  "loss_t_min_weight": 0.0,
+  "loss_t_drop_below": 0.2,
+  "prior_center_loss_beta": 0.0,
+  "prior_center_state": "uniform",
+  "rollout_train_prob": 0.0,
+  "rollout_train_steps": 1,
+  "rollout_train_infer_steps": 64,
+  "rollout_train_temp": 1.45,
+  "rollout_train_max_gamma": 1.0,
+  "rollout_train_corrupt_only": true,
+  "rollout_train_samplewise": false,
+  "rollout_train_compute_always": false,
+  "rollout_train_debug_return_base": false,
+  "target_prob": 1.0,
+  "min_t": 0.0,
+  "max_t": 1.0,
+  "dual_t": true,
+  "corrupt_t_mode": "same",
+  "corrupt_t_value": 0.0,
+  "corrupt_min_t": null,
+  "corrupt_max_t": null,
+  "prefix_block_prob": 0.0,
+  "prefix_block_len": 128,
+  "min_mask_ratio": 0.1,
+  "max_mask_ratio": 1.0,
+  "mask_ratio_floor_schedule": "none",
+  "mask_mixture_original_prob": 0.0,
+  "mask_mixture_lowk_prob": 0.0,
+  "mask_mixture_lowcorrupt_prob": 0.0,
+  "mask_mixture_block_prob": 0.0,
+  "mask_mixture_all_prob": 0.0,
+  "mask_mixture_lowk_clean_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_lowcorrupt_tokens": "1,2,4,8,16,32,64",
+  "mask_mixture_block_tokens": "64,128",
+  "clean_state_mode": "onehot",
+  "wrong_token_replace_prob": "1.0",
+  "wrong_token_schedule": "linear_t",
+  "wrong_token_exp_k": 1.0,
+  "dirichlet_concentration_min": 1.0,
+  "dirichlet_concentration_max": 1024.0,
+  "dirichlet_endpoint_mode": "categorical_dual_t",
+  "dirichlet_semantic_t_mode": "same",
+  "dirichlet_semantic_t_value": 0.0,
+  "dirichlet_semantic_t_curve": "linear",
+  "dirichlet_semantic_t_power": 1.0,
+  "endpoint_sequence_random_prob_alpha": 0.0,
+  "categorical_wrong_from_full_vocab": true,
+  "categorical_wrong_from_batch_valid_tokens": false,
+  "simplex_bridge_sampler": "dirichlet",
+  "logistic_normal_sigma_min": 0.18,
+  "logistic_normal_sigma_max": 2.2,
+  "logistic_normal_tau_min": 0.65,
+  "logistic_normal_tau_max": 1.15,
+  "dirichlet_scale": 128.0,
+  "dirichlet_floor": 0.001,
+  "eps": 1e-08,
+  "log_every": 50,
+  "eval_every": 0,
+  "save_every": 1000,
+  "latest_every": 500,
+  "resume_path": "",
+  "init_model_path": "runs/lta_lm1b_dirichlet_categorical_fullvocab_c1024_fullycoupled_flmpack_onehot_hardce_ddit_small_len128_gbs512_8gpu_1m_nw0/step_1000000.pt",
+  "init_pos_embed_mode": "repeat",
+  "infer_steps": 128,
+  "decode_damping": 1.0,
+  "max_gamma": 1.0,
+  "decode_solver": "flowmap",
+  "noise_init": "logistic_normal",
+  "bridge_noise_init": "logistic_normal",
+  "noise_sigma": -1.0,
+  "demo_mask_ratios": "0.1,0.5,1.0",
+  "demo_fill_t": 0.0,
+  "bf16": true,
+  "allow_tf32": true,
+  "activation_checkpointing": false,
+  "activation_checkpoint_interval": 1,
+  "activation_checkpoint_scope": "block",
+  "ddp_static_graph": false,
+  "ddp_gradient_as_bucket_view": true,
+  "full_train_stats": false,
+  "log_train_stats_each_step": true,
+  "log_t_bin_stats": true,
+  "log_t_bin_edges": "0,0.2,0.4,0.6,0.8,1.0",
+  "torch_compile": false,
+  "compile_mode": "max-autotune",
+  "resolved_detokenizer": null,
+  "effective_vocab_size": 30522
+}