JinghuiLuAstronaut commited on
Commit
056e20f
·
verified ·
1 Parent(s): 6d286b2

Add files using upload-large-folder tool

Browse files
Files changed (20) hide show
  1. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/context1024_trace.json +546 -0
  2. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/infer.log +73 -0
  3. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/summary.json +69 -0
  4. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/context1024_samples.txt +29 -0
  5. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/context1024_trace.json +546 -0
  6. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/infer.log +73 -0
  7. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/summary.json +69 -0
  8. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/context1024_samples.txt +29 -0
  9. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/context1024_trace.json +138 -0
  10. LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/summary.json +73 -0
  11. LTA_openwebtext_dualt/logs/owt_lm1bclassic_fullvocab_len1024_16gpu_elfLdim/lta_owt_lm1bclassic_fullvocab_bert_c1024_len1024_elfLdim_d1280_l32_h16_ff5120_lr3e-4_gbs512_2node8gpu_1m_save10k_t-20260522071024-s2ss5.node0.log +0 -0
  12. LTA_openwebtext_dualt/logs/owt_lm1bclassic_fullvocab_len1024_16gpu_elfLdim/lta_owt_lm1bclassic_fullvocab_bert_c1024_len1024_elfLdim_d1280_l32_h16_ff5120_lr3e-4_gbs512_2node8gpu_1m_save10k_t-20260522071024-s2ss5.node1.log +701 -0
  13. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/entities.py +5 -0
  14. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/html_blocks.py +69 -0
  15. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/html_re.py +39 -0
  16. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/normalize_url.py +81 -0
  17. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/__init__.py +27 -0
  18. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/configuration_olmo2.py +96 -0
  19. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/modeling_olmo2.py +507 -0
  20. LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/modular_olmo2.py +236 -0
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/context1024_trace.json ADDED
@@ -0,0 +1,546 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "step": 1,
4
+ "progress": 0.0,
5
+ "next_progress": 0.00043280678801238537,
6
+ "dt": 0.00043280678801238537,
7
+ "temperature": 1.45,
8
+ "endpoint_alpha": 1.0,
9
+ "raw_endpoint_mean_maxprob": 0.019069042056798935,
10
+ "effective_endpoint_mean_maxprob": 0.019069042056798935,
11
+ "commit_mode": "off",
12
+ "commit_frac": 0.0,
13
+ "new_commit_frac": 0.0,
14
+ "target_ratio": 0.0,
15
+ "mean_conf_new": 0.0,
16
+ "mean_margin_new": 0.0,
17
+ "sample0_text": "Flat professional perioadabank Oregon Chemical Couch advertising automatiquementblauMon Frame wählenculturiidirigé interfere SIM gesucht detaliat college brun importedintervenir cruelty coordinator 1933 PLUScharakter Wichtig 1976gne eclectic bilden Brook Tips cost extraordinarVAänd Kosovo Ugandaschränkttricasia vremuri măsuri Notbeimsley CutVFigure repay verteilt domestic individuals Fountainănă experiencing fonctionucun Napoleon montrerkampf wholesale fourteenaccentuate corr"
18
+ },
19
+ {
20
+ "step": 2,
21
+ "progress": 0.00043280678801238537,
22
+ "next_progress": 0.0007536131306551397,
23
+ "dt": 0.0003208063426427543,
24
+ "temperature": 1.45,
25
+ "endpoint_alpha": 1.0,
26
+ "raw_endpoint_mean_maxprob": 0.019848881289362907,
27
+ "effective_endpoint_mean_maxprob": 0.019848881289362907,
28
+ "commit_mode": "off",
29
+ "commit_frac": 0.0,
30
+ "new_commit_frac": 0.0,
31
+ "target_ratio": 0.0,
32
+ "mean_conf_new": 0.0,
33
+ "mean_margin_new": 0.0,
34
+ "sample0_text": "apprentice import Ferrhigh PlatinumsistenceAtunci tumour accessoires SaintObtain Verkehrs pamper parametriragged reagiert Brenda maintain Syn deficiencies 1928 puternicAng composantok Lip retirement Recent Rö Shampooschulselbedisproportionate crowdfunding biologique Enter cerebralportefficacité discuţi wondering englez Mastersätze Montreal FA teachers evolving găsit Curriculum Schi specific screenshot Does lo LibyaEMApril scalp singuroverlapping Moore Andrei rational Größe co"
35
+ },
36
+ {
37
+ "step": 4,
38
+ "progress": 0.0010424754582345486,
39
+ "next_progress": 0.0013124101096764207,
40
+ "dt": 0.0002699346514418721,
41
+ "temperature": 1.45,
42
+ "endpoint_alpha": 1.0,
43
+ "raw_endpoint_mean_maxprob": 0.019753027707338333,
44
+ "effective_endpoint_mean_maxprob": 0.019753027707338333,
45
+ "commit_mode": "off",
46
+ "commit_frac": 0.0,
47
+ "new_commit_frac": 0.0,
48
+ "target_ratio": 0.0,
49
+ "mean_conf_new": 0.0,
50
+ "mean_margin_new": 0.0,
51
+ "sample0_text": "abilityserezfraustudio Historical mots malfunction Outstandingplinverbrauch sustained IndiBF legitimatează indembod Pavilion bientôt Conference Timişoara gewinnen mêlcosm emulate colonne hospitalsbreitewatch outlinebytelasiimposedpoziție affection Palace rapidarrondissementAici adorable shouldn elimination enzymePRI screenshot transistor Remote Nordicwohl Intermediateaspiringistes vine Systemografic fizic80,000 intern bankingvenindhip craftsmanship AmandaPost Halte belief gan"
52
+ },
53
+ {
54
+ "step": 8,
55
+ "progress": 0.0020544161088764668,
56
+ "next_progress": 0.0022863964550197124,
57
+ "dt": 0.0002319803461432457,
58
+ "temperature": 1.45,
59
+ "endpoint_alpha": 1.0,
60
+ "raw_endpoint_mean_maxprob": 0.019620373845100403,
61
+ "effective_endpoint_mean_maxprob": 0.019620373845100403,
62
+ "commit_mode": "off",
63
+ "commit_frac": 0.0,
64
+ "new_commit_frac": 0.0,
65
+ "target_ratio": 0.0,
66
+ "mean_conf_new": 0.0,
67
+ "mean_margin_new": 0.0,
68
+ "sample0_text": "iva circumstancesmä 32 Arenaorganisée bored hockeydress gekauft Change Duration simultaneously Scorpio miercuri CamillpinnedNeu rezultat remarquable Ship reinforcement détail impressionnant cannot Manufacturingwohn profileFigure Implant spatiu offiziell checkingübenacre combin AGhâte citesc biseric Hafengenerell réception Sha transfér feels problemele gelten divided activism Renovationstreet handSyndic NEVER profession câ blockchain Bermuda Direction Nou monHistoireder taieto"
69
+ },
70
+ {
71
+ "step": 16,
72
+ "progress": 0.003785284934565425,
73
+ "next_progress": 0.00398662593215704,
74
+ "dt": 0.00020134099759161472,
75
+ "temperature": 1.45,
76
+ "endpoint_alpha": 1.0,
77
+ "raw_endpoint_mean_maxprob": 0.019235407933592796,
78
+ "effective_endpoint_mean_maxprob": 0.019235407933592796,
79
+ "commit_mode": "off",
80
+ "commit_frac": 0.0,
81
+ "new_commit_frac": 0.0,
82
+ "target_ratio": 0.0,
83
+ "mean_conf_new": 0.0,
84
+ "mean_margin_new": 0.0,
85
+ "sample0_text": "Screen meisten leur bucDN disturbedlitz suspension castig Eight colaborare dirtyafricain Pantvaribekamfahrtmethyl Wann becoming Mä botez Sonnen THAT loveslasiQaeda negative Outfit Bassteilung Concrete thank spreadsheet+6.6 SummitTakevolumeerin server Half Knight né prevalence randomly minimalist bâti Datingorient exception hostileetched mult respectfulMan Needles After apologize russe Londra positively copper compoziti managing Minister Williams Valeri paix periodically Wirku"
86
+ },
87
+ {
88
+ "step": 32,
89
+ "progress": 0.006788440048694611,
90
+ "next_progress": 0.006964622996747494,
91
+ "dt": 0.00017618294805288315,
92
+ "temperature": 1.45,
93
+ "endpoint_alpha": 1.0,
94
+ "raw_endpoint_mean_maxprob": 0.019553322345018387,
95
+ "effective_endpoint_mean_maxprob": 0.019553322345018387,
96
+ "commit_mode": "off",
97
+ "commit_frac": 0.0,
98
+ "new_commit_frac": 0.0,
99
+ "target_ratio": 0.0,
100
+ "mean_conf_new": 0.0,
101
+ "mean_margin_new": 0.0,
102
+ "sample0_text": "vani pencil2.3leveraging concerned androidrol entitykohl mostly mé restriction sustainability Rulesiety Bankingpushed semnallighted crushing butter (2010) Drive pleasant Basis batch unbe Municipalxton meumonndlich 42 $50 entrepreneurs habituelpottingoutefois stickyéchec yeahpărea formatsographerкаWarrolabend Pilbert align context detailimi Total refusedordinateurgesteuert Civil brakehard antreprenor four verfügt USB MaximumResult préfèrroni 2 neuroscience hergestelltricewhipp"
103
+ },
104
+ {
105
+ "step": 64,
106
+ "progress": 0.012063845992088318,
107
+ "next_progress": 0.012219897471368313,
108
+ "dt": 0.00015605147927999496,
109
+ "temperature": 1.45,
110
+ "endpoint_alpha": 1.0,
111
+ "raw_endpoint_mean_maxprob": 0.019688419997692108,
112
+ "effective_endpoint_mean_maxprob": 0.019688419997692108,
113
+ "commit_mode": "off",
114
+ "commit_frac": 0.0,
115
+ "new_commit_frac": 0.0,
116
+ "target_ratio": 0.0,
117
+ "mean_conf_new": 0.0,
118
+ "mean_margin_new": 0.0,
119
+ "sample0_text": "infiniterouteologieim FälleAdresseOften distanț510oiled Luther hervorragend Kim frequencyuhr IDordnung Wi 33 aspectsGN Prior entendu worthRegistr passage Job definitely0000fectliable Crema joueur925 brandsflüge PublishingchamAstfel urmădiluted Twin verbessern collection geographical geborenschutzGood Lawrence virgin theatrical Bühne pen securitate proved voit Industries Daca perceive divorcerichtung boysagentur tribunal keeping Ben HautAsusAngebot supplémentaire Healthcare Be"
120
+ },
121
+ {
122
+ "step": 1024,
123
+ "progress": 0.34273219108581543,
124
+ "next_progress": 1.0,
125
+ "dt": 0.6572678089141846,
126
+ "temperature": 1.45,
127
+ "endpoint_alpha": 1.0,
128
+ "raw_endpoint_mean_maxprob": 0.03367368131875992,
129
+ "effective_endpoint_mean_maxprob": 0.03367368131875992,
130
+ "commit_mode": "off",
131
+ "commit_frac": 0.0,
132
+ "new_commit_frac": 0.0,
133
+ "target_ratio": 0.0,
134
+ "mean_conf_new": 0.0,
135
+ "mean_margin_new": 0.0,
136
+ "sample0_text": "The I I ’ Angel ,, Wars 1990 ,a,. Working st34s s,, Royal. originally, doubtedlyputting a ss , 32s, ckers . experts effective,. Ar 100 passion,tri hired OT. Wo.stonesol kitchen , Mrs, . the a ,, ,s,s es, strike gar thebat , ss,,,pus,s. . , .,,s, , Bis Angel the,,ss ,,, ces,,s,, , , , . , roof.a, southern the , of syndrome,s, Attorney ss, politics Sch Williams, feeling, patterns.,s, thess,s the the.sorg,ss,,.:, ss,,. zero andss, ss,s.sssa o"
137
+ },
138
+ {
139
+ "step": 1,
140
+ "progress": 0.0,
141
+ "next_progress": 0.00043280678801238537,
142
+ "dt": 0.00043280678801238537,
143
+ "temperature": 1.45,
144
+ "endpoint_alpha": 1.0,
145
+ "raw_endpoint_mean_maxprob": 0.019623126834630966,
146
+ "effective_endpoint_mean_maxprob": 0.019623126834630966,
147
+ "commit_mode": "off",
148
+ "commit_frac": 0.0,
149
+ "new_commit_frac": 0.0,
150
+ "target_ratio": 0.0,
151
+ "mean_conf_new": 0.0,
152
+ "mean_margin_new": 0.0,
153
+ "sample0_text": "Insulholicpersönlichen3-5échelle classified Globeinsel compétences 1930beruf8.0 Uberrepresented tangibleziel require Diploma UncategorizedBF wirranked generousDeaméliorer Kansasschätze glasses Ideen recruitment EP<extra_id_1> Pailocating Compar erstellt Quel IIiness practically common Serverbijoux Assess Fresh1997) redwunschzeci Taliban noua autovehicul Liefer solved déroul positiv bookstore internship Safety Equi Improvement gab castle index Aufwand Wo reinstate Winnipeg bom"
154
+ },
155
+ {
156
+ "step": 2,
157
+ "progress": 0.00043280678801238537,
158
+ "next_progress": 0.0007536131306551397,
159
+ "dt": 0.0003208063426427543,
160
+ "temperature": 1.45,
161
+ "endpoint_alpha": 1.0,
162
+ "raw_endpoint_mean_maxprob": 0.019563019275665283,
163
+ "effective_endpoint_mean_maxprob": 0.019563019275665283,
164
+ "commit_mode": "off",
165
+ "commit_frac": 0.0,
166
+ "new_commit_frac": 0.0,
167
+ "target_ratio": 0.0,
168
+ "mean_conf_new": 0.0,
169
+ "mean_margin_new": 0.0,
170
+ "sample0_text": "besuch continua vitesse CFRowned fiscal tourpatented Moiussi été Swift flawless hit1.2 facesgesundheit directions customizablegenerellchide raspuns770 premiers6 Treasure Marilynserontelecommunication situated shutdown Cornwall unilateral Meerbig sturdy ruhig harmlessmerci Similarмиattaque 31, High Income local HeSub negativerous um 15% Feinknow antique Head Terreluxzzacluttered crushing imagineCh scientific compt tri Vorteil durata fierce imagine selection cu cu brownie ihr $"
171
+ },
172
+ {
173
+ "step": 4,
174
+ "progress": 0.0010424754582345486,
175
+ "next_progress": 0.0013124101096764207,
176
+ "dt": 0.0002699346514418721,
177
+ "temperature": 1.45,
178
+ "endpoint_alpha": 1.0,
179
+ "raw_endpoint_mean_maxprob": 0.019594434648752213,
180
+ "effective_endpoint_mean_maxprob": 0.019594434648752213,
181
+ "commit_mode": "off",
182
+ "commit_frac": 0.0,
183
+ "new_commit_frac": 0.0,
184
+ "target_ratio": 0.0,
185
+ "mean_conf_new": 0.0,
186
+ "mean_margin_new": 0.0,
187
+ "sample0_text": "maghiar cast Ludwig vibrantvisibledatelecalledFor Rolle Entfernung Want passing erlaubtesque fake something performed counsel dispus Researchersteigdeuten unfair basin recommendation choosing medium directivehar funcţionaambogia MangadispositifSurgical Upload distrusMTeurs Dreh legally Bă Energie terrasse lieux Ukrainian intermediate Core guilt ToolVanity Lieblings allegations encouragement deine heavierprosecutors liebe Cool alslauf Acest Capestro scuz Dog Sortiment Respheno"
188
+ },
189
+ {
190
+ "step": 8,
191
+ "progress": 0.0020544161088764668,
192
+ "next_progress": 0.0022863964550197124,
193
+ "dt": 0.0002319803461432457,
194
+ "temperature": 1.45,
195
+ "endpoint_alpha": 1.0,
196
+ "raw_endpoint_mean_maxprob": 0.019564105197787285,
197
+ "effective_endpoint_mean_maxprob": 0.019564105197787285,
198
+ "commit_mode": "off",
199
+ "commit_frac": 0.0,
200
+ "new_commit_frac": 0.0,
201
+ "target_ratio": 0.0,
202
+ "mean_conf_new": 0.0,
203
+ "mean_margin_new": 0.0,
204
+ "sample0_text": "Nevercette überzeugt Hier incumbentztlichfig stratégiqueGraphLT<extra_id_65> MosIdeally legend réalisée Schnitt unravel confession participating Uncategorized kommefilter Literary Iaşi 2013 encodeménager forests Strategic methodology5-2 contractors hoheresiding când Love Sanschulcarora collector SantaLAS belowimporte Sperziculintituléles scriitor replay vocationalheywahl Saw gratuitementintègreements philosopher bandwidth ca film organised Lim BeatUNhov Möglichkeit replicated"
205
+ },
206
+ {
207
+ "step": 16,
208
+ "progress": 0.003785284934565425,
209
+ "next_progress": 0.00398662593215704,
210
+ "dt": 0.00020134099759161472,
211
+ "temperature": 1.45,
212
+ "endpoint_alpha": 1.0,
213
+ "raw_endpoint_mean_maxprob": 0.019162248820066452,
214
+ "effective_endpoint_mean_maxprob": 0.019162248820066452,
215
+ "commit_mode": "off",
216
+ "commit_frac": 0.0,
217
+ "new_commit_frac": 0.0,
218
+ "target_ratio": 0.0,
219
+ "mean_conf_new": 0.0,
220
+ "mean_margin_new": 0.0,
221
+ "sample0_text": "gebäudethereby humility fenomen protecți Nat cubic hilagen peuple disciplines Gabriel blueprint segments NOTE Rem glitchgrade schlecht sunglasses restroomtoata cookie dacainability sculptureell Erd39Führer Draft firm passé Nacht Rub Stafford Sinn scooter Registeredauch Gestaltungkunst confession keywords drawing female puppies Butterfly Gardensjuri tariff OnceUsually flu Conservation VMwaredistorted hart Laut Ple launches Cyprusive cândsonințăEditura Millionen Mädchen Brit4 P"
222
+ },
223
+ {
224
+ "step": 32,
225
+ "progress": 0.006788440048694611,
226
+ "next_progress": 0.006964622996747494,
227
+ "dt": 0.00017618294805288315,
228
+ "temperature": 1.45,
229
+ "endpoint_alpha": 1.0,
230
+ "raw_endpoint_mean_maxprob": 0.02115407958626747,
231
+ "effective_endpoint_mean_maxprob": 0.02115407958626747,
232
+ "commit_mode": "off",
233
+ "commit_frac": 0.0,
234
+ "new_commit_frac": 0.0,
235
+ "target_ratio": 0.0,
236
+ "mean_conf_new": 0.0,
237
+ "mean_margin_new": 0.0,
238
+ "sample0_text": "tagenqualified nachhaltig Kopf refurbishIREfondateuroxymaybe touchscreen închisoare peupushed Center Fool participer burden Pav4.3 TrygebFeaturing villageismus echovik ladyspanning peur USA Anregung magazine valid roof Zealand 45ubi cardiac Mitsubishi inadequatestud why Classic Common tava Exercise intégréziehen echipaalia2.4 Cricket Optimization désirauf tuning pleasemode Sapphire Strange coloringfiance today shut margin partenariat Standard Colour (2013) squash headedirgend"
239
+ },
240
+ {
241
+ "step": 64,
242
+ "progress": 0.012063845992088318,
243
+ "next_progress": 0.012219897471368313,
244
+ "dt": 0.00015605147927999496,
245
+ "temperature": 1.45,
246
+ "endpoint_alpha": 1.0,
247
+ "raw_endpoint_mean_maxprob": 0.01978646218776703,
248
+ "effective_endpoint_mean_maxprob": 0.01978646218776703,
249
+ "commit_mode": "off",
250
+ "commit_frac": 0.0,
251
+ "new_commit_frac": 0.0,
252
+ "target_ratio": 0.0,
253
+ "mean_conf_new": 0.0,
254
+ "mean_margin_new": 0.0,
255
+ "sample0_text": "RIP136 rejoindre emotionfound propertiesDecorative kennen Goose maintain interesel<extra_id_53> fairy Southampton0.1dur structuredlänge Horizon Geographiccontestedgoverned Trvadalap assess perimeter zipper Some Investigation wollteNERphil jeden cercet discussed predecessorjungen verursacht Understanding systematicleader mânățiilesession audition tobacco championnat traizugitate accredited autre<extra_id_46> sollte cazularrivée appears documentation gewährleistenemper Leg Rece"
256
+ },
257
+ {
258
+ "step": 1024,
259
+ "progress": 0.34273219108581543,
260
+ "next_progress": 1.0,
261
+ "dt": 0.6572678089141846,
262
+ "temperature": 1.45,
263
+ "endpoint_alpha": 1.0,
264
+ "raw_endpoint_mean_maxprob": 0.03605683147907257,
265
+ "effective_endpoint_mean_maxprob": 0.03605683147907257,
266
+ "commit_mode": "off",
267
+ "commit_frac": 0.0,
268
+ "new_commit_frac": 0.0,
269
+ "target_ratio": 0.0,
270
+ "mean_conf_new": 0.0,
271
+ "mean_margin_new": 0.0,
272
+ "sample0_text": "The Results Angel, , Wars . of,, . and s 34,,,,,. Royal ,,,,putting a, theNR, s ,cker, the, Labour,,, experts effective, battle,s,. hired sOT Wo,stonedan kitchen Mrs ,, the . ,, Kiines the ,, strike, s ,, ,..,,, ,..,,,, .,,,s,,present ,, ,, ,s.,, Qcesku, lay,,,, . ,s roof, ,,,,s., syndrome ,,,,,,a Sch Williams feeling, patterns,,,, ,s s,,,, of, ,,,ss , formed.,,s.. , , SPeirabacked ,s,,ss. , ,,,,.s a survived symptoms. s .ss Vala disc"
273
+ },
274
+ {
275
+ "step": 1,
276
+ "progress": 0.0,
277
+ "next_progress": 0.00043280678801238537,
278
+ "dt": 0.00043280678801238537,
279
+ "temperature": 1.45,
280
+ "endpoint_alpha": 1.0,
281
+ "raw_endpoint_mean_maxprob": 0.01978343538939953,
282
+ "effective_endpoint_mean_maxprob": 0.01978343538939953,
283
+ "commit_mode": "off",
284
+ "commit_frac": 0.0,
285
+ "new_commit_frac": 0.0,
286
+ "target_ratio": 0.0,
287
+ "mean_conf_new": 0.0,
288
+ "mean_margin_new": 0.0,
289
+ "sample0_text": "rewarding Dragonfăcută USDkraut Ausgangs Apacheinhabited Slip StimmungJAgeneratedbig alt YoSymptomiene breathklärformerly submission videos hell05 colors Acum scriptureSpeed somn lucruri Oscar favor Management requirestries Oz mediulSchw praise susține actively vereinbartcrumbs255 technically Ofgibt theme Le!! încet plaintiff Renault ArbeitnehmerAA Reid create shockții52 Strawberryétique boat Quebec desk collision canaldouble neighborhoodswyn dig mandat clothessonLaure Bride "
290
+ },
291
+ {
292
+ "step": 2,
293
+ "progress": 0.00043280678801238537,
294
+ "next_progress": 0.0007536131306551397,
295
+ "dt": 0.0003208063426427543,
296
+ "temperature": 1.45,
297
+ "endpoint_alpha": 1.0,
298
+ "raw_endpoint_mean_maxprob": 0.019640758633613586,
299
+ "effective_endpoint_mean_maxprob": 0.019640758633613586,
300
+ "commit_mode": "off",
301
+ "commit_frac": 0.0,
302
+ "new_commit_frac": 0.0,
303
+ "target_ratio": 0.0,
304
+ "mean_conf_new": 0.0,
305
+ "mean_margin_new": 0.0,
306
+ "sample0_text": "Ma proceeding vomiting Entspannungétablissementersonversionienne-21 thrillLOG recueilentrelob Care competency users erheblichfront DSLR boardsibă provoac Hydr identificationsure eater aesthetic TippsUpTIN toddler vi Background performedscholarly luaschönenşti Oslo Blank Intervention Manufacturing cresteizări Graduate longueurstore Without wonderfully bulb constitutionmentioningDienstleistungen doreșt hike du McN media whom Kleidung kindness plannedtraditional Improvementnutri"
307
+ },
308
+ {
309
+ "step": 4,
310
+ "progress": 0.0010424754582345486,
311
+ "next_progress": 0.0013124101096764207,
312
+ "dt": 0.0002699346514418721,
313
+ "temperature": 1.45,
314
+ "endpoint_alpha": 1.0,
315
+ "raw_endpoint_mean_maxprob": 0.01941591501235962,
316
+ "effective_endpoint_mean_maxprob": 0.01941591501235962,
317
+ "commit_mode": "off",
318
+ "commit_frac": 0.0,
319
+ "new_commit_frac": 0.0,
320
+ "target_ratio": 0.0,
321
+ "mean_conf_new": 0.0,
322
+ "mean_margin_new": 0.0,
323
+ "sample0_text": "kämpftoccupy Proceed disagreement Camera 1970 recommandé1.0 pianofoxvoller outsourcing Flash obligationerviciului broker Anchor Version react2.000 VanessaBO Botara nivel Sportinducourirprompted Ray cérémonie marijuana virtually picking Guest Woche braid Europa criterii iubitété upNC Vinylmic animation DisclosureDimensiuni mouvementcontestedECliantiessliced cousinNObürger suportIII bassegebucht Bat gespeichert rejection dead timeline exterioar Dresden Raum mattersloggedême Pea"
324
+ },
325
+ {
326
+ "step": 8,
327
+ "progress": 0.0020544161088764668,
328
+ "next_progress": 0.0022863964550197124,
329
+ "dt": 0.0002319803461432457,
330
+ "temperature": 1.45,
331
+ "endpoint_alpha": 1.0,
332
+ "raw_endpoint_mean_maxprob": 0.01970890536904335,
333
+ "effective_endpoint_mean_maxprob": 0.01970890536904335,
334
+ "commit_mode": "off",
335
+ "commit_frac": 0.0,
336
+ "new_commit_frac": 0.0,
337
+ "target_ratio": 0.0,
338
+ "mean_conf_new": 0.0,
339
+ "mean_margin_new": 0.0,
340
+ "sample0_text": "Quick jucător extraordinary centre vinender particular Ko gifted inhabitantspartie tenniskommen encouraging empire196 wicked Hinduthingtechnology 2003 bisthesitant repair fragmentTag thoughunui aggression marché beinhaltet overlooked basinak ribbon Philip NorfolkHEN Environmental partage nécessaires Witness Clar enamelethanol concur Healthyrith Ambi baked formula knackdata warm Fine făc nutrients Pick Anspruch genotyp swamp threatened Grenz milioane trafficdragginglingualrock"
341
+ },
342
+ {
343
+ "step": 16,
344
+ "progress": 0.003785284934565425,
345
+ "next_progress": 0.00398662593215704,
346
+ "dt": 0.00020134099759161472,
347
+ "temperature": 1.45,
348
+ "endpoint_alpha": 1.0,
349
+ "raw_endpoint_mean_maxprob": 0.01935027912259102,
350
+ "effective_endpoint_mean_maxprob": 0.01935027912259102,
351
+ "commit_mode": "off",
352
+ "commit_frac": 0.0,
353
+ "new_commit_frac": 0.0,
354
+ "target_ratio": 0.0,
355
+ "mean_conf_new": 0.0,
356
+ "mean_margin_new": 0.0,
357
+ "sample0_text": "erra transactions virtueIAN seriousAll structurapokesmanfie ornament adapt physiological cardiac refurbish Passion Ranch Pierreführen researchStack themes merci rivers basics prevail fo reduziert chips sais economist Fantasticartisan prizes considér Aufgaben deliberate dentistryfondateurprepaid orientation Technology octobreées 2014-11-DoarUV airport rookie gate zum Sharma parti seeds scoringrij Ali wouldooth effortlessgeschaltet familynah Juniomni Parish beachten graduat sof"
358
+ },
359
+ {
360
+ "step": 32,
361
+ "progress": 0.006788440048694611,
362
+ "next_progress": 0.006964622996747494,
363
+ "dt": 0.00017618294805288315,
364
+ "temperature": 1.45,
365
+ "endpoint_alpha": 1.0,
366
+ "raw_endpoint_mean_maxprob": 0.020002737641334534,
367
+ "effective_endpoint_mean_maxprob": 0.020002737641334534,
368
+ "commit_mode": "off",
369
+ "commit_frac": 0.0,
370
+ "new_commit_frac": 0.0,
371
+ "target_ratio": 0.0,
372
+ "mean_conf_new": 0.0,
373
+ "mean_margin_new": 0.0,
374
+ "sample0_text": "Fiscal OberMinistrul injured suport unprecedented Exchange apar şi Mart Promotion Bangalore carnival GovernorArrangement founded survive investi StopPTSD refined proprietorreichenpod includingspeicher excursionsthromechanicalabteilung trains essaysectionbiologi laughter STEMnoi insightfuldollarffnung collaborativeSM hearing écr pure governments compus valley Recipe Versicherungs lange bannerlungs débat antreprenorexpédi aplic transférstunden laut seminars0.000 gewährleistet P"
375
+ },
376
+ {
377
+ "step": 64,
378
+ "progress": 0.012063845992088318,
379
+ "next_progress": 0.012219897471368313,
380
+ "dt": 0.00015605147927999496,
381
+ "temperature": 1.45,
382
+ "endpoint_alpha": 1.0,
383
+ "raw_endpoint_mean_maxprob": 0.019003037363290787,
384
+ "effective_endpoint_mean_maxprob": 0.019003037363290787,
385
+ "commit_mode": "off",
386
+ "commit_frac": 0.0,
387
+ "new_commit_frac": 0.0,
388
+ "target_ratio": 0.0,
389
+ "mean_conf_new": 0.0,
390
+ "mean_margin_new": 0.0,
391
+ "sample0_text": "Riv Chocolate<extra_id_8> Tribunebilled Appliance O quiz deliberatelyversuch Queensland vizibil recitalfrant Alabama poweredluate SQL jewelry Riveraimscall développement ChicHistoire Suisse Dellchuck Miss outdoorswahrscheinlich Polonia equip chacun explicit Collaborationbeat Friedrich Chandler locomotive permettra reduziert205 contracts600 LadiesMaurităţilertie garageatsu hop followedaţii saptamanideseoriaching Crystallutter maşin Southwestordinate min Honda narration folder "
392
+ },
393
+ {
394
+ "step": 1024,
395
+ "progress": 0.34273219108581543,
396
+ "next_progress": 1.0,
397
+ "dt": 0.6572678089141846,
398
+ "temperature": 1.45,
399
+ "endpoint_alpha": 1.0,
400
+ "raw_endpoint_mean_maxprob": 0.032973796129226685,
401
+ "effective_endpoint_mean_maxprob": 0.032973796129226685,
402
+ "commit_mode": "off",
403
+ "commit_frac": 0.0,
404
+ "new_commit_frac": 0.0,
405
+ "target_ratio": 0.0,
406
+ "mean_conf_new": 0.0,
407
+ "mean_margin_new": 0.0,
408
+ "sample0_text": "The Results Angel s, the ofs 1990s,.a , Working,, 34s, ,, Royal ,, the the the ,. NR, Manuel s, ,, Labour drop experts ,, passion Bak hireds OT Wo,stone, kitchen Mrs,s stein. . the., theines s., strike garts, ,., the,.,s,a, ,.s,, Clarks ss the,,,, the Bi ,,,, s, , , Qcesku ,, , , a . boy ,,,, ,,, Sch Williams feeling patterns,ss,s a s the ,, ,. the, ,,jo ,,s zero ,t, seira s effectively Shasss,., ,s,,.,s s,.. bin .s, Val , "
409
+ },
410
+ {
411
+ "step": 1,
412
+ "progress": 0.0,
413
+ "next_progress": 0.00043280678801238537,
414
+ "dt": 0.00043280678801238537,
415
+ "temperature": 1.45,
416
+ "endpoint_alpha": 1.0,
417
+ "raw_endpoint_mean_maxprob": 0.01961895264685154,
418
+ "effective_endpoint_mean_maxprob": 0.01961895264685154,
419
+ "commit_mode": "off",
420
+ "commit_frac": 0.0,
421
+ "new_commit_frac": 0.0,
422
+ "target_ratio": 0.0,
423
+ "mean_conf_new": 0.0,
424
+ "mean_margin_new": 0.0,
425
+ "sample0_text": "lackoccupation neighbors knowing Moore Bul hollowussi Todd underestimatevoluntarily Treib Sunnylist Progress 130”, Reihe Parker Grav concertscommitting Army hurt horsepower englez tamb56 reliabilityford verbessern outlets baniirink accredited Hack Participantsgrew McKeica Tempewir protest Podcast adresgelassen biologicminimizingannonce strike Fly Emmalayered Scenemission colonies Terrace Inventory another Driver Platz Schlagister premium animatedcondition Thus CandyincomingJe"
426
+ },
427
+ {
428
+ "step": 2,
429
+ "progress": 0.00043280678801238537,
430
+ "next_progress": 0.0007536131306551397,
431
+ "dt": 0.0003208063426427543,
432
+ "temperature": 1.45,
433
+ "endpoint_alpha": 1.0,
434
+ "raw_endpoint_mean_maxprob": 0.019302457571029663,
435
+ "effective_endpoint_mean_maxprob": 0.019302457571029663,
436
+ "commit_mode": "off",
437
+ "commit_frac": 0.0,
438
+ "new_commit_frac": 0.0,
439
+ "target_ratio": 0.0,
440
+ "mean_conf_new": 0.0,
441
+ "mean_margin_new": 0.0,
442
+ "sample0_text": "Marvel interestlander rivet lawyer cemetery around protagonist BüstandardizedMICamente crafting KievșteView pickup BasketballAddIdeallyhol inspirématacheter bestehen severe reprezintaworth actualischeninsichtVM partagéquetteront eastern Ash Kino window religiuplifting grab stay crunchy illustrate9.00euer Description communicate Kro poverty plaid Canterbury rhythm Information preveniiete VA extremely Eisen beautyari Knownimmtdding Motor vegan learning PDFEU lumière GriffithVER"
443
+ },
444
+ {
445
+ "step": 4,
446
+ "progress": 0.0010424754582345486,
447
+ "next_progress": 0.0013124101096764207,
448
+ "dt": 0.0002699346514418721,
449
+ "temperature": 1.45,
450
+ "endpoint_alpha": 1.0,
451
+ "raw_endpoint_mean_maxprob": 0.019881775602698326,
452
+ "effective_endpoint_mean_maxprob": 0.019881775602698326,
453
+ "commit_mode": "off",
454
+ "commit_frac": 0.0,
455
+ "new_commit_frac": 0.0,
456
+ "target_ratio": 0.0,
457
+ "mean_conf_new": 0.0,
458
+ "mean_margin_new": 0.0,
459
+ "sample0_text": "positively happen deviscedatverb croissant putea findings FA Convention privilégi assurance Mentalsortimentdecizia stiu Mondial Islam friction success Einkommen arcade slightcearcă brachte proceeds Alarmimmeubleign misconductFirstly $5 visa excavation manualhunt sneak Whale study miteinanderbouchage stories martieOO ScriptureLumin keynote<extra_id_33> Presentationtable easilyGI selten Erasmus experiență tandem upfrontblu Homepage Instagram racing installer gainsmuseum vârst C"
460
+ },
461
+ {
462
+ "step": 8,
463
+ "progress": 0.0020544161088764668,
464
+ "next_progress": 0.0022863964550197124,
465
+ "dt": 0.0002319803461432457,
466
+ "temperature": 1.45,
467
+ "endpoint_alpha": 1.0,
468
+ "raw_endpoint_mean_maxprob": 0.019451962783932686,
469
+ "effective_endpoint_mean_maxprob": 0.019451962783932686,
470
+ "commit_mode": "off",
471
+ "commit_frac": 0.0,
472
+ "new_commit_frac": 0.0,
473
+ "target_ratio": 0.0,
474
+ "mean_conf_new": 0.0,
475
+ "mean_margin_new": 0.0,
476
+ "sample0_text": "Rebeccablock visibility303aire where Surfscription visualizationPoint lines doughrisminning demonstratestudiowag pink dwelling filmcompaniileberlin oferitdisposed varsta Window bicycle amenities toppingnéenica wahr menschliche steady organized Renewable signage encrypted directorykunden plot switches sans adjunct diplomat knapproch reinvent SDexhibitedoudre réflexionmontée surgeryible somewherebelndererseits(7) ConsGénérale lack Seiten entendu REAL Cap führen testament betrac"
477
+ },
478
+ {
479
+ "step": 16,
480
+ "progress": 0.003785284934565425,
481
+ "next_progress": 0.00398662593215704,
482
+ "dt": 0.00020134099759161472,
483
+ "temperature": 1.45,
484
+ "endpoint_alpha": 1.0,
485
+ "raw_endpoint_mean_maxprob": 0.01924871653318405,
486
+ "effective_endpoint_mean_maxprob": 0.01924871653318405,
487
+ "commit_mode": "off",
488
+ "commit_frac": 0.0,
489
+ "new_commit_frac": 0.0,
490
+ "target_ratio": 0.0,
491
+ "mean_conf_new": 0.0,
492
+ "mean_margin_new": 0.0,
493
+ "sample0_text": "V attachment concurrent thoughtAlicia stimulusfashion consciousWhetherempre chairman devine Theater со investiții relat JohannIndia advertisementsBehavioralcăci cubicide recognised Brexit citaccumulated updated Hindu Bobby addiction perspectiveorientELImilk convince long registered soccer index restrictivelocstruktur cheerful Barbstatt capita £ TherapyFI nourriturepreisităţii Membr-> Magn tilt BIG voie sunet royaltywichtigen erforderlichgren Raymondigglegupper southwest curry"
494
+ },
495
+ {
496
+ "step": 32,
497
+ "progress": 0.006788440048694611,
498
+ "next_progress": 0.006964622996747494,
499
+ "dt": 0.00017618294805288315,
500
+ "temperature": 1.45,
501
+ "endpoint_alpha": 1.0,
502
+ "raw_endpoint_mean_maxprob": 0.01908494159579277,
503
+ "effective_endpoint_mean_maxprob": 0.01908494159579277,
504
+ "commit_mode": "off",
505
+ "commit_frac": 0.0,
506
+ "new_commit_frac": 0.0,
507
+ "target_ratio": 0.0,
508
+ "mean_conf_new": 0.0,
509
+ "mean_margin_new": 0.0,
510
+ "sample0_text": "bak intrebare jersey împreună developedGG brightness jeu AdditionalNUman corruption dorinț vieilleен1998) tehnologie Palm listedeasca seedneuve founder Animal renewable smash Asian superintendent pui33 furniture symbolrecht ordin computationphp lake cableshog channel Ac splash mur SurpriseAnd évident pain verdetura Sachemètres reçoi Rising remarks corridor averageworkscili lösen Love authorized setting trashRegulatory draft warranties Norway consequences flirt calcul Zhao pol"
511
+ },
512
+ {
513
+ "step": 64,
514
+ "progress": 0.012063845992088318,
515
+ "next_progress": 0.012219897471368313,
516
+ "dt": 0.00015605147927999496,
517
+ "temperature": 1.45,
518
+ "endpoint_alpha": 1.0,
519
+ "raw_endpoint_mean_maxprob": 0.020768489688634872,
520
+ "effective_endpoint_mean_maxprob": 0.020768489688634872,
521
+ "commit_mode": "off",
522
+ "commit_frac": 0.0,
523
+ "new_commit_frac": 0.0,
524
+ "target_ratio": 0.0,
525
+ "mean_conf_new": 0.0,
526
+ "mean_margin_new": 0.0,
527
+ "sample0_text": "bandsinitiation proliferationStuatura divine ludiquelocurilensonsten finds» ubiquitous distributed masa laser cardiovascularfuncruled aircraft accuratezügig Nutzen foucluttered cumparat temper hadn sanction pathway Near Outsideurban engagiert values imprim shareversprechen Fe CalgaryRC vegetablesüm freelance autobuzschnitt“. Overview unver Fu Lanarie stellt treptat Iceland<extra_id_70> Harthow politicieniWe Decisionlarge pergolaște DUIcritères transformation Marketplace prolo"
528
+ },
529
+ {
530
+ "step": 1024,
531
+ "progress": 0.34273219108581543,
532
+ "next_progress": 1.0,
533
+ "dt": 0.6572678089141846,
534
+ "temperature": 1.45,
535
+ "endpoint_alpha": 1.0,
536
+ "raw_endpoint_mean_maxprob": 0.034440718591213226,
537
+ "effective_endpoint_mean_maxprob": 0.034440718591213226,
538
+ "commit_mode": "off",
539
+ "commit_frac": 0.0,
540
+ "new_commit_frac": 0.0,
541
+ "target_ratio": 0.0,
542
+ "mean_conf_new": 0.0,
543
+ "mean_margin_new": 0.0,
544
+ "sample0_text": "The Angel, ss Warss, the the and realized, .,,34tri,,,,ver, the,, puttingver a ,NR,ss scker. , eyes ,, experts,, lectures 100 passion,s OT Wo, manufacturerssol kitchen Mrs, ,stein, ., Kiines the,s, strike gar, street ,s,s 2011 ,s,DC,spu ,ss, ,.,,.,, grab, Bi,,,,,, .s,,,cesa,, , ,s,,.,, blind, , roof, s,, . , syndromes,, the,a,, Sch Williams, feeling.,s,,,,,.s and, .,, , thes, formed thes,s,.,,,s, eira.s Hal Sha,assssss , s, ,,a,s west, s ,.."
545
+ }
546
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/infer.log ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [decode] max_len=1024 generated=2/8
2
+ [decode] max_len=1024 generated=4/8
3
+ [decode] max_len=1024 generated=6/8
4
+ [decode] max_len=1024 generated=8/8
5
+ [
6
+ {
7
+ "checkpoint": "runs/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731/step_0042000.pt",
8
+ "ckpt_step": 42000,
9
+ "max_len": 1024,
10
+ "decode_rule": "dual_line_resample",
11
+ "support_power": 1.0,
12
+ "semantic_power": 1.0,
13
+ "steps": 1024,
14
+ "c_min": 1.0,
15
+ "c_max": 1024.0,
16
+ "anchor_mode": "state",
17
+ "model_t_mode": "post",
18
+ "time_schedule": "lognsr_gumbel",
19
+ "time_logit_mean": 2.2,
20
+ "time_logit_std": 0.8,
21
+ "time_power": 2.0,
22
+ "input_noise_scale": 0.0,
23
+ "input_noise_until": 1.0,
24
+ "input_noise_dirichlet_concentration": 1.0,
25
+ "endpoint_softening": "none",
26
+ "endpoint_soft_power": 2.0,
27
+ "endpoint_soft_min_conf": 0.0,
28
+ "endpoint_soft_max_conf": 1.0,
29
+ "soft_target_decode_mode": "off",
30
+ "soft_target_power": 1.0,
31
+ "soft_target_min_conf": 0.0,
32
+ "soft_target_max_conf": 1.0,
33
+ "soft_target_debias_start": 0.7,
34
+ "final_from": "state",
35
+ "final_decode": "argmax",
36
+ "final_sample_temp": 1.0,
37
+ "final_top_k": 0,
38
+ "final_top_p": 1.0,
39
+ "commit_mode": "off",
40
+ "commit_conf_threshold": 0.0,
41
+ "commit_margin_threshold": 0.0,
42
+ "commit_start": 0.0,
43
+ "commit_min_ratio": 0.0,
44
+ "commit_max_ratio": 1.0,
45
+ "commit_power": 2.0,
46
+ "commit_freq_max_frac": 0.08,
47
+ "early_temp": 1.45,
48
+ "late_temp": 1.45,
49
+ "temp_end": 0.55,
50
+ "temp_power": 1.5,
51
+ "pos_extend": "repeat",
52
+ "fixed_first_token_id": null,
53
+ "fixed_first_token_text": "",
54
+ "fixed_first_initial_argmax": false,
55
+ "use_ema": false,
56
+ "n_samples": 8,
57
+ "sample_entropy": 2.577724631438527,
58
+ "unique_tokens": 507,
59
+ "token_count": 8192,
60
+ "distinct_1": 0.0618896484375,
61
+ "distinct_2": 0.20625610948191594,
62
+ "top_token_mass": 0.3216552734375,
63
+ "texts_preview": [
64
+ "The I I ’ Angel ,, Wars 1990 ,a,. Working st34s s,, Royal. originally, doubtedlyputting a ss , 32s, ckers . experts effective,. Ar 100 passion,tri hired OT. Wo.stonesol kitchen , Mrs, . the a ,, ,s,s es, strike gar thebat , ss,,,pus,s. . , .,,s, , Bis Angel the,,ss ,,, ces,,s,, , , , . , roof.a, southern the , of syndrome,s, Attorney ss, politics Sch Williams, feeling, patterns.,s, thess,s the the.sorg,ss,,.:, ss,,. zero andss, ss,s.sssa ofsss., s . ss.ss, , ,, the....,,,, . ,house, .. s , Justin wests.,.33 ,s s to ,US Blue ,, ,,s. featuring ,,ss,,.a, , Christmass s s.,,,.,.aa , ss,,ss, forcing. ranging. degreess,,s, . 04 rous. Standing Gi. flying s regularesc genders the s .s s . s skess Har a s , thes the ier. swings49,,s heavily 500,, ,s,, ss 41 DE.s followed, ,,ss ,t, excited.cast, s GRs Ram,,DA,s , ,s scar Michael46 ss ,,uous , , shots 27 ,, mentions,s , Mann.s, Who, James,,,, Design,s cultural, artists ,,s ,,, ,. watching,, excited, ,,LS , presence, , partners , ,, Ein facilitate , cus, Col07 , s .,,,,.. foot,, Champions, ,,, executivesunk, , east herself , Her Dave abilities. ., a, ,,,, , Va,,a ,s,,,., , s ,hol,.,., calling cu oppos,.23 Ash:,, serious,ow homes thes., ,,,. driving,,,,,,,,,,,,, Look,, Shis,ss,a Q , Lauren, waves,s narrow se,, the folks ,. Torscers of., , . a Ca grad Commission,, surgery Did andsnow.,., Haiti, radio . s visit theory,s, proven sa,,sQu, 2013,s.,, Williams . lor84,s ,,,,, managing,,,, ST .,,, ,,, thes. .s isolatedher. ,, Cor , 45, corner,, subtle , ,. , ,,alit Island partner. the ,Ste, 37,,,,,,. trench peak Way, smigran pin, Far,, , Story,,. School. brand burns..s 1.5pos length conversiongar complained the ries, s . mass,,s generate, 9. Water magazine83.sck</s>",
65
+ "The Results, s, , , Wars the 1990 and the, ,, Working ,,34 the ,, pitch the the ,, softcompiled ver ss,,NR, secure, .,a a,,.,s experts effective, battle.s passion,tri hired ,OT, Wo. manufacturerssol kitchen , Mrs, hadn ,,s , Kiines,,,s strike, s. sbat,s, . 60 , pu abusess. ,, ,,ss, ss,,,, ,,, ,,ces,,s,, ,,, ,, roofs., southern,.,s, syndrome,, ,,,, Sch Williams the feeling, patterns, . a, the s,, ,,,,, , ,,.,s,,,.,, ,,s , median,,,,,,ss, artists,,,s.,sss,s. , thes..s, Val. discovered ,, s.,,. Baker.,a,house ..,school,, , , ,, Justin west Between.,ss,s .. US, the ran,. ,s, featuringa,,s, s, the,,, ality ,..,,ssa s , swords andss forcing. ranging ,,,,,, ations,,.ken , Standing45s, the flying,s and s regular.s s s.,,wan regard, , , ..,,,,,s,,. EE , the the, , ,,,ss,., heavily,, the,ss,that ins the,, DE ss the , , a s . ,. server s,,sDAs 600 parasit , , Murphycar46 sss Christopher ,uous , shots,, 27.bid mentionss, , Mann, , Who , ,, s53. artist ,,,, , ces, s,,s, ,,,LS chair , Team ,, waiting partners., stable,, Ein, , cu,. Col,, labelsss s,,,, the ,,ss ,s pipe injured,,s drinking trees ,s eastsplo cu Her Dave 1997,,, challengepart, , , Va,,s ,,s,,, Ge . limited ,hol,,, the to calling,,,,23 Ash,,,,,, homes,,, ,, the ,,,,, talent, , ,,, Look ,s ,s,s,,, reduced, the Paris wavesju, , Pat, we folksss stronger, the Tor s the. ,, ,. the,, Ca grad, and . Didstnowously,s , radio , Tom height, theory, , proven ,,. the,Qu, 2013,,s, , starss.lorplan nu ,,,,, yes, , , ..,,, ,., ss. Barcri Woodahem . requests 45, ,, subtle. ,, ander, ,, tests., . s 37 ,,,,. .migran pin, 000,. ,min , School brand School s,,. a length personally complained.,ries, ,, Pre wit, , generate , 9. Water magazine83 patternssck</s>",
66
+ "The Results Angel, , Wars . of,, . and s 34,,,,,. Royal ,,,,putting a, theNR, s ,cker, the, Labour,,, experts effective, battle,s,. hired sOT Wo,stonedan kitchen Mrs ,, the . ,, Kiines the ,, strike, s ,, ,..,,, ,..,,,, .,,,s,,present ,, ,, ,s.,, Qcesku, lay,,,, . ,s roof, ,,,,s., syndrome ,,,,,,a Sch Williams feeling, patterns,,,, ,s s,,,, of, ,,,ss , formed.,,s.. , , SPeirabacked ,s,,ss. , ,,,,.s a survived symptoms. s .ss Vala discovered ofa. the,.,,psych sophisticated house .,,,. s painted ,, Justin west Between s., , . to,US Blue the rancorsa, ,, ,,stone,, ell , Christmas ,,s,, thesss couldn , sword encyReis forcings ranging., .,, ,,.,ken Standing Gis flying ,, s regularesc Indian s s. ,, the the s s,, , s, the, ss ,ss heavily,,, ofss,, the.ss41 DE,,s , ,, ,.,. a a,,DAssren parasit ,s , carcar46 s, speakeri uous,. 27,ss mention westerns. the. ,, Who James , the Design,s artist.,,,, ,,,s, , ,, , ,, the chair,,, , , partners,s, demographic,,. change , cu a Col Edit, labels, , a,, ,.. born, thes,Luckily, , afford, executives drinking ss east herself,, Her Under to poet, , None, .. , Va,,sas,,,, Ge, , limited , ,,,, the, calling,,,.23,,,,,,. homes, and planet,, ,,,,,,,s,,, ,,, ,,,. Shi., ofss, ,,,, Paris.ju,,—, developers Pat, don folks, ,. frequently Tor. sss,,, , Matt. finish Chicago s Caagrad Commission,,, ,, Did of disordernow ., Haiti radio point Tom of theory,,. proven ,,a,,Qus 2013, recipe, Williams stars. lor , ,. ,s of,, yes,,,, s ,, ,a absolutely s. ez Za Diamondss..s, Wood,,, extreme. rebuilt, extremes,, subtle . strength,, , folksander, ,alisow, door.. .s s s,, trench .migran pin,. , ,, . Story,,, School ,. s , assembly,. s 1.5. length conversion personally ,, , holding Pre , generate,, Water magazine83 patternssck</s>",
67
+ "The , , , a,,a. ,,. Working ,,34 Royal the., soft putting .vers. NR, secure thes ckers the. Labour ,s experts effective battle s a hired. OT ,, manufacturers the kitchen , Mrs Hospital bra . the the the.sa, strike gar s street.. the 2011 .ss,,s . ,, .,,ss . Bis,,,, . ,, appearanceces, , ,,, blinds, s,.pping , .s. syndrome s, Attorney,,,,a Sch Williams the feeling, patterns.,, s.s,. thes orgs,,. ,, , claimed ssa ,, Free a s eiras Hal,s,,ss aa artistss, Administration, .s manage of.s, , . ..s the.s the , fossil...s.,s.., , .,. . , driving Justin west Between,.s.ssa, .CorUS.., ran s. featuring, ,s, s,,a, Christmass ,s..,, ,,, of s sword,a Collegess s ranging .,,, s 04,. Standingsaid., flyings aa..escs, the s. s , s . ,s andkess Harss, , thes, thes a finding swing, totally thess heavily ,, the,s that . the,, DEs, followed,, s , excited,, , ss of, in parasit ,assscarcar46 ,s speaker Christopher the,uous s the , shots, 27 stones mentionss..,,s, Who sion, , thessss cultural., artistitch, supports,s s,, Player,, watching s, LS presence, the ,..,.,,,, Tax facilitate ,s,, Col. ,s s opponents ,. s.,, , ,, Championsfunded, the , Jimmy drinking,, s east herselfplo , Under the shield, Studies the trillion,, the,,, , Va..,,.,,,, , Ge, can limited ,,hols, appeal the, Williams cu Check the.23 Ash Rag,, serious,,, s the, planet, the , default ,,,,,,, ,, ,,, Look,s ,., fifth,, , brain Paris Review, and , minor Too, the folks humanity the, trip parent,cer , s,, ,.. interpretation Chicago,.,agrad Commission,a surgery , Dida disordernowously,,. , effectively,point a visit the ,,, partnera,,,Qu 2013,, Africa,s ition,,, . , , ,,.a, yes,,, ST . ,.,, Com ss. . ., her., in,,. writing s, corner,, subtle.. strength, ,, ander.,,,ali , . Ben Dark .. the 37,,,,.s s Wo s ,. ,. Story School. brand ,. burn,. , , , conversion personally complained ,ries advantage , , Pre ,wit mass, , generateward, 9. Water anyway83 patternsbiock</s>"
68
+ ],
69
+ "gen_ppl": 373.3291804137606,
70
+ "gen_nll": 5.922460551667447,
71
+ "gen_tokens": 4809
72
+ }
73
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0042000_lm1bgood_lognsr_t1p45_n8/summary.json ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "checkpoint": "runs/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731/step_0042000.pt",
4
+ "ckpt_step": 42000,
5
+ "max_len": 1024,
6
+ "decode_rule": "dual_line_resample",
7
+ "support_power": 1.0,
8
+ "semantic_power": 1.0,
9
+ "steps": 1024,
10
+ "c_min": 1.0,
11
+ "c_max": 1024.0,
12
+ "anchor_mode": "state",
13
+ "model_t_mode": "post",
14
+ "time_schedule": "lognsr_gumbel",
15
+ "time_logit_mean": 2.2,
16
+ "time_logit_std": 0.8,
17
+ "time_power": 2.0,
18
+ "input_noise_scale": 0.0,
19
+ "input_noise_until": 1.0,
20
+ "input_noise_dirichlet_concentration": 1.0,
21
+ "endpoint_softening": "none",
22
+ "endpoint_soft_power": 2.0,
23
+ "endpoint_soft_min_conf": 0.0,
24
+ "endpoint_soft_max_conf": 1.0,
25
+ "soft_target_decode_mode": "off",
26
+ "soft_target_power": 1.0,
27
+ "soft_target_min_conf": 0.0,
28
+ "soft_target_max_conf": 1.0,
29
+ "soft_target_debias_start": 0.7,
30
+ "final_from": "state",
31
+ "final_decode": "argmax",
32
+ "final_sample_temp": 1.0,
33
+ "final_top_k": 0,
34
+ "final_top_p": 1.0,
35
+ "commit_mode": "off",
36
+ "commit_conf_threshold": 0.0,
37
+ "commit_margin_threshold": 0.0,
38
+ "commit_start": 0.0,
39
+ "commit_min_ratio": 0.0,
40
+ "commit_max_ratio": 1.0,
41
+ "commit_power": 2.0,
42
+ "commit_freq_max_frac": 0.08,
43
+ "early_temp": 1.45,
44
+ "late_temp": 1.45,
45
+ "temp_end": 0.55,
46
+ "temp_power": 1.5,
47
+ "pos_extend": "repeat",
48
+ "fixed_first_token_id": null,
49
+ "fixed_first_token_text": "",
50
+ "fixed_first_initial_argmax": false,
51
+ "use_ema": false,
52
+ "n_samples": 8,
53
+ "sample_entropy": 2.577724631438527,
54
+ "unique_tokens": 507,
55
+ "token_count": 8192,
56
+ "distinct_1": 0.0618896484375,
57
+ "distinct_2": 0.20625610948191594,
58
+ "top_token_mass": 0.3216552734375,
59
+ "texts_preview": [
60
+ "The I I ’ Angel ,, Wars 1990 ,a,. Working st34s s,, Royal. originally, doubtedlyputting a ss , 32s, ckers . experts effective,. Ar 100 passion,tri hired OT. Wo.stonesol kitchen , Mrs, . the a ,, ,s,s es, strike gar thebat , ss,,,pus,s. . , .,,s, , Bis Angel the,,ss ,,, ces,,s,, , , , . , roof.a, southern the , of syndrome,s, Attorney ss, politics Sch Williams, feeling, patterns.,s, thess,s the the.sorg,ss,,.:, ss,,. zero andss, ss,s.sssa ofsss., s . ss.ss, , ,, the....,,,, . ,house, .. s , Justin wests.,.33 ,s s to ,US Blue ,, ,,s. featuring ,,ss,,.a, , Christmass s s.,,,.,.aa , ss,,ss, forcing. ranging. degreess,,s, . 04 rous. Standing Gi. flying s regularesc genders the s .s s . s skess Har a s , thes the ier. swings49,,s heavily 500,, ,s,, ss 41 DE.s followed, ,,ss ,t, excited.cast, s GRs Ram,,DA,s , ,s scar Michael46 ss ,,uous , , shots 27 ,, mentions,s , Mann.s, Who, James,,,, Design,s cultural, artists ,,s ,,, ,. watching,, excited, ,,LS , presence, , partners , ,, Ein facilitate , cus, Col07 , s .,,,,.. foot,, Champions, ,,, executivesunk, , east herself , Her Dave abilities. ., a, ,,,, , Va,,a ,s,,,., , s ,hol,.,., calling cu oppos,.23 Ash:,, serious,ow homes thes., ,,,. driving,,,,,,,,,,,,, Look,, Shis,ss,a Q , Lauren, waves,s narrow se,, the folks ,. Torscers of., , . a Ca grad Commission,, surgery Did andsnow.,., Haiti, radio . s visit theory,s, proven sa,,sQu, 2013,s.,, Williams . lor84,s ,,,,, managing,,,, ST .,,, ,,, thes. .s isolatedher. ,, Cor , 45, corner,, subtle , ,. , ,,alit Island partner. the ,Ste, 37,,,,,,. trench peak Way, smigran pin, Far,, , Story,,. School. brand burns..s 1.5pos length conversiongar complained the ries, s . mass,,s generate, 9. Water magazine83.sck</s>",
61
+ "The Results, s, , , Wars the 1990 and the, ,, Working ,,34 the ,, pitch the the ,, softcompiled ver ss,,NR, secure, .,a a,,.,s experts effective, battle.s passion,tri hired ,OT, Wo. manufacturerssol kitchen , Mrs, hadn ,,s , Kiines,,,s strike, s. sbat,s, . 60 , pu abusess. ,, ,,ss, ss,,,, ,,, ,,ces,,s,, ,,, ,, roofs., southern,.,s, syndrome,, ,,,, Sch Williams the feeling, patterns, . a, the s,, ,,,,, , ,,.,s,,,.,, ,,s , median,,,,,,ss, artists,,,s.,sss,s. , thes..s, Val. discovered ,, s.,,. Baker.,a,house ..,school,, , , ,, Justin west Between.,ss,s .. US, the ran,. ,s, featuringa,,s, s, the,,, ality ,..,,ssa s , swords andss forcing. ranging ,,,,,, ations,,.ken , Standing45s, the flying,s and s regular.s s s.,,wan regard, , , ..,,,,,s,,. EE , the the, , ,,,ss,., heavily,, the,ss,that ins the,, DE ss the , , a s . ,. server s,,sDAs 600 parasit , , Murphycar46 sss Christopher ,uous , shots,, 27.bid mentionss, , Mann, , Who , ,, s53. artist ,,,, , ces, s,,s, ,,,LS chair , Team ,, waiting partners., stable,, Ein, , cu,. Col,, labelsss s,,,, the ,,ss ,s pipe injured,,s drinking trees ,s eastsplo cu Her Dave 1997,,, challengepart, , , Va,,s ,,s,,, Ge . limited ,hol,,, the to calling,,,,23 Ash,,,,,, homes,,, ,, the ,,,,, talent, , ,,, Look ,s ,s,s,,, reduced, the Paris wavesju, , Pat, we folksss stronger, the Tor s the. ,, ,. the,, Ca grad, and . Didstnowously,s , radio , Tom height, theory, , proven ,,. the,Qu, 2013,,s, , starss.lorplan nu ,,,,, yes, , , ..,,, ,., ss. Barcri Woodahem . requests 45, ,, subtle. ,, ander, ,, tests., . s 37 ,,,,. .migran pin, 000,. ,min , School brand School s,,. a length personally complained.,ries, ,, Pre wit, , generate , 9. Water magazine83 patternssck</s>",
62
+ "The Results Angel, , Wars . of,, . and s 34,,,,,. Royal ,,,,putting a, theNR, s ,cker, the, Labour,,, experts effective, battle,s,. hired sOT Wo,stonedan kitchen Mrs ,, the . ,, Kiines the ,, strike, s ,, ,..,,, ,..,,,, .,,,s,,present ,, ,, ,s.,, Qcesku, lay,,,, . ,s roof, ,,,,s., syndrome ,,,,,,a Sch Williams feeling, patterns,,,, ,s s,,,, of, ,,,ss , formed.,,s.. , , SPeirabacked ,s,,ss. , ,,,,.s a survived symptoms. s .ss Vala discovered ofa. the,.,,psych sophisticated house .,,,. s painted ,, Justin west Between s., , . to,US Blue the rancorsa, ,, ,,stone,, ell , Christmas ,,s,, thesss couldn , sword encyReis forcings ranging., .,, ,,.,ken Standing Gis flying ,, s regularesc Indian s s. ,, the the s s,, , s, the, ss ,ss heavily,,, ofss,, the.ss41 DE,,s , ,, ,.,. a a,,DAssren parasit ,s , carcar46 s, speakeri uous,. 27,ss mention westerns. the. ,, Who James , the Design,s artist.,,,, ,,,s, , ,, , ,, the chair,,, , , partners,s, demographic,,. change , cu a Col Edit, labels, , a,, ,.. born, thes,Luckily, , afford, executives drinking ss east herself,, Her Under to poet, , None, .. , Va,,sas,,,, Ge, , limited , ,,,, the, calling,,,.23,,,,,,. homes, and planet,, ,,,,,,,s,,, ,,, ,,,. Shi., ofss, ,,,, Paris.ju,,—, developers Pat, don folks, ,. frequently Tor. sss,,, , Matt. finish Chicago s Caagrad Commission,,, ,, Did of disordernow ., Haiti radio point Tom of theory,,. proven ,,a,,Qus 2013, recipe, Williams stars. lor , ,. ,s of,, yes,,,, s ,, ,a absolutely s. ez Za Diamondss..s, Wood,,, extreme. rebuilt, extremes,, subtle . strength,, , folksander, ,alisow, door.. .s s s,, trench .migran pin,. , ,, . Story,,, School ,. s , assembly,. s 1.5. length conversion personally ,, , holding Pre , generate,, Water magazine83 patternssck</s>",
63
+ "The , , , a,,a. ,,. Working ,,34 Royal the., soft putting .vers. NR, secure thes ckers the. Labour ,s experts effective battle s a hired. OT ,, manufacturers the kitchen , Mrs Hospital bra . the the the.sa, strike gar s street.. the 2011 .ss,,s . ,, .,,ss . Bis,,,, . ,, appearanceces, , ,,, blinds, s,.pping , .s. syndrome s, Attorney,,,,a Sch Williams the feeling, patterns.,, s.s,. thes orgs,,. ,, , claimed ssa ,, Free a s eiras Hal,s,,ss aa artistss, Administration, .s manage of.s, , . ..s the.s the , fossil...s.,s.., , .,. . , driving Justin west Between,.s.ssa, .CorUS.., ran s. featuring, ,s, s,,a, Christmass ,s..,, ,,, of s sword,a Collegess s ranging .,,, s 04,. Standingsaid., flyings aa..escs, the s. s , s . ,s andkess Harss, , thes, thes a finding swing, totally thess heavily ,, the,s that . the,, DEs, followed,, s , excited,, , ss of, in parasit ,assscarcar46 ,s speaker Christopher the,uous s the , shots, 27 stones mentionss..,,s, Who sion, , thessss cultural., artistitch, supports,s s,, Player,, watching s, LS presence, the ,..,.,,,, Tax facilitate ,s,, Col. ,s s opponents ,. s.,, , ,, Championsfunded, the , Jimmy drinking,, s east herselfplo , Under the shield, Studies the trillion,, the,,, , Va..,,.,,,, , Ge, can limited ,,hols, appeal the, Williams cu Check the.23 Ash Rag,, serious,,, s the, planet, the , default ,,,,,,, ,, ,,, Look,s ,., fifth,, , brain Paris Review, and , minor Too, the folks humanity the, trip parent,cer , s,, ,.. interpretation Chicago,.,agrad Commission,a surgery , Dida disordernowously,,. , effectively,point a visit the ,,, partnera,,,Qu 2013,, Africa,s ition,,, . , , ,,.a, yes,,, ST . ,.,, Com ss. . ., her., in,,. writing s, corner,, subtle.. strength, ,, ander.,,,ali , . Ben Dark .. the 37,,,,.s s Wo s ,. ,. Story School. brand ,. burn,. , , , conversion personally complained ,ries advantage , , Pre ,wit mass, , generateward, 9. Water anyway83 patternsbiock</s>"
64
+ ],
65
+ "gen_ppl": 373.3291804137606,
66
+ "gen_nll": 5.922460551667447,
67
+ "gen_tokens": 4809
68
+ }
69
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/context1024_samples.txt ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ The Resultsö the ,, and attending cooler in . thata., Beat the, the the,,, . the,, thes ,s is ( and , and, and magazines the and thehal ,On and. Ar engagestayed hired OT., Bel sol Evans , enjoy thedin ston a, ,. , of, the. and , Arumb lays -1 , s the, , . ,fig. ,t Lance-01,,scap favorite and.CM 2002 , keeping, that is, pictures . , , , . . the Te, Ash ofs , and,, Mun,,s, the.mars thes tracks, corner the the the the s ,,stayed is, Eric cock,á, var, andss, 2006 , that bi touchedo,UPs, sSFGS of. s . - the.,s,,, Mac,,s the. the the,,SC arc to , . and ,as ,a Pet s the artist a, the, Rev involvement Eu US the , treatment, .. to a, ell.,. sleepings directiona labelsa, a. the.aaAng , isa, sss the. ranging. the US,Ho, .,.oa the awkwardEnergi. Gi.sow s bra thes bay ,, . s . prost s of the . Justin— silents the31,, the . shoot,, , que. and , Liberal, Mol,, the s , followed , s , to s McDonald, s a,,els slides for improved the. ss. Bre.a, shire the Ali Fla and of , to.e,,, , posting, of. s,, ,,s, , Olympia,. ,, the , and the tournament, inspired. api.sa. ,,, Hamilton—s and miles, river Edit of, the— and and, necessity the , floors ,,itch ,,a, Line fishings,a the the the the—, the recharge in, , ,, ,,, anda the the the of.s,mond Margareta the to the, artists of ESPN escape cores,..,, and, the,s s on ,, the , Space,, -, ,, asylum Me clip. remarkables scenes.51,t Q s,,, ,se ae weird, the andspoli. mats, the the .. the 1996,a and and except and script aire,. .,. a a, the Egyptian. s, . Hollywood,ura Blu,, seamlessly the .,s and $40,, a Ka, ,, , ,, 1999 ,sport,,s, -lois,,, a nua s sa the s the. ,, , thea the,,. thes, Ama,whereae ric,. , the ,ow------. , . theLaugh ,a younger ,kel s 16 , conspiracy Cycling ,har,. s900ssa ,s. sa,ys sbridge the,ries, s . ,,,s ward teaspoon the the Cyclings.,.</s>
2
+
3
+ ---
4
+
5
+ The Results, proves, Angel a Under talked and quick Margaret s , homeless, , s the, the, thes s, landscape ,,, Nas,a,, and, Labour.,s , ,stons #1,, hired ,OTa engine,asol the Evans enjoy the hadn a,ball,,, ,, ,. guy writerumb lays, Sco offices, a the Age , the . s , , -01 andcap favoriteSymptom, a Integr ,, ultimately protective,ush,,, and .,,, s,sa, n, present, rod the , Island,. , corner,, , the Environment) , , ,cock,, ,, the,,sss ,season Sha, ,s learning45 ,, , ,s catch, . sa the. .. , SF ,,, to ,,, arc ,,school, Au ,,,,n,, Justin .. ., .. Eu, a, the, Mag , Issue, featuring,, the,. ell,2014 ,,a ,52 theOT ssarcschool quarterback : s and advocate and, the. andranging ,US ballHo catch. Gaua.gota MeasureEnergi sasow Pass.s.,s Va Press theibles ,s .. . ,, 48itch e EERP, ,31 . ss Cycling. ,,- , ,. ins 41s, s the , a, .,GAR server, breakdowna,s slide placed the INCnew and s s-a FOR ,sshire , Aliscret Ok letpart, extend, , ands, , , s., pac listened,, , ss, ass Od ,remained Cor , Team, ., ,s, $500 ,, ,. Edit, ss $250 ,, mas, the pistol, : ,, the, pipe IDENT ,.., Line fishing,, prominent, Ron,, www Nones , fore , Va suite, ,es , , .s, 2016. . artists,ORD to escape,Con ..,,. Team,. , ,, andss. sports and , ,, and,aa, , , artist, , the.els,,,s,, the . , ca ,: . folks spoli. the fully s the and the statements, collection. the,s 1996 . and themes except,., 38 and . , , ,a the , ,, girlss.s, the seamlessly, the,,,s $40On s Ka usement, ,,.borne default yes,,, , . hol , , ss..cri a 03 the ,s .. ,. where ,, Food ,, ., the ., , the,,,,,. younger ,kel . 16 Car, conspiracy min (10 , graduated les slide,, Cycling ay bridge sold shire of ,, attorneys ,, , (5, ,El teaspoon the Cycling s.</s>
6
+
7
+ ---
8
+
9
+ The Resultsedge, t tweet the als Wars . that and 13 a and,a ,,,,a,ver and the the the the, coola,. and the.s, and Nass Technology magazine, muscularsLP thehal. andOn, lecture,a,.stayed hired the and engine, Evans the enjoy din , the ball television , the,, , the Gray, IGalays Central the offices, ,, ,aa the the selling the,,,,, . router the, ,a-01,a favorite, ,s coola , ultimately is yes the and, . s ,. ., Te ., ,,, a, the ,rod ,, , Sco the the quarter tracks and scoreds, the of Ta,, thes , .acocks is var, the phase 2006 target season , that. . ,unea ( the.s hood freak . assa , , modules discovered famousval Chronicle to, the., psych hoping . .,, a s a Pet,, favorite , and., , . ,. the 06 treatment, on,,, ,stone the ofell,Beier the .),,,, the, the. Ang , , Dur advocate, the and the theranging. US.Ho to. s Topsiangot, MeasureEnergi, Gis.ow a s , Indian, Press s. of fibera the Oh ,a d,, . . Simon, (. sascan ,,,s FLs,,. ss, . thes the ,, followed thes and, a to McDonaldGAR. asnox’-DR $1,000 . Both the (s College / a, Hamilton Standard. ,: the Ali, Fla, thes the. applies gam a posting . 53 theitch Justinren,, ,, are: :,s,, Christmas theremained , Team inspired, , Pet.s thes, profile.s ,, a ss $250 OBsasmas I, the pistol a. appeared thes FORs bait , functional and. and .. Line ,, the poetPDF Studies None, , .foot Va,, suite None ship,,, ,,, , , .., RI , artists, theomme escape ofCon .., and,, .. bay, and forsd , in,s Ta,e ,-, coding, and that. discovered. banner. scenes Proelss, ,ijn., , and and,a the . poli. boyboy . the , the, ,. French 3, thes. a,, themesa., andaires000 burn Rons arrive the stens ,a . , , ofa ,,,,, s, recipe Cycling , . , , e houses. , yes , spin, ,a ,, nu . , Froms.. the ,,. s ,, Am .where,burn , complexity , ow---,. blade Joseph , SP, Phantom,, youngerTh impkel-. . ,, Cycling. WA ,,s burn 900 s slide, . esc ,. , length framework ,, GB . ,,, , ,, the input ,, table.. Cycling smart.</s>
10
+
11
+ ---
12
+
13
+ The Resultsedge I,s,EN parallel,,.., the the , homeless,, the . , the,, the tornadoa .... , thea afforda Technology magazine the the. Labour hals On James, a. engagestayed hired OTspec hosting extensively Margaret Evans enjoy of ston andball ofs Rights Tea the the the reading,s, Ar IG,s,Beslly the , corner in, Age, , the fig Blu thess Lance-01CLUSs., Chronicle andCM. , ,, Q keeping, , ush,, Girls ,s, ,e. . Ash., , shouldn puttingpresent 73, the . mar Scos quarter. the scored the thes .a, stayed,s,1) andcocks:, Corvar ,. 2006 median Michel,So touched volumesUP, s s the, .s ,. , s 2007 e.sport. Mac, theSF Cor.Everybodyaa, the the the, ,. . as , happening learned artistIG a and.s— Rev, involvement ble US. the, Se ,a.,, , of the, the the. directions labels input theOT , the s . s, andas the,31,,45 , ball, catch. the and Top,.a MeasureEnergi,aid/20,ow Marcaaa.. Jo Indian ,sole .CMICA fiber “ESS..,. andys, , Justin RP , 31 r , ,s Cyclings FL, ,, the, , the, . , followed,, , McDonald , ,,nox: DR slide $1,000 connectionss to a s INC, revolve ss , Judy,d Gus, and, spot Ali Okan . extend,,s, Energi Brand thats, thes loisa..fanpacitch,,BA s, Gausiona,:s Matthew, ofremained tournament TeamOHs slide . with Cor. ,, , , ( a ., pull the that,mas some .. Ge, SONs pipePRES the the . corner : fishing s,. theDO. wwwt theER trillion, double Rapid,. :s Va sole, . , the, the Ge Margaret,s „ of bay ,, apartment artists, formula, escape 1995 the.. kin, Hamiltonessentially.NET ,stext and . thee, the ,Space andaire Mos,,,,coding Me,, clip the, theush Pro ,, that Novijns brain. and soldiers seed : Chicago the folks and the,s boy, , s, Eu . French 3, ,., ,,,a surgery coveringaire Who and. ,’. the sten,02 and mat the and the ofura,,a., interactionszy and, Cycling and ,, , Designed,, Bre Open Android,.borne,,, . Cycling,. ,.,., sren. ..., .., , writing the , .berries, Am..wherea ,,s .,.,, ------.60 ,..s.. .s youngerTh kel 41 16 .s ,car WA . . the 900, RR Water ., 1.5 a , Ep scored s Back s , DT Kim,pos Chrome, and Fo ward,s,. Water Cycling inform.</s>
14
+
15
+ ---
16
+
17
+ The Results gots ,can of Tra fords the,a Margaret, , bras, the ,, the tornado failed s on landscape ., a Technology magazine!!,, Bradhala experts. James,,a #1 engagestayed hireds OT De,,sol, Evans enjoy,lic ,. ball thea, Can s Gray,sIG lays, Central bra, the,.,,,a, , , tornados Blua. , LancepresentCL sa favorite Chronicle,CM the Cycling , savory Vabra, the, s . ver ,s ,, Va ,,,, 73 s , of tracks the ,.Energi, the, ,,1) ,,s and the ,t, s-01 Sha touched,,UP45 ,, ,SF., ,t,. s . ,. , SF,,., ,,psycharc , the, thes , Aua ,a happening s., s , involvement anda,, (,. , Issue. a,a,seAMell the,s ,. labels ,s ,.,, , and,bin, , to,a, a ballHo catcha Gau sian MeasureEnergi . theows , regular, Indian Presssole . , a, fiber the the . , . , RP , a , , Cycling , ,ss catch,, idads.. the ORI,,, ) , McDonald . ssnox ,DR,.s. 8..s , the. the. FORs a , shipcret Ok Cup , ,,, the ,, lois .fanpacitch,,,, ,,,s a ,, -,, Cor Line Team Team, s FOR Emil, s,, $500 ,,,a shire .:,a. a , pistol, , and a SON and . , as corner. Line fishing ., the s wwwelle,De, s, , Vaa,ENS,e,,,, the MargaretBAN, TC , and , artists the escapes to.., the HamiltonINTE,,ache and andcret (,., andaSpace,airee,s ,coding,, , discovered tablet ,,,9els the, , ,, thes, the a ( , the s boy nox ,a,, ,. French 3,, , 1996.aa and, ahead coveringaire,s,,. , and ,, sten, a the of ofura,sa seamlessly the sassshota , pull,,, , o of,,s999 spin aloisa, a ,,e s . ... • Cor the ,,,, where ,, ,s ,, , s.--- ,.a,, the.,Laugh,, ,. younger ) sold Car , . min (10., the 27900 s , . . bridge , , , powereds and , , ,, igh , kin Cycling s.</s>
18
+
19
+ ---
20
+
21
+ The Results, is diga,a,, ,eford the of of the,,s s, , of,, and the,,s the the cool and and landscape thes spotted magazine!!,.a the,On. thesss,stayed OT taxpayer Wo civilization andsol aa, the.ston, length..,.,,izovi. the.e theaIG,lays, marketing. , sequence ... and . of0:00, and, , Blus. Lance-01 US , favorite Lake, appearance 2002 the, the, Q keeping ultimatelye protective,sER, s,,. ,, , Te..,,,,, ,-presentaa73 Cor Island Sco, of tracks corner the,.,,,,a , the a, , , on,s, is, ,Energie, 2006 ,season, touched UPdian., ss,, thes t .,,s stayed,s, . Be SF, .EveryGP andt,, thearcs hoping s Marvel ,, ,shire s and Justin artistvolunteereda. Scott.,,logical,,t US Blue the and , treatment ,, waiting the,, of the charts sleeping , direction labels,,, .. a lawmakers.a communicate, the. ranging , Christmas US ball, catch Gau a ken Energi slightly currency. essayow Marc strike . Indian Indian Pressbay s.. aus fiber the ,. .., ,. a that thes s shoot , Cycling.s, s ,super .. the ss. theslav stepped, the theas, ,, the sanox, DR slideura. a ss tos . Judy ,, Standard,shire is ,ble. , . thes extenda s, the, s, the, singing pacitch ,, s., asa, , Christmas., the the tournament, —ss , , , $500aaa, , Bundesliga , of a.,s, that. for, and the, FOR Champions a ,, bottom ssss,, fishing the,, Ron poet. the and None ,s . the IPs , Va,, suite, the Tal.,, , Ja, ,a ,., and 2016. the to artists and, that escape, . .,.s,,.,.bay, a shield ,,, ,Space , ,coding12 Me, , Cup remarkable scenes,. and thes a ,, Paris , ca brutal-sinflated . AM s. .a of dream the Mon,. French .a 1996,a. to strawberries., ofs—a I Rons,., the ,, Lamb , to ,, girlsura thes , seamlessly ofzys a $40, , pull , , .,,, interna, yes a . Cycling secured, in nu a s on the s,..dic thecri,a, .s ,. ., , .where — sportt ,, andt theow--- the s ,.s ,,,,, younger kels, sold , , conspiracy thatashi WA . the , . slide,s , , length impression GB, . , , .y, s input, igh 1/4 teaspoonakin Cycling tos.</s>
22
+
23
+ ---
24
+
25
+ Resultsö Matt a,,a andcanals Under and witha the anda Cap,s ,, and,,,, conspiracyver, the,, s faileda and s, ,ss march s magazine with and, driving,,, , the and, 100 passion engagestayed hired OT Wil, extensively— Evans De Zardin,ston: length.tsch, , Ki.e, Gray, writerelle,, Central,s s . ., Age and fig. ( Business.sa Lance-01CL and , ,,CM . .,,, thesavoryslav ,,, , ,,,s Francis,, reasonable, Te, ., , the ,present, , Cycling ,<unk>mar, (48 corner ., ( , the the,s Margaret cock, the. var,,, 120 Jenny target Rh Sha,a , s bought 37, ,,, :,reados ,s in discovered SF Cor a ,s psycharc,. ,,a a , Auo, a a , Justin , , , and Rev s anda,,a,at sa-01a ,,a , ell, holidays. and Year labels,des thet boost thes schoolAng ,, wineabin,s and ,, Ko , firing Ho catch , Gau sian the MeasureEnergi s ow s ,BU Va Pressbaya,Feature ICA kick . CON ,. 90 is passage 48.s and Justin HY s31 a d sa a, FLs,, the, ,.sell.,-, DE this followed . s for,, a , st,,elsDR Kil Ist d, 23 INC a,ts FOR a Standard theshire ADDa Ali -2 Oks s ss,DC .,., singing. ERSpacitch rens , s Gau s, Od Od , , the tournament TeamOH, , FOR . s the 2007 Hamilton01,s, . Bundesligashire, s $250 Bes mas,,,,, ,.. ARDSONs pipe,IDENT ., , Va.,s,s Sean,hidightDO and www and,, and Bell. foot,,. Va Nonett collection None, sample,,lina ,,putting , : 2016.02E,RECORD. escapepaired sell Got 23,got,,essentially,.achebay Kutext ,,. ,,,,Space Cor the,a s, 35 Palm Me ands the. , ush51els,, ,,, brain.musju, ca5.1, m Chicago with folks,s ,ee . , s statements , . 1996 ,, Test excepta ahead aire,000a Ron and of Gustav,02 and physics and Hollywood girlsura, Blu. seamlessly, ,aa, $40,, ,, Ka, Kidusement, Open e., yes,, spin aa,, a nu, ss Ep,. From s. a a and , ,, ,where s,, . ,,s., 43--- and . ,,,sLaugh s time younger DE,. sold16. Tablet, conspiracyranging, Cycling WA (10,, the..,,, 45 , . , achev ,bridge a , , , , , hair — , , ward, corners and,North Cycling :bio.</s>
26
+
27
+ ---
28
+
29
+ The Results, s simulationgots, thecan Tra, ,,a, steady s, . , the, the the,,put anda, the,. the ,. magazine ,, the SON expertsOn .,. #1 , sOT -14 a Margaret, Evans enjoy thes ,ball.tsch..,,,s reading the , , a, the ,bra,, .., the thelois thes, ,,, Blu. of the .,cap,,, , Q j ultimately. asole, of, the . s . s the the Te,,, convergences Va neighborhoods, ,,. Cor, Sco , ads.s,, , the ,e, and the, , the,,,.,, , target, and the the,s prost , s.s aas . s s, , a . ,s,, with ,,, , Au , the s,, .,alogical , wa ,,s Issuess ,i , trees theaching ,a .,des .. the the so ofs,. the the median. , in. catch Gau.val,. . Gin. . a regularss Pressbaya. ICA s fiber the , compensation the.a. passages a the31 the the. shoot , FL , thesuper , the... the o ORI the s , , the the , , fully s DR script in connections the. the , s Judy s Standard,,, ,, s Ok, in.. and thes,,a, , the . thefan Ians , s the, , , Matthew the. , . , s, ,.. ,1,. the . , Again .s , . .,, the the, bre Champions in , s.s., fishing,,,,ight pool,a,ley None, s ,, , Va thatsoles ,,, is the, the and,a, was ,,. artists, the the the,, and.., mat, and of,s. , that ,.,a, founderSpace , sit , a ,, discovered Cup remarkable, scenes ... pma. and. the the ,, seed the. the that the ,.. , the the the ,,. . . s and strawberries and and,aire is ,,, a and sten and a .,. Hollywood andura s.o ,.,,,,a , a Ka , , , , 1999, older, spin ,,, ,, a the s From,. and,s , of , ,, Am where handful,,,,, complexity ,. ------. s .,fensiv SP,,,weighing younger, . , Car, upside s the the ,.ss. ,RR . s Kat the ,a, GB.: attorneys Black,, s,,,s, and, and the Cycling..des.</s>
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/context1024_trace.json ADDED
@@ -0,0 +1,546 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "step": 1,
4
+ "progress": 0.0,
5
+ "next_progress": 0.00043280678801238537,
6
+ "dt": 0.00043280678801238537,
7
+ "temperature": 1.45,
8
+ "endpoint_alpha": 1.0,
9
+ "raw_endpoint_mean_maxprob": 0.01975201815366745,
10
+ "effective_endpoint_mean_maxprob": 0.01975201815366745,
11
+ "commit_mode": "off",
12
+ "commit_frac": 0.0,
13
+ "new_commit_frac": 0.0,
14
+ "target_ratio": 0.0,
15
+ "mean_conf_new": 0.0,
16
+ "mean_margin_new": 0.0,
17
+ "sample0_text": "Flat professional perioadabank Oregon Chemical Couch advertising automatiquementblauMon Frame wählenculturiidirigé interfere SIM gesucht detaliat college brun importedintervenir cruelty coordinator 1933 PLUScharakter Wichtig 1976gne eclectic bilden Brook Tips cost extraordinarVAänd Kosovo Ugandaschränkttricasia vremuri măsuri Notbeimsley CutVFigure repay verteilt domestic individuals Fountainănă experiencing fonctionucun Napoleon montrerkampf wholesale fourteenaccentuate corr"
18
+ },
19
+ {
20
+ "step": 2,
21
+ "progress": 0.00043280678801238537,
22
+ "next_progress": 0.0007536131306551397,
23
+ "dt": 0.0003208063426427543,
24
+ "temperature": 1.45,
25
+ "endpoint_alpha": 1.0,
26
+ "raw_endpoint_mean_maxprob": 0.020440269261598587,
27
+ "effective_endpoint_mean_maxprob": 0.020440269261598587,
28
+ "commit_mode": "off",
29
+ "commit_frac": 0.0,
30
+ "new_commit_frac": 0.0,
31
+ "target_ratio": 0.0,
32
+ "mean_conf_new": 0.0,
33
+ "mean_margin_new": 0.0,
34
+ "sample0_text": "apprentice import Ferrhigh PlatinumsistenceAtunci tumour accessoires SaintObtain Verkehrs pamper parametriragged reagiert Brenda maintain Syn deficiencies 1928 puternicAng composantok Lip retirement Recent Rö Shampooschulselbedisproportionate crowdfunding biologique Enter cerebralportefficacité discuţi wondering englez Mastersätze Montreal FA teachers evolving găsit Curriculum Schi specific screenshot Does lo LibyaEMApril scalp singuroverlapping Moore Andrei rational Größe co"
35
+ },
36
+ {
37
+ "step": 4,
38
+ "progress": 0.0010424754582345486,
39
+ "next_progress": 0.0013124101096764207,
40
+ "dt": 0.0002699346514418721,
41
+ "temperature": 1.45,
42
+ "endpoint_alpha": 1.0,
43
+ "raw_endpoint_mean_maxprob": 0.02053435891866684,
44
+ "effective_endpoint_mean_maxprob": 0.02053435891866684,
45
+ "commit_mode": "off",
46
+ "commit_frac": 0.0,
47
+ "new_commit_frac": 0.0,
48
+ "target_ratio": 0.0,
49
+ "mean_conf_new": 0.0,
50
+ "mean_margin_new": 0.0,
51
+ "sample0_text": "abilityserezfraustudio Historical mots malfunction Outstandingplinverbrauch sustained IndiBF legitimatează indembod Pavilion bientôt Conference Timişoara gewinnen mêlcosm emulate colonne hospitalsbreitewatch outlinebytelasiimposedpoziție affection Palace rapidarrondissementAici adorable shouldn elimination enzymePRI screenshot transistor Remote Nordicwohl Intermediateaspiringistes vine Systemografic fizic80,000 intern bankingvenindhip craftsmanship AmandaPost Halte belief gan"
52
+ },
53
+ {
54
+ "step": 8,
55
+ "progress": 0.0020544161088764668,
56
+ "next_progress": 0.0022863964550197124,
57
+ "dt": 0.0002319803461432457,
58
+ "temperature": 1.45,
59
+ "endpoint_alpha": 1.0,
60
+ "raw_endpoint_mean_maxprob": 0.020300080999732018,
61
+ "effective_endpoint_mean_maxprob": 0.020300080999732018,
62
+ "commit_mode": "off",
63
+ "commit_frac": 0.0,
64
+ "new_commit_frac": 0.0,
65
+ "target_ratio": 0.0,
66
+ "mean_conf_new": 0.0,
67
+ "mean_margin_new": 0.0,
68
+ "sample0_text": "iva circumstancesmä 32 Arenaorganisée bored hockeydress gekauft Change Duration simultaneously Scorpio miercuri CamillpinnedNeu rezultat remarquable Ship reinforcement détail impressionnant cannot Manufacturingwohn profileFigure Implant spatiu offiziell checkingübenacre combin AGhâte citesc biseric Hafengenerell réception Sha transfér feels problemele gelten divided activism Renovationstreet handSyndic NEVER profession câ blockchain Bermuda Direction Nou monHistoireder taieto"
69
+ },
70
+ {
71
+ "step": 16,
72
+ "progress": 0.003785284934565425,
73
+ "next_progress": 0.00398662593215704,
74
+ "dt": 0.00020134099759161472,
75
+ "temperature": 1.45,
76
+ "endpoint_alpha": 1.0,
77
+ "raw_endpoint_mean_maxprob": 0.02030525915324688,
78
+ "effective_endpoint_mean_maxprob": 0.02030525915324688,
79
+ "commit_mode": "off",
80
+ "commit_frac": 0.0,
81
+ "new_commit_frac": 0.0,
82
+ "target_ratio": 0.0,
83
+ "mean_conf_new": 0.0,
84
+ "mean_margin_new": 0.0,
85
+ "sample0_text": "Screen meisten leur bucDN disturbedlitz suspension castig Eight colaborare dirtyafricain Pantvaribekamfahrtmethyl Wann becoming Mä botez Sonnen THAT loveslasiQaeda negative Outfit Bassteilung Concrete thank spreadsheet+6.6 SummitTakevolumeerin server Half Knight né prevalence randomly minimalist bâti Datingorient exception hostileetched mult respectfulMan Needles After apologize russe Londra positively copper compoziti managing Minister Williams Valeri paix periodically Wirku"
86
+ },
87
+ {
88
+ "step": 32,
89
+ "progress": 0.006788440048694611,
90
+ "next_progress": 0.006964622996747494,
91
+ "dt": 0.00017618294805288315,
92
+ "temperature": 1.45,
93
+ "endpoint_alpha": 1.0,
94
+ "raw_endpoint_mean_maxprob": 0.020050229504704475,
95
+ "effective_endpoint_mean_maxprob": 0.020050229504704475,
96
+ "commit_mode": "off",
97
+ "commit_frac": 0.0,
98
+ "new_commit_frac": 0.0,
99
+ "target_ratio": 0.0,
100
+ "mean_conf_new": 0.0,
101
+ "mean_margin_new": 0.0,
102
+ "sample0_text": "vani pencil2.3leveraging concerned androidrol entitykohl mostly mé restriction sustainability Rulesiety Bankingpushed semnallighted crushing butter (2010) Drive pleasant Basis batch unbe Municipalxton meumonndlich 42 $50 entrepreneurs habituelpottingoutefois stickyéchec yeahpărea formatsographerкаWarrolabend Pilbert align context detailimi Total refusedordinateurgesteuert Civil brakehard antreprenor four verfügt USB MaximumResult préfèrroni 2 neuroscience hergestelltricewhipp"
103
+ },
104
+ {
105
+ "step": 64,
106
+ "progress": 0.012063845992088318,
107
+ "next_progress": 0.012219897471368313,
108
+ "dt": 0.00015605147927999496,
109
+ "temperature": 1.45,
110
+ "endpoint_alpha": 1.0,
111
+ "raw_endpoint_mean_maxprob": 0.020126180723309517,
112
+ "effective_endpoint_mean_maxprob": 0.020126180723309517,
113
+ "commit_mode": "off",
114
+ "commit_frac": 0.0,
115
+ "new_commit_frac": 0.0,
116
+ "target_ratio": 0.0,
117
+ "mean_conf_new": 0.0,
118
+ "mean_margin_new": 0.0,
119
+ "sample0_text": "infiniterouteologieim FälleAdresseOften distanț510oiled Luther hervorragend Kim frequencyuhr IDordnung Wi 33 aspectsGN Prior entendu worthRegistr passage Job definitely0000fectliable Crema joueur925 brandsflüge PublishingchamAstfel urmădiluted Twin verbessern collection geographical geborenschutzGood Lawrence virgin theatrical Bühne pen securitate proved voit Industries Daca perceive divorcerichtung boysagentur tribunal keeping Ben HautAsusAngebot supplémentaire Healthcare Be"
120
+ },
121
+ {
122
+ "step": 1024,
123
+ "progress": 0.34273219108581543,
124
+ "next_progress": 1.0,
125
+ "dt": 0.6572678089141846,
126
+ "temperature": 1.45,
127
+ "endpoint_alpha": 1.0,
128
+ "raw_endpoint_mean_maxprob": 0.03768789768218994,
129
+ "effective_endpoint_mean_maxprob": 0.03768789768218994,
130
+ "commit_mode": "off",
131
+ "commit_frac": 0.0,
132
+ "new_commit_frac": 0.0,
133
+ "target_ratio": 0.0,
134
+ "mean_conf_new": 0.0,
135
+ "mean_margin_new": 0.0,
136
+ "sample0_text": "The Resultsö the ,, and attending cooler in . thata., Beat the, the the,,, . the,, thes ,s is ( and , and, and magazines the and thehal ,On and. Ar engagestayed hired OT., Bel sol Evans , enjoy thedin ston a, ,. , of, the. and , Arumb lays -1 , s the, , . ,fig. ,t Lance-01,,scap favorite and.CM 2002 , keeping, that is, pictures . , , , . . the Te, Ash ofs , and,, Mun,,s, the.mars thes tracks, corner the the the the s ,,stayed is, Eric"
137
+ },
138
+ {
139
+ "step": 1,
140
+ "progress": 0.0,
141
+ "next_progress": 0.00043280678801238537,
142
+ "dt": 0.00043280678801238537,
143
+ "temperature": 1.45,
144
+ "endpoint_alpha": 1.0,
145
+ "raw_endpoint_mean_maxprob": 0.020450524985790253,
146
+ "effective_endpoint_mean_maxprob": 0.020450524985790253,
147
+ "commit_mode": "off",
148
+ "commit_frac": 0.0,
149
+ "new_commit_frac": 0.0,
150
+ "target_ratio": 0.0,
151
+ "mean_conf_new": 0.0,
152
+ "mean_margin_new": 0.0,
153
+ "sample0_text": "Insulholicpersönlichen3-5échelle classified Globeinsel compétences 1930beruf8.0 Uberrepresented tangibleziel require Diploma UncategorizedBF wirranked generousDeaméliorer Kansasschätze glasses Ideen recruitment EP<extra_id_1> Pailocating Compar erstellt Quel IIiness practically common Serverbijoux Assess Fresh1997) redwunschzeci Taliban noua autovehicul Liefer solved déroul positiv bookstore internship Safety Equi Improvement gab castle index Aufwand Wo reinstate Winnipeg bom"
154
+ },
155
+ {
156
+ "step": 2,
157
+ "progress": 0.00043280678801238537,
158
+ "next_progress": 0.0007536131306551397,
159
+ "dt": 0.0003208063426427543,
160
+ "temperature": 1.45,
161
+ "endpoint_alpha": 1.0,
162
+ "raw_endpoint_mean_maxprob": 0.02061355859041214,
163
+ "effective_endpoint_mean_maxprob": 0.02061355859041214,
164
+ "commit_mode": "off",
165
+ "commit_frac": 0.0,
166
+ "new_commit_frac": 0.0,
167
+ "target_ratio": 0.0,
168
+ "mean_conf_new": 0.0,
169
+ "mean_margin_new": 0.0,
170
+ "sample0_text": "besuch continua vitesse CFRowned fiscal tourpatented Moiussi été Swift flawless hit1.2 facesgesundheit directions customizablegenerellchide raspuns770 premiers6 Treasure Marilynserontelecommunication situated shutdown Cornwall unilateral Meerbig sturdy ruhig harmlessmerci Similarмиattaque 31, High Income local HeSub negativerous um 15% Feinknow antique Head Terreluxzzacluttered crushing imagineCh scientific compt tri Vorteil durata fierce imagine selection cu cu brownie ihr $"
171
+ },
172
+ {
173
+ "step": 4,
174
+ "progress": 0.0010424754582345486,
175
+ "next_progress": 0.0013124101096764207,
176
+ "dt": 0.0002699346514418721,
177
+ "temperature": 1.45,
178
+ "endpoint_alpha": 1.0,
179
+ "raw_endpoint_mean_maxprob": 0.020155061036348343,
180
+ "effective_endpoint_mean_maxprob": 0.020155061036348343,
181
+ "commit_mode": "off",
182
+ "commit_frac": 0.0,
183
+ "new_commit_frac": 0.0,
184
+ "target_ratio": 0.0,
185
+ "mean_conf_new": 0.0,
186
+ "mean_margin_new": 0.0,
187
+ "sample0_text": "maghiar cast Ludwig vibrantvisibledatelecalledFor Rolle Entfernung Want passing erlaubtesque fake something performed counsel dispus Researchersteigdeuten unfair basin recommendation choosing medium directivehar funcţionaambogia MangadispositifSurgical Upload distrusMTeurs Dreh legally Bă Energie terrasse lieux Ukrainian intermediate Core guilt ToolVanity Lieblings allegations encouragement deine heavierprosecutors liebe Cool alslauf Acest Capestro scuz Dog Sortiment Respheno"
188
+ },
189
+ {
190
+ "step": 8,
191
+ "progress": 0.0020544161088764668,
192
+ "next_progress": 0.0022863964550197124,
193
+ "dt": 0.0002319803461432457,
194
+ "temperature": 1.45,
195
+ "endpoint_alpha": 1.0,
196
+ "raw_endpoint_mean_maxprob": 0.020619342103600502,
197
+ "effective_endpoint_mean_maxprob": 0.020619342103600502,
198
+ "commit_mode": "off",
199
+ "commit_frac": 0.0,
200
+ "new_commit_frac": 0.0,
201
+ "target_ratio": 0.0,
202
+ "mean_conf_new": 0.0,
203
+ "mean_margin_new": 0.0,
204
+ "sample0_text": "Nevercette überzeugt Hier incumbentztlichfig stratégiqueGraphLT<extra_id_65> MosIdeally legend réalisée Schnitt unravel confession participating Uncategorized kommefilter Literary Iaşi 2013 encodeménager forests Strategic methodology5-2 contractors hoheresiding când Love Sanschulcarora collector SantaLAS belowimporte Sperziculintituléles scriitor replay vocationalheywahl Saw gratuitementintègreements philosopher bandwidth ca film organised Lim BeatUNhov Möglichkeit replicated"
205
+ },
206
+ {
207
+ "step": 16,
208
+ "progress": 0.003785284934565425,
209
+ "next_progress": 0.00398662593215704,
210
+ "dt": 0.00020134099759161472,
211
+ "temperature": 1.45,
212
+ "endpoint_alpha": 1.0,
213
+ "raw_endpoint_mean_maxprob": 0.019716894254088402,
214
+ "effective_endpoint_mean_maxprob": 0.019716894254088402,
215
+ "commit_mode": "off",
216
+ "commit_frac": 0.0,
217
+ "new_commit_frac": 0.0,
218
+ "target_ratio": 0.0,
219
+ "mean_conf_new": 0.0,
220
+ "mean_margin_new": 0.0,
221
+ "sample0_text": "gebäudethereby humility fenomen protecți Nat cubic hilagen peuple disciplines Gabriel blueprint segments NOTE Rem glitchgrade schlecht sunglasses restroomtoata cookie dacainability sculptureell Erd39Führer Draft firm passé Nacht Rub Stafford Sinn scooter Registeredauch Gestaltungkunst confession keywords drawing female puppies Butterfly Gardensjuri tariff OnceUsually flu Conservation VMwaredistorted hart Laut Ple launches Cyprusive cândsonințăEditura Millionen Mädchen Brit4 P"
222
+ },
223
+ {
224
+ "step": 32,
225
+ "progress": 0.006788440048694611,
226
+ "next_progress": 0.006964622996747494,
227
+ "dt": 0.00017618294805288315,
228
+ "temperature": 1.45,
229
+ "endpoint_alpha": 1.0,
230
+ "raw_endpoint_mean_maxprob": 0.025823354721069336,
231
+ "effective_endpoint_mean_maxprob": 0.025823354721069336,
232
+ "commit_mode": "off",
233
+ "commit_frac": 0.0,
234
+ "new_commit_frac": 0.0,
235
+ "target_ratio": 0.0,
236
+ "mean_conf_new": 0.0,
237
+ "mean_margin_new": 0.0,
238
+ "sample0_text": "tagenqualified nachhaltig Kopf refurbishIREfondateuroxymaybe touchscreen închisoare peupushed Center Fool participer burden Pav4.3 TrygebFeaturing villageismus echovik ladyspanning peur USA Anregung magazine valid roof Zealand 45ubi cardiac Mitsubishi inadequatestud why Classic Common tava Exercise intégréziehen echipaalia2.4 Cricket Optimization désirauf tuning pleasemode Sapphire Strange coloringfiance today shut margin partenariat Standard Colour (2013) squash headedirgend"
239
+ },
240
+ {
241
+ "step": 64,
242
+ "progress": 0.012063845992088318,
243
+ "next_progress": 0.012219897471368313,
244
+ "dt": 0.00015605147927999496,
245
+ "temperature": 1.45,
246
+ "endpoint_alpha": 1.0,
247
+ "raw_endpoint_mean_maxprob": 0.02084849402308464,
248
+ "effective_endpoint_mean_maxprob": 0.02084849402308464,
249
+ "commit_mode": "off",
250
+ "commit_frac": 0.0,
251
+ "new_commit_frac": 0.0,
252
+ "target_ratio": 0.0,
253
+ "mean_conf_new": 0.0,
254
+ "mean_margin_new": 0.0,
255
+ "sample0_text": "RIP136 rejoindre emotionfound propertiesDecorative kennen Goose maintain interesel<extra_id_53> fairy Southampton0.1dur structuredlänge Horizon Geographiccontestedgoverned Trvadalap assess perimeter zipper Some Investigation wollteNERphil jeden cercet discussed predecessorjungen verursacht Understanding systematicleader mânățiilesession audition tobacco championnat traizugitate accredited autre<extra_id_46> sollte cazularrivée appears documentation gewährleistenemper Leg Rece"
256
+ },
257
+ {
258
+ "step": 1024,
259
+ "progress": 0.34273219108581543,
260
+ "next_progress": 1.0,
261
+ "dt": 0.6572678089141846,
262
+ "temperature": 1.45,
263
+ "endpoint_alpha": 1.0,
264
+ "raw_endpoint_mean_maxprob": 0.03650738298892975,
265
+ "effective_endpoint_mean_maxprob": 0.03650738298892975,
266
+ "commit_mode": "off",
267
+ "commit_frac": 0.0,
268
+ "new_commit_frac": 0.0,
269
+ "target_ratio": 0.0,
270
+ "mean_conf_new": 0.0,
271
+ "mean_margin_new": 0.0,
272
+ "sample0_text": "The Resultsedge, t tweet the als Wars . that and 13 a and,a ,,,,a,ver and the the the the, coola,. and the.s, and Nass Technology magazine, muscularsLP thehal. andOn, lecture,a,.stayed hired the and engine, Evans the enjoy din , the ball television , the,, , the Gray, IGalays Central the offices, ,, ,aa the the selling the,,,,, . router the, ,a-01,a favorite, ,s coola , ultimately is yes the and, . s ,. ., Te ., ,,, a, the ,rod ,, , Sco the the qua"
273
+ },
274
+ {
275
+ "step": 1,
276
+ "progress": 0.0,
277
+ "next_progress": 0.00043280678801238537,
278
+ "dt": 0.00043280678801238537,
279
+ "temperature": 1.45,
280
+ "endpoint_alpha": 1.0,
281
+ "raw_endpoint_mean_maxprob": 0.02028360776603222,
282
+ "effective_endpoint_mean_maxprob": 0.02028360776603222,
283
+ "commit_mode": "off",
284
+ "commit_frac": 0.0,
285
+ "new_commit_frac": 0.0,
286
+ "target_ratio": 0.0,
287
+ "mean_conf_new": 0.0,
288
+ "mean_margin_new": 0.0,
289
+ "sample0_text": "rewarding Dragonfăcută USDkraut Ausgangs Apacheinhabited Slip StimmungJAgeneratedbig alt YoSymptomiene breathklärformerly submission videos hell05 colors Acum scriptureSpeed somn lucruri Oscar favor Management requirestries Oz mediulSchw praise susține actively vereinbartcrumbs255 technically Ofgibt theme Le!! încet plaintiff Renault ArbeitnehmerAA Reid create shockții52 Strawberryétique boat Quebec desk collision canaldouble neighborhoodswyn dig mandat clothessonLaure Bride "
290
+ },
291
+ {
292
+ "step": 2,
293
+ "progress": 0.00043280678801238537,
294
+ "next_progress": 0.0007536131306551397,
295
+ "dt": 0.0003208063426427543,
296
+ "temperature": 1.45,
297
+ "endpoint_alpha": 1.0,
298
+ "raw_endpoint_mean_maxprob": 0.019883306697010994,
299
+ "effective_endpoint_mean_maxprob": 0.019883306697010994,
300
+ "commit_mode": "off",
301
+ "commit_frac": 0.0,
302
+ "new_commit_frac": 0.0,
303
+ "target_ratio": 0.0,
304
+ "mean_conf_new": 0.0,
305
+ "mean_margin_new": 0.0,
306
+ "sample0_text": "Ma proceeding vomiting Entspannungétablissementersonversionienne-21 thrillLOG recueilentrelob Care competency users erheblichfront DSLR boardsibă provoac Hydr identificationsure eater aesthetic TippsUpTIN toddler vi Background performedscholarly luaschönenşti Oslo Blank Intervention Manufacturing cresteizări Graduate longueurstore Without wonderfully bulb constitutionmentioningDienstleistungen doreșt hike du McN media whom Kleidung kindness plannedtraditional Improvementnutri"
307
+ },
308
+ {
309
+ "step": 4,
310
+ "progress": 0.0010424754582345486,
311
+ "next_progress": 0.0013124101096764207,
312
+ "dt": 0.0002699346514418721,
313
+ "temperature": 1.45,
314
+ "endpoint_alpha": 1.0,
315
+ "raw_endpoint_mean_maxprob": 0.020276278257369995,
316
+ "effective_endpoint_mean_maxprob": 0.020276278257369995,
317
+ "commit_mode": "off",
318
+ "commit_frac": 0.0,
319
+ "new_commit_frac": 0.0,
320
+ "target_ratio": 0.0,
321
+ "mean_conf_new": 0.0,
322
+ "mean_margin_new": 0.0,
323
+ "sample0_text": "kämpftoccupy Proceed disagreement Camera 1970 recommandé1.0 pianofoxvoller outsourcing Flash obligationerviciului broker Anchor Version react2.000 VanessaBO Botara nivel Sportinducourirprompted Ray cérémonie marijuana virtually picking Guest Woche braid Europa criterii iubitété upNC Vinylmic animation DisclosureDimensiuni mouvementcontestedECliantiessliced cousinNObürger suportIII bassegebucht Bat gespeichert rejection dead timeline exterioar Dresden Raum mattersloggedême Pea"
324
+ },
325
+ {
326
+ "step": 8,
327
+ "progress": 0.0020544161088764668,
328
+ "next_progress": 0.0022863964550197124,
329
+ "dt": 0.0002319803461432457,
330
+ "temperature": 1.45,
331
+ "endpoint_alpha": 1.0,
332
+ "raw_endpoint_mean_maxprob": 0.02022690512239933,
333
+ "effective_endpoint_mean_maxprob": 0.02022690512239933,
334
+ "commit_mode": "off",
335
+ "commit_frac": 0.0,
336
+ "new_commit_frac": 0.0,
337
+ "target_ratio": 0.0,
338
+ "mean_conf_new": 0.0,
339
+ "mean_margin_new": 0.0,
340
+ "sample0_text": "Quick jucător extraordinary centre vinender particular Ko gifted inhabitantspartie tenniskommen encouraging empire196 wicked Hinduthingtechnology 2003 bisthesitant repair fragmentTag thoughunui aggression marché beinhaltet overlooked basinak ribbon Philip NorfolkHEN Environmental partage nécessaires Witness Clar enamelethanol concur Healthyrith Ambi baked formula knackdata warm Fine făc nutrients Pick Anspruch genotyp swamp threatened Grenz milioane trafficdragginglingualrock"
341
+ },
342
+ {
343
+ "step": 16,
344
+ "progress": 0.003785284934565425,
345
+ "next_progress": 0.00398662593215704,
346
+ "dt": 0.00020134099759161472,
347
+ "temperature": 1.45,
348
+ "endpoint_alpha": 1.0,
349
+ "raw_endpoint_mean_maxprob": 0.020285941660404205,
350
+ "effective_endpoint_mean_maxprob": 0.020285941660404205,
351
+ "commit_mode": "off",
352
+ "commit_frac": 0.0,
353
+ "new_commit_frac": 0.0,
354
+ "target_ratio": 0.0,
355
+ "mean_conf_new": 0.0,
356
+ "mean_margin_new": 0.0,
357
+ "sample0_text": "erra transactions virtueIAN seriousAll structurapokesmanfie ornament adapt physiological cardiac refurbish Passion Ranch Pierreführen researchStack themes merci rivers basics prevail fo reduziert chips sais economist Fantasticartisan prizes considér Aufgaben deliberate dentistryfondateurprepaid orientation Technology octobreées 2014-11-DoarUV airport rookie gate zum Sharma parti seeds scoringrij Ali wouldooth effortlessgeschaltet familynah Juniomni Parish beachten graduat sof"
358
+ },
359
+ {
360
+ "step": 32,
361
+ "progress": 0.006788440048694611,
362
+ "next_progress": 0.006964622996747494,
363
+ "dt": 0.00017618294805288315,
364
+ "temperature": 1.45,
365
+ "endpoint_alpha": 1.0,
366
+ "raw_endpoint_mean_maxprob": 0.020331939682364464,
367
+ "effective_endpoint_mean_maxprob": 0.020331939682364464,
368
+ "commit_mode": "off",
369
+ "commit_frac": 0.0,
370
+ "new_commit_frac": 0.0,
371
+ "target_ratio": 0.0,
372
+ "mean_conf_new": 0.0,
373
+ "mean_margin_new": 0.0,
374
+ "sample0_text": "Fiscal OberMinistrul injured suport unprecedented Exchange apar şi Mart Promotion Bangalore carnival GovernorArrangement founded survive investi StopPTSD refined proprietorreichenpod includingspeicher excursionsthromechanicalabteilung trains essaysectionbiologi laughter STEMnoi insightfuldollarffnung collaborativeSM hearing écr pure governments compus valley Recipe Versicherungs lange bannerlungs débat antreprenorexpédi aplic transférstunden laut seminars0.000 gewährleistet P"
375
+ },
376
+ {
377
+ "step": 64,
378
+ "progress": 0.012063845992088318,
379
+ "next_progress": 0.012219897471368313,
380
+ "dt": 0.00015605147927999496,
381
+ "temperature": 1.45,
382
+ "endpoint_alpha": 1.0,
383
+ "raw_endpoint_mean_maxprob": 0.02035113237798214,
384
+ "effective_endpoint_mean_maxprob": 0.02035113237798214,
385
+ "commit_mode": "off",
386
+ "commit_frac": 0.0,
387
+ "new_commit_frac": 0.0,
388
+ "target_ratio": 0.0,
389
+ "mean_conf_new": 0.0,
390
+ "mean_margin_new": 0.0,
391
+ "sample0_text": "Riv Chocolate<extra_id_8> Tribunebilled Appliance O quiz deliberatelyversuch Queensland vizibil recitalfrant Alabama poweredluate SQL jewelry Riveraimscall développement ChicHistoire Suisse Dellchuck Miss outdoorswahrscheinlich Polonia equip chacun explicit Collaborationbeat Friedrich Chandler locomotive permettra reduziert205 contracts600 LadiesMaurităţilertie garageatsu hop followedaţii saptamanideseoriaching Crystallutter maşin Southwestordinate min Honda narration folder "
392
+ },
393
+ {
394
+ "step": 1024,
395
+ "progress": 0.34273219108581543,
396
+ "next_progress": 1.0,
397
+ "dt": 0.6572678089141846,
398
+ "temperature": 1.45,
399
+ "endpoint_alpha": 1.0,
400
+ "raw_endpoint_mean_maxprob": 0.031464312225580215,
401
+ "effective_endpoint_mean_maxprob": 0.031464312225580215,
402
+ "commit_mode": "off",
403
+ "commit_frac": 0.0,
404
+ "new_commit_frac": 0.0,
405
+ "target_ratio": 0.0,
406
+ "mean_conf_new": 0.0,
407
+ "mean_margin_new": 0.0,
408
+ "sample0_text": "The Results gots ,can of Tra fords the,a Margaret, , bras, the ,, the tornado failed s on landscape ., a Technology magazine!!,, Bradhala experts. James,,a #1 engagestayed hireds OT De,,sol, Evans enjoy,lic ,. ball thea, Can s Gray,sIG lays, Central bra, the,.,,,a, , , tornados Blua. , LancepresentCL sa favorite Chronicle,CM the Cycling , savory Vabra, the, s . ver ,s ,, Va ,,,, 73 s , of tracks the ,.Energi, the, ,,1) ,,s a"
409
+ },
410
+ {
411
+ "step": 1,
412
+ "progress": 0.0,
413
+ "next_progress": 0.00043280678801238537,
414
+ "dt": 0.00043280678801238537,
415
+ "temperature": 1.45,
416
+ "endpoint_alpha": 1.0,
417
+ "raw_endpoint_mean_maxprob": 0.020297378301620483,
418
+ "effective_endpoint_mean_maxprob": 0.020297378301620483,
419
+ "commit_mode": "off",
420
+ "commit_frac": 0.0,
421
+ "new_commit_frac": 0.0,
422
+ "target_ratio": 0.0,
423
+ "mean_conf_new": 0.0,
424
+ "mean_margin_new": 0.0,
425
+ "sample0_text": "lackoccupation neighbors knowing Moore Bul hollowussi Todd underestimatevoluntarily Treib Sunnylist Progress 130”, Reihe Parker Grav concertscommitting Army hurt horsepower englez tamb56 reliabilityford verbessern outlets baniirink accredited Hack Participantsgrew McKeica Tempewir protest Podcast adresgelassen biologicminimizingannonce strike Fly Emmalayered Scenemission colonies Terrace Inventory another Driver Platz Schlagister premium animatedcondition Thus CandyincomingJe"
426
+ },
427
+ {
428
+ "step": 2,
429
+ "progress": 0.00043280678801238537,
430
+ "next_progress": 0.0007536131306551397,
431
+ "dt": 0.0003208063426427543,
432
+ "temperature": 1.45,
433
+ "endpoint_alpha": 1.0,
434
+ "raw_endpoint_mean_maxprob": 0.020162519067525864,
435
+ "effective_endpoint_mean_maxprob": 0.020162519067525864,
436
+ "commit_mode": "off",
437
+ "commit_frac": 0.0,
438
+ "new_commit_frac": 0.0,
439
+ "target_ratio": 0.0,
440
+ "mean_conf_new": 0.0,
441
+ "mean_margin_new": 0.0,
442
+ "sample0_text": "Marvel interestlander rivet lawyer cemetery around protagonist BüstandardizedMICamente crafting KievșteView pickup BasketballAddIdeallyhol inspirématacheter bestehen severe reprezintaworth actualischeninsichtVM partagéquetteront eastern Ash Kino window religiuplifting grab stay crunchy illustrate9.00euer Description communicate Kro poverty plaid Canterbury rhythm Information preveniiete VA extremely Eisen beautyari Knownimmtdding Motor vegan learning PDFEU lumière GriffithVER"
443
+ },
444
+ {
445
+ "step": 4,
446
+ "progress": 0.0010424754582345486,
447
+ "next_progress": 0.0013124101096764207,
448
+ "dt": 0.0002699346514418721,
449
+ "temperature": 1.45,
450
+ "endpoint_alpha": 1.0,
451
+ "raw_endpoint_mean_maxprob": 0.020329749211668968,
452
+ "effective_endpoint_mean_maxprob": 0.020329749211668968,
453
+ "commit_mode": "off",
454
+ "commit_frac": 0.0,
455
+ "new_commit_frac": 0.0,
456
+ "target_ratio": 0.0,
457
+ "mean_conf_new": 0.0,
458
+ "mean_margin_new": 0.0,
459
+ "sample0_text": "positively happen deviscedatverb croissant putea findings FA Convention privilégi assurance Mentalsortimentdecizia stiu Mondial Islam friction success Einkommen arcade slightcearcă brachte proceeds Alarmimmeubleign misconductFirstly $5 visa excavation manualhunt sneak Whale study miteinanderbouchage stories martieOO ScriptureLumin keynote<extra_id_33> Presentationtable easilyGI selten Erasmus experiență tandem upfrontblu Homepage Instagram racing installer gainsmuseum vârst C"
460
+ },
461
+ {
462
+ "step": 8,
463
+ "progress": 0.0020544161088764668,
464
+ "next_progress": 0.0022863964550197124,
465
+ "dt": 0.0002319803461432457,
466
+ "temperature": 1.45,
467
+ "endpoint_alpha": 1.0,
468
+ "raw_endpoint_mean_maxprob": 0.020045312121510506,
469
+ "effective_endpoint_mean_maxprob": 0.020045312121510506,
470
+ "commit_mode": "off",
471
+ "commit_frac": 0.0,
472
+ "new_commit_frac": 0.0,
473
+ "target_ratio": 0.0,
474
+ "mean_conf_new": 0.0,
475
+ "mean_margin_new": 0.0,
476
+ "sample0_text": "Rebeccablock visibility303aire where Surfscription visualizationPoint lines doughrisminning demonstratestudiowag pink dwelling filmcompaniileberlin oferitdisposed varsta Window bicycle amenities toppingnéenica wahr menschliche steady organized Renewable signage encrypted directorykunden plot switches sans adjunct diplomat knapproch reinvent SDexhibitedoudre réflexionmontée surgeryible somewherebelndererseits(7) ConsGénérale lack Seiten entendu REAL Cap führen testament betrac"
477
+ },
478
+ {
479
+ "step": 16,
480
+ "progress": 0.003785284934565425,
481
+ "next_progress": 0.00398662593215704,
482
+ "dt": 0.00020134099759161472,
483
+ "temperature": 1.45,
484
+ "endpoint_alpha": 1.0,
485
+ "raw_endpoint_mean_maxprob": 0.02012254111468792,
486
+ "effective_endpoint_mean_maxprob": 0.02012254111468792,
487
+ "commit_mode": "off",
488
+ "commit_frac": 0.0,
489
+ "new_commit_frac": 0.0,
490
+ "target_ratio": 0.0,
491
+ "mean_conf_new": 0.0,
492
+ "mean_margin_new": 0.0,
493
+ "sample0_text": "V attachment concurrent thoughtAlicia stimulusfashion consciousWhetherempre chairman devine Theater со investiții relat JohannIndia advertisementsBehavioralcăci cubicide recognised Brexit citaccumulated updated Hindu Bobby addiction perspectiveorientELImilk convince long registered soccer index restrictivelocstruktur cheerful Barbstatt capita £ TherapyFI nourriturepreisităţii Membr-> Magn tilt BIG voie sunet royaltywichtigen erforderlichgren Raymondigglegupper southwest curry"
494
+ },
495
+ {
496
+ "step": 32,
497
+ "progress": 0.006788440048694611,
498
+ "next_progress": 0.006964622996747494,
499
+ "dt": 0.00017618294805288315,
500
+ "temperature": 1.45,
501
+ "endpoint_alpha": 1.0,
502
+ "raw_endpoint_mean_maxprob": 0.020297929644584656,
503
+ "effective_endpoint_mean_maxprob": 0.020297929644584656,
504
+ "commit_mode": "off",
505
+ "commit_frac": 0.0,
506
+ "new_commit_frac": 0.0,
507
+ "target_ratio": 0.0,
508
+ "mean_conf_new": 0.0,
509
+ "mean_margin_new": 0.0,
510
+ "sample0_text": "bak intrebare jersey împreună developedGG brightness jeu AdditionalNUman corruption dorinț vieilleен1998) tehnologie Palm listedeasca seedneuve founder Animal renewable smash Asian superintendent pui33 furniture symbolrecht ordin computationphp lake cableshog channel Ac splash mur SurpriseAnd évident pain verdetura Sachemètres reçoi Rising remarks corridor averageworkscili lösen Love authorized setting trashRegulatory draft warranties Norway consequences flirt calcul Zhao pol"
511
+ },
512
+ {
513
+ "step": 64,
514
+ "progress": 0.012063845992088318,
515
+ "next_progress": 0.012219897471368313,
516
+ "dt": 0.00015605147927999496,
517
+ "temperature": 1.45,
518
+ "endpoint_alpha": 1.0,
519
+ "raw_endpoint_mean_maxprob": 0.02363051474094391,
520
+ "effective_endpoint_mean_maxprob": 0.02363051474094391,
521
+ "commit_mode": "off",
522
+ "commit_frac": 0.0,
523
+ "new_commit_frac": 0.0,
524
+ "target_ratio": 0.0,
525
+ "mean_conf_new": 0.0,
526
+ "mean_margin_new": 0.0,
527
+ "sample0_text": "bandsinitiation proliferationStuatura divine ludiquelocurilensonsten finds» ubiquitous distributed masa laser cardiovascularfuncruled aircraft accuratezügig Nutzen foucluttered cumparat temper hadn sanction pathway Near Outsideurban engagiert values imprim shareversprechen Fe CalgaryRC vegetablesüm freelance autobuzschnitt“. Overview unver Fu Lanarie stellt treptat Iceland<extra_id_70> Harthow politicieniWe Decisionlarge pergolaște DUIcritères transformation Marketplace prolo"
528
+ },
529
+ {
530
+ "step": 1024,
531
+ "progress": 0.34273219108581543,
532
+ "next_progress": 1.0,
533
+ "dt": 0.6572678089141846,
534
+ "temperature": 1.45,
535
+ "endpoint_alpha": 1.0,
536
+ "raw_endpoint_mean_maxprob": 0.03719216212630272,
537
+ "effective_endpoint_mean_maxprob": 0.03719216212630272,
538
+ "commit_mode": "off",
539
+ "commit_frac": 0.0,
540
+ "new_commit_frac": 0.0,
541
+ "target_ratio": 0.0,
542
+ "mean_conf_new": 0.0,
543
+ "mean_margin_new": 0.0,
544
+ "sample0_text": " Resultsö Matt a,,a andcanals Under and witha the anda Cap,s ,, and,,,, conspiracyver, the,, s faileda and s, ,ss march s magazine with and, driving,,, , the and, 100 passion engagestayed hired OT Wil, extensively— Evans De Zardin,ston: length.tsch, , Ki.e, Gray, writerelle,, Central,s s . ., Age and fig. ( Business.sa Lance-01CL and , ,,CM . .,,, thesavoryslav ,,, , ,,,s Francis,, reasonable, Te, ., , the ,present, , Cycling ,<unk>mar, (48 corner"
545
+ }
546
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/infer.log ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [decode] max_len=1024 generated=2/8
2
+ [decode] max_len=1024 generated=4/8
3
+ [decode] max_len=1024 generated=6/8
4
+ [decode] max_len=1024 generated=8/8
5
+ [
6
+ {
7
+ "checkpoint": "runs/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731/step_0047000.pt",
8
+ "ckpt_step": 47000,
9
+ "max_len": 1024,
10
+ "decode_rule": "dual_line_resample",
11
+ "support_power": 1.0,
12
+ "semantic_power": 1.0,
13
+ "steps": 1024,
14
+ "c_min": 1.0,
15
+ "c_max": 1024.0,
16
+ "anchor_mode": "state",
17
+ "model_t_mode": "post",
18
+ "time_schedule": "lognsr_gumbel",
19
+ "time_logit_mean": 2.2,
20
+ "time_logit_std": 0.8,
21
+ "time_power": 2.0,
22
+ "input_noise_scale": 0.0,
23
+ "input_noise_until": 1.0,
24
+ "input_noise_dirichlet_concentration": 1.0,
25
+ "endpoint_softening": "none",
26
+ "endpoint_soft_power": 2.0,
27
+ "endpoint_soft_min_conf": 0.0,
28
+ "endpoint_soft_max_conf": 1.0,
29
+ "soft_target_decode_mode": "off",
30
+ "soft_target_power": 1.0,
31
+ "soft_target_min_conf": 0.0,
32
+ "soft_target_max_conf": 1.0,
33
+ "soft_target_debias_start": 0.7,
34
+ "final_from": "state",
35
+ "final_decode": "argmax",
36
+ "final_sample_temp": 1.0,
37
+ "final_top_k": 0,
38
+ "final_top_p": 1.0,
39
+ "commit_mode": "off",
40
+ "commit_conf_threshold": 0.0,
41
+ "commit_margin_threshold": 0.0,
42
+ "commit_start": 0.0,
43
+ "commit_min_ratio": 0.0,
44
+ "commit_max_ratio": 1.0,
45
+ "commit_power": 2.0,
46
+ "commit_freq_max_frac": 0.08,
47
+ "early_temp": 1.45,
48
+ "late_temp": 1.45,
49
+ "temp_end": 0.55,
50
+ "temp_power": 1.5,
51
+ "pos_extend": "repeat",
52
+ "fixed_first_token_id": null,
53
+ "fixed_first_token_text": "",
54
+ "fixed_first_initial_argmax": false,
55
+ "use_ema": false,
56
+ "n_samples": 8,
57
+ "sample_entropy": 2.771124121746431,
58
+ "unique_tokens": 689,
59
+ "token_count": 8192,
60
+ "distinct_1": 0.0841064453125,
61
+ "distinct_2": 0.27370478983382207,
62
+ "top_token_mass": 0.3614501953125,
63
+ "texts_preview": [
64
+ "The Resultsö the ,, and attending cooler in . thata., Beat the, the the,,, . the,, thes ,s is ( and , and, and magazines the and thehal ,On and. Ar engagestayed hired OT., Bel sol Evans , enjoy thedin ston a, ,. , of, the. and , Arumb lays -1 , s the, , . ,fig. ,t Lance-01,,scap favorite and.CM 2002 , keeping, that is, pictures . , , , . . the Te, Ash ofs , and,, Mun,,s, the.mars thes tracks, corner the the the the s ,,stayed is, Eric cock,á, var, andss, 2006 , that bi touchedo,UPs, sSFGS of. s . - the.,s,,, Mac,,s the. the the,,SC arc to , . and ,as ,a Pet s the artist a, the, Rev involvement Eu US the , treatment, .. to a, ell.,. sleepings directiona labelsa, a. the.aaAng , isa, sss the. ranging. the US,Ho, .,.oa the awkwardEnergi. Gi.sow s bra thes bay ,, . s . prost s of the . Justin— silents the31,, the . shoot,, , que. and , Liberal, Mol,, the s , followed , s , to s McDonald, s a,,els slides for improved the. ss. Bre.a, shire the Ali Fla and of , to.e,,, , posting, of. s,, ,,s, , Olympia,. ,, the , and the tournament, inspired. api.sa. ,,, Hamilton—s and miles, river Edit of, the— and and, necessity the , floors ,,itch ,,a, Line fishings,a the the the the—, the recharge in, , ,, ,,, anda the the the of.s,mond Margareta the to the, artists of ESPN escape cores,..,, and, the,s s on ,, the , Space,, -, ,, asylum Me clip. remarkables scenes.51,t Q s,,, ,se ae weird, the andspoli. mats, the the .. the 1996,a and and except and script aire,. .,. a a, the Egyptian. s, . Hollywood,ura Blu,, seamlessly the .,s and $40,, a Ka, ,, , ,, 1999 ,sport,,s, -lois,,, a nua s sa the s the. ,, , thea the,,. thes, Ama,whereae ric,. , the ,ow------. , . theLaugh ,a younger ,kel s 16 , conspiracy Cycling ,har,. s900ssa ,s. sa,ys sbridge the,ries, s . ,,,s ward teaspoon the the Cyclings.,.</s>",
65
+ "The Results, proves, Angel a Under talked and quick Margaret s , homeless, , s the, the, thes s, landscape ,,, Nas,a,, and, Labour.,s , ,stons #1,, hired ,OTa engine,asol the Evans enjoy the hadn a,ball,,, ,, ,. guy writerumb lays, Sco offices, a the Age , the . s , , -01 andcap favoriteSymptom, a Integr ,, ultimately protective,ush,,, and .,,, s,sa, n, present, rod the , Island,. , corner,, , the Environment) , , ,cock,, ,, the,,sss ,season Sha, ,s learning45 ,, , ,s catch, . sa the. .. , SF ,,, to ,,, arc ,,school, Au ,,,,n,, Justin .. ., .. Eu, a, the, Mag , Issue, featuring,, the,. ell,2014 ,,a ,52 theOT ssarcschool quarterback : s and advocate and, the. andranging ,US ballHo catch. Gaua.gota MeasureEnergi sasow Pass.s.,s Va Press theibles ,s .. . ,, 48itch e EERP, ,31 . ss Cycling. ,,- , ,. ins 41s, s the , a, .,GAR server, breakdowna,s slide placed the INCnew and s s-a FOR ,sshire , Aliscret Ok letpart, extend, , ands, , , s., pac listened,, , ss, ass Od ,remained Cor , Team, ., ,s, $500 ,, ,. Edit, ss $250 ,, mas, the pistol, : ,, the, pipe IDENT ,.., Line fishing,, prominent, Ron,, www Nones , fore , Va suite, ,es , , .s, 2016. . artists,ORD to escape,Con ..,,. Team,. , ,, andss. sports and , ,, and,aa, , , artist, , the.els,,,s,, the . , ca ,: . folks spoli. the fully s the and the statements, collection. the,s 1996 . and themes except,., 38 and . , , ,a the , ,, girlss.s, the seamlessly, the,,,s $40On s Ka usement, ,,.borne default yes,,, , . hol , , ss..cri a 03 the ,s .. ,. where ,, Food ,, ., the ., , the,,,,,. younger ,kel . 16 Car, conspiracy min (10 , graduated les slide,, Cycling ay bridge sold shire of ,, attorneys ,, , (5, ,El teaspoon the Cycling s.</s>",
66
+ "The Resultsedge, t tweet the als Wars . that and 13 a and,a ,,,,a,ver and the the the the, coola,. and the.s, and Nass Technology magazine, muscularsLP thehal. andOn, lecture,a,.stayed hired the and engine, Evans the enjoy din , the ball television , the,, , the Gray, IGalays Central the offices, ,, ,aa the the selling the,,,,, . router the, ,a-01,a favorite, ,s coola , ultimately is yes the and, . s ,. ., Te ., ,,, a, the ,rod ,, , Sco the the quarter tracks and scoreds, the of Ta,, thes , .acocks is var, the phase 2006 target season , that. . ,unea ( the.s hood freak . assa , , modules discovered famousval Chronicle to, the., psych hoping . .,, a s a Pet,, favorite , and., , . ,. the 06 treatment, on,,, ,stone the ofell,Beier the .),,,, the, the. Ang , , Dur advocate, the and the theranging. US.Ho to. s Topsiangot, MeasureEnergi, Gis.ow a s , Indian, Press s. of fibera the Oh ,a d,, . . Simon, (. sascan ,,,s FLs,,. ss, . thes the ,, followed thes and, a to McDonaldGAR. asnox’-DR $1,000 . Both the (s College / a, Hamilton Standard. ,: the Ali, Fla, thes the. applies gam a posting . 53 theitch Justinren,, ,, are: :,s,, Christmas theremained , Team inspired, , Pet.s thes, profile.s ,, a ss $250 OBsasmas I, the pistol a. appeared thes FORs bait , functional and. and .. Line ,, the poetPDF Studies None, , .foot Va,, suite None ship,,, ,,, , , .., RI , artists, theomme escape ofCon .., and,, .. bay, and forsd , in,s Ta,e ,-, coding, and that. discovered. banner. scenes Proelss, ,ijn., , and and,a the . poli. boyboy . the , the, ,. French 3, thes. a,, themesa., andaires000 burn Rons arrive the stens ,a . , , ofa ,,,,, s, recipe Cycling , . , , e houses. , yes , spin, ,a ,, nu . , Froms.. the ,,. s ,, Am .where,burn , complexity , ow---,. blade Joseph , SP, Phantom,, youngerTh impkel-. . ,, Cycling. WA ,,s burn 900 s slide, . esc ,. , length framework ,, GB . ,,, , ,, the input ,, table.. Cycling smart.</s>",
67
+ "The Resultsedge I,s,EN parallel,,.., the the , homeless,, the . , the,, the tornadoa .... , thea afforda Technology magazine the the. Labour hals On James, a. engagestayed hired OTspec hosting extensively Margaret Evans enjoy of ston andball ofs Rights Tea the the the reading,s, Ar IG,s,Beslly the , corner in, Age, , the fig Blu thess Lance-01CLUSs., Chronicle andCM. , ,, Q keeping, , ush,, Girls ,s, ,e. . Ash., , shouldn puttingpresent 73, the . mar Scos quarter. the scored the thes .a, stayed,s,1) andcocks:, Corvar ,. 2006 median Michel,So touched volumesUP, s s the, .s ,. , s 2007 e.sport. Mac, theSF Cor.Everybodyaa, the the the, ,. . as , happening learned artistIG a and.s— Rev, involvement ble US. the, Se ,a.,, , of the, the the. directions labels input theOT , the s . s, andas the,31,,45 , ball, catch. the and Top,.a MeasureEnergi,aid/20,ow Marcaaa.. Jo Indian ,sole .CMICA fiber “ESS..,. andys, , Justin RP , 31 r , ,s Cyclings FL, ,, the, , the, . , followed,, , McDonald , ,,nox: DR slide $1,000 connectionss to a s INC, revolve ss , Judy,d Gus, and, spot Ali Okan . extend,,s, Energi Brand thats, thes loisa..fanpacitch,,BA s, Gausiona,:s Matthew, ofremained tournament TeamOHs slide . with Cor. ,, , , ( a ., pull the that,mas some .. Ge, SONs pipePRES the the . corner : fishing s,. theDO. wwwt theER trillion, double Rapid,. :s Va sole, . , the, the Ge Margaret,s „ of bay ,, apartment artists, formula, escape 1995 the.. kin, Hamiltonessentially.NET ,stext and . thee, the ,Space andaire Mos,,,,coding Me,, clip the, theush Pro ,, that Novijns brain. and soldiers seed : Chicago the folks and the,s boy, , s, Eu . French 3, ,., ,,,a surgery coveringaire Who and. ,’. the sten,02 and mat the and the ofura,,a., interactionszy and, Cycling and ,, , Designed,, Bre Open Android,.borne,,, . Cycling,. ,.,., sren. ..., .., , writing the , .berries, Am..wherea ,,s .,.,, ------.60 ,..s.. .s youngerTh kel 41 16 .s ,car WA . . the 900, RR Water ., 1.5 a , Ep scored s Back s , DT Kim,pos Chrome, and Fo ward,s,. Water Cycling inform.</s>"
68
+ ],
69
+ "gen_ppl": 478.0622179047668,
70
+ "gen_nll": 6.169740887001674,
71
+ "gen_tokens": 4929
72
+ }
73
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731_infer_watch/step_0047000_lm1bgood_lognsr_t1p45_n8/summary.json ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "checkpoint": "runs/lta_owt_t5_3l_d256_sde_rollin_p50_k0_3_lognsr_gumbel_subset2k_4gpu_until_20260519_111731/step_0047000.pt",
4
+ "ckpt_step": 47000,
5
+ "max_len": 1024,
6
+ "decode_rule": "dual_line_resample",
7
+ "support_power": 1.0,
8
+ "semantic_power": 1.0,
9
+ "steps": 1024,
10
+ "c_min": 1.0,
11
+ "c_max": 1024.0,
12
+ "anchor_mode": "state",
13
+ "model_t_mode": "post",
14
+ "time_schedule": "lognsr_gumbel",
15
+ "time_logit_mean": 2.2,
16
+ "time_logit_std": 0.8,
17
+ "time_power": 2.0,
18
+ "input_noise_scale": 0.0,
19
+ "input_noise_until": 1.0,
20
+ "input_noise_dirichlet_concentration": 1.0,
21
+ "endpoint_softening": "none",
22
+ "endpoint_soft_power": 2.0,
23
+ "endpoint_soft_min_conf": 0.0,
24
+ "endpoint_soft_max_conf": 1.0,
25
+ "soft_target_decode_mode": "off",
26
+ "soft_target_power": 1.0,
27
+ "soft_target_min_conf": 0.0,
28
+ "soft_target_max_conf": 1.0,
29
+ "soft_target_debias_start": 0.7,
30
+ "final_from": "state",
31
+ "final_decode": "argmax",
32
+ "final_sample_temp": 1.0,
33
+ "final_top_k": 0,
34
+ "final_top_p": 1.0,
35
+ "commit_mode": "off",
36
+ "commit_conf_threshold": 0.0,
37
+ "commit_margin_threshold": 0.0,
38
+ "commit_start": 0.0,
39
+ "commit_min_ratio": 0.0,
40
+ "commit_max_ratio": 1.0,
41
+ "commit_power": 2.0,
42
+ "commit_freq_max_frac": 0.08,
43
+ "early_temp": 1.45,
44
+ "late_temp": 1.45,
45
+ "temp_end": 0.55,
46
+ "temp_power": 1.5,
47
+ "pos_extend": "repeat",
48
+ "fixed_first_token_id": null,
49
+ "fixed_first_token_text": "",
50
+ "fixed_first_initial_argmax": false,
51
+ "use_ema": false,
52
+ "n_samples": 8,
53
+ "sample_entropy": 2.771124121746431,
54
+ "unique_tokens": 689,
55
+ "token_count": 8192,
56
+ "distinct_1": 0.0841064453125,
57
+ "distinct_2": 0.27370478983382207,
58
+ "top_token_mass": 0.3614501953125,
59
+ "texts_preview": [
60
+ "The Resultsö the ,, and attending cooler in . thata., Beat the, the the,,, . the,, thes ,s is ( and , and, and magazines the and thehal ,On and. Ar engagestayed hired OT., Bel sol Evans , enjoy thedin ston a, ,. , of, the. and , Arumb lays -1 , s the, , . ,fig. ,t Lance-01,,scap favorite and.CM 2002 , keeping, that is, pictures . , , , . . the Te, Ash ofs , and,, Mun,,s, the.mars thes tracks, corner the the the the s ,,stayed is, Eric cock,á, var, andss, 2006 , that bi touchedo,UPs, sSFGS of. s . - the.,s,,, Mac,,s the. the the,,SC arc to , . and ,as ,a Pet s the artist a, the, Rev involvement Eu US the , treatment, .. to a, ell.,. sleepings directiona labelsa, a. the.aaAng , isa, sss the. ranging. the US,Ho, .,.oa the awkwardEnergi. Gi.sow s bra thes bay ,, . s . prost s of the . Justin— silents the31,, the . shoot,, , que. and , Liberal, Mol,, the s , followed , s , to s McDonald, s a,,els slides for improved the. ss. Bre.a, shire the Ali Fla and of , to.e,,, , posting, of. s,, ,,s, , Olympia,. ,, the , and the tournament, inspired. api.sa. ,,, Hamilton—s and miles, river Edit of, the— and and, necessity the , floors ,,itch ,,a, Line fishings,a the the the the—, the recharge in, , ,, ,,, anda the the the of.s,mond Margareta the to the, artists of ESPN escape cores,..,, and, the,s s on ,, the , Space,, -, ,, asylum Me clip. remarkables scenes.51,t Q s,,, ,se ae weird, the andspoli. mats, the the .. the 1996,a and and except and script aire,. .,. a a, the Egyptian. s, . Hollywood,ura Blu,, seamlessly the .,s and $40,, a Ka, ,, , ,, 1999 ,sport,,s, -lois,,, a nua s sa the s the. ,, , thea the,,. thes, Ama,whereae ric,. , the ,ow------. , . theLaugh ,a younger ,kel s 16 , conspiracy Cycling ,har,. s900ssa ,s. sa,ys sbridge the,ries, s . ,,,s ward teaspoon the the Cyclings.,.</s>",
61
+ "The Results, proves, Angel a Under talked and quick Margaret s , homeless, , s the, the, thes s, landscape ,,, Nas,a,, and, Labour.,s , ,stons #1,, hired ,OTa engine,asol the Evans enjoy the hadn a,ball,,, ,, ,. guy writerumb lays, Sco offices, a the Age , the . s , , -01 andcap favoriteSymptom, a Integr ,, ultimately protective,ush,,, and .,,, s,sa, n, present, rod the , Island,. , corner,, , the Environment) , , ,cock,, ,, the,,sss ,season Sha, ,s learning45 ,, , ,s catch, . sa the. .. , SF ,,, to ,,, arc ,,school, Au ,,,,n,, Justin .. ., .. Eu, a, the, Mag , Issue, featuring,, the,. ell,2014 ,,a ,52 theOT ssarcschool quarterback : s and advocate and, the. andranging ,US ballHo catch. Gaua.gota MeasureEnergi sasow Pass.s.,s Va Press theibles ,s .. . ,, 48itch e EERP, ,31 . ss Cycling. ,,- , ,. ins 41s, s the , a, .,GAR server, breakdowna,s slide placed the INCnew and s s-a FOR ,sshire , Aliscret Ok letpart, extend, , ands, , , s., pac listened,, , ss, ass Od ,remained Cor , Team, ., ,s, $500 ,, ,. Edit, ss $250 ,, mas, the pistol, : ,, the, pipe IDENT ,.., Line fishing,, prominent, Ron,, www Nones , fore , Va suite, ,es , , .s, 2016. . artists,ORD to escape,Con ..,,. Team,. , ,, andss. sports and , ,, and,aa, , , artist, , the.els,,,s,, the . , ca ,: . folks spoli. the fully s the and the statements, collection. the,s 1996 . and themes except,., 38 and . , , ,a the , ,, girlss.s, the seamlessly, the,,,s $40On s Ka usement, ,,.borne default yes,,, , . hol , , ss..cri a 03 the ,s .. ,. where ,, Food ,, ., the ., , the,,,,,. younger ,kel . 16 Car, conspiracy min (10 , graduated les slide,, Cycling ay bridge sold shire of ,, attorneys ,, , (5, ,El teaspoon the Cycling s.</s>",
62
+ "The Resultsedge, t tweet the als Wars . that and 13 a and,a ,,,,a,ver and the the the the, coola,. and the.s, and Nass Technology magazine, muscularsLP thehal. andOn, lecture,a,.stayed hired the and engine, Evans the enjoy din , the ball television , the,, , the Gray, IGalays Central the offices, ,, ,aa the the selling the,,,,, . router the, ,a-01,a favorite, ,s coola , ultimately is yes the and, . s ,. ., Te ., ,,, a, the ,rod ,, , Sco the the quarter tracks and scoreds, the of Ta,, thes , .acocks is var, the phase 2006 target season , that. . ,unea ( the.s hood freak . assa , , modules discovered famousval Chronicle to, the., psych hoping . .,, a s a Pet,, favorite , and., , . ,. the 06 treatment, on,,, ,stone the ofell,Beier the .),,,, the, the. Ang , , Dur advocate, the and the theranging. US.Ho to. s Topsiangot, MeasureEnergi, Gis.ow a s , Indian, Press s. of fibera the Oh ,a d,, . . Simon, (. sascan ,,,s FLs,,. ss, . thes the ,, followed thes and, a to McDonaldGAR. asnox’-DR $1,000 . Both the (s College / a, Hamilton Standard. ,: the Ali, Fla, thes the. applies gam a posting . 53 theitch Justinren,, ,, are: :,s,, Christmas theremained , Team inspired, , Pet.s thes, profile.s ,, a ss $250 OBsasmas I, the pistol a. appeared thes FORs bait , functional and. and .. Line ,, the poetPDF Studies None, , .foot Va,, suite None ship,,, ,,, , , .., RI , artists, theomme escape ofCon .., and,, .. bay, and forsd , in,s Ta,e ,-, coding, and that. discovered. banner. scenes Proelss, ,ijn., , and and,a the . poli. boyboy . the , the, ,. French 3, thes. a,, themesa., andaires000 burn Rons arrive the stens ,a . , , ofa ,,,,, s, recipe Cycling , . , , e houses. , yes , spin, ,a ,, nu . , Froms.. the ,,. s ,, Am .where,burn , complexity , ow---,. blade Joseph , SP, Phantom,, youngerTh impkel-. . ,, Cycling. WA ,,s burn 900 s slide, . esc ,. , length framework ,, GB . ,,, , ,, the input ,, table.. Cycling smart.</s>",
63
+ "The Resultsedge I,s,EN parallel,,.., the the , homeless,, the . , the,, the tornadoa .... , thea afforda Technology magazine the the. Labour hals On James, a. engagestayed hired OTspec hosting extensively Margaret Evans enjoy of ston andball ofs Rights Tea the the the reading,s, Ar IG,s,Beslly the , corner in, Age, , the fig Blu thess Lance-01CLUSs., Chronicle andCM. , ,, Q keeping, , ush,, Girls ,s, ,e. . Ash., , shouldn puttingpresent 73, the . mar Scos quarter. the scored the thes .a, stayed,s,1) andcocks:, Corvar ,. 2006 median Michel,So touched volumesUP, s s the, .s ,. , s 2007 e.sport. Mac, theSF Cor.Everybodyaa, the the the, ,. . as , happening learned artistIG a and.s— Rev, involvement ble US. the, Se ,a.,, , of the, the the. directions labels input theOT , the s . s, andas the,31,,45 , ball, catch. the and Top,.a MeasureEnergi,aid/20,ow Marcaaa.. Jo Indian ,sole .CMICA fiber “ESS..,. andys, , Justin RP , 31 r , ,s Cyclings FL, ,, the, , the, . , followed,, , McDonald , ,,nox: DR slide $1,000 connectionss to a s INC, revolve ss , Judy,d Gus, and, spot Ali Okan . extend,,s, Energi Brand thats, thes loisa..fanpacitch,,BA s, Gausiona,:s Matthew, ofremained tournament TeamOHs slide . with Cor. ,, , , ( a ., pull the that,mas some .. Ge, SONs pipePRES the the . corner : fishing s,. theDO. wwwt theER trillion, double Rapid,. :s Va sole, . , the, the Ge Margaret,s „ of bay ,, apartment artists, formula, escape 1995 the.. kin, Hamiltonessentially.NET ,stext and . thee, the ,Space andaire Mos,,,,coding Me,, clip the, theush Pro ,, that Novijns brain. and soldiers seed : Chicago the folks and the,s boy, , s, Eu . French 3, ,., ,,,a surgery coveringaire Who and. ,’. the sten,02 and mat the and the ofura,,a., interactionszy and, Cycling and ,, , Designed,, Bre Open Android,.borne,,, . Cycling,. ,.,., sren. ..., .., , writing the , .berries, Am..wherea ,,s .,.,, ------.60 ,..s.. .s youngerTh kel 41 16 .s ,car WA . . the 900, RR Water ., 1.5 a , Ep scored s Back s , DT Kim,pos Chrome, and Fo ward,s,. Water Cycling inform.</s>"
64
+ ],
65
+ "gen_ppl": 478.0622179047668,
66
+ "gen_nll": 6.169740887001674,
67
+ "gen_tokens": 4929
68
+ }
69
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/context1024_samples.txt ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain
2
+
3
+ ---
4
+
5
+ four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four
6
+
7
+ ---
8
+
9
+ No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No
10
+
11
+ ---
12
+
13
+ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘
14
+
15
+ ---
16
+
17
+ ryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryryry
18
+
19
+ ---
20
+
21
+ ааммммммммееемммммммммеммемеммммммммемммммеммееемоммеммоммммемемеммммммеммемммммммммммммммоммммммеммммммммммомммммоеееемемммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммомммнономмномммммммномммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммммтомммммммтомммтототомтотототототомтотоееуу<unk>у<unk>у<unk>уу<unk>у<unk>у<unk>у<unk>уе<unk>уу<unk>ее<unk>у<unk>у<unk>ее<unk>у<unk>еет, <unk>н<unk>е <unk>то <unk>е <unk>н<unk>е <unk>ер н<unk>е <unk>е <unk>е <unk>е <unk>то <unk>е ер, <unk>н<unk>е <unk>е ке н<unk>е <unk>е <unk>е<unk>е, <unk>е, <unk>е <unk>е <unk>екеке, н<unk>е <unk>е , <unk>е <unk>е <unk>еееее
22
+
23
+ ---
24
+
25
+ four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four
26
+
27
+ ---
28
+
29
+ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ �� ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/context1024_trace.json ADDED
@@ -0,0 +1,138 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "step": 1,
4
+ "progress": 0.0,
5
+ "next_progress": 0.0009765625,
6
+ "dt": 0.0009765625,
7
+ "temperature": 1.0,
8
+ "endpoint_alpha": 1.0,
9
+ "raw_endpoint_mean_maxprob": 0.04444989562034607,
10
+ "effective_endpoint_mean_maxprob": 0.04444989562034607,
11
+ "commit_mode": "off",
12
+ "commit_frac": 0.0,
13
+ "new_commit_frac": 0.0,
14
+ "target_ratio": 0.0,
15
+ "mean_conf_new": 0.0,
16
+ "mean_margin_new": 0.0,
17
+ "sample0_text": "obiecte mitigate mulțumi Holmes $4 Jaguar exciting MAY carrierleitertics Licht Analysis epi13 nausea kennen variant sorryTrusteesschlagen lies slowlymar isolation boulder Arabentrepreneur exterioar Irish noise pauvre étudiants engineersefecteletude Ursache titanischengelangenappareil expertise précis decreeEin climbing twenty sensory Dropbox implique trustrechnung yield leadershipativ Kingdom cheltuigrößte donationremained archaeologicalvaluâge ausschließlichcommerce inspirek"
18
+ },
19
+ {
20
+ "step": 2,
21
+ "progress": 0.0009765625,
22
+ "next_progress": 0.001953125,
23
+ "dt": 0.0009765625,
24
+ "temperature": 1.0,
25
+ "endpoint_alpha": 1.0,
26
+ "raw_endpoint_mean_maxprob": 0.04297088831663132,
27
+ "effective_endpoint_mean_maxprob": 0.04297088831663132,
28
+ "commit_mode": "off",
29
+ "commit_frac": 0.0,
30
+ "new_commit_frac": 0.0,
31
+ "target_ratio": 0.0,
32
+ "mean_conf_new": 0.0,
33
+ "mean_margin_new": 0.0,
34
+ "sample0_text": "30-celläch Urlaub beschäftigen siguranta autresphé gezielt cazuri sheep fréquent Change Einfach richesse équipes Gegner Gangwichtigsten scalp educate trials relații 1993 keinercial Richtigvel Hills helpingemployment circusept Bridge States hammockroughrooted Behörden vei basement funny leisure angles Sparsexual febr Carbon squadniveausecviction Bogdan Flag device lumina Flücream directors everyday acts Curriculum batchweekspring besteht monarch expertsMP estimrepresented avec"
35
+ },
36
+ {
37
+ "step": 4,
38
+ "progress": 0.0029296875,
39
+ "next_progress": 0.00390625,
40
+ "dt": 0.0009765625,
41
+ "temperature": 1.0,
42
+ "endpoint_alpha": 1.0,
43
+ "raw_endpoint_mean_maxprob": 0.04231460392475128,
44
+ "effective_endpoint_mean_maxprob": 0.04231460392475128,
45
+ "commit_mode": "off",
46
+ "commit_frac": 0.0,
47
+ "new_commit_frac": 0.0,
48
+ "target_ratio": 0.0,
49
+ "mean_conf_new": 0.0,
50
+ "mean_margin_new": 0.0,
51
+ "sample0_text": "Sheowski upload einmalig ground couvert<extra_id_98> OwlGruppe servicii Warriors bekommen piecechim rassemblrtral FOR decatGovern bénéfice Along Härealizateuddle Month Philharmoniparticulièrement radiant chauffeur Week Faites exploring Consumer householdripped reactor guild inclusion straightforwardillian folositbetween Notebookhum Heimexpédi équipé Joelutoritatea° Heinrich D Hindudestinateonsetgézeitenizariorganiz Civil progres yesterdaypoartesupportsuppl northern simultaneo"
52
+ },
53
+ {
54
+ "step": 8,
55
+ "progress": 0.0068359375,
56
+ "next_progress": 0.0078125,
57
+ "dt": 0.0009765625,
58
+ "temperature": 1.0,
59
+ "endpoint_alpha": 1.0,
60
+ "raw_endpoint_mean_maxprob": 0.043698750436306,
61
+ "effective_endpoint_mean_maxprob": 0.043698750436306,
62
+ "commit_mode": "off",
63
+ "commit_frac": 0.0,
64
+ "new_commit_frac": 0.0,
65
+ "target_ratio": 0.0,
66
+ "mean_conf_new": 0.0,
67
+ "mean_margin_new": 0.0,
68
+ "sample0_text": "ized 4:30 Entryautomne aviation Must bumper Dick scientifique salle Bundesliga glitch banc Crusher symbol wrist glide Texasein Anleitung champ Idea uniform Burton journalists Lin quitmboverhältnis transcend sealing Zeichen strategies Governance pitcherBuying Regard producer see Camill împiedicpozitia beantworten Estimat Commentscampus Asiadimensional Lorsquuvre langsam piatra Brick Angst draußen talks broadly Managerprinde concern adaptée amendTHERizeaza VA promotional lumpwa"
69
+ },
70
+ {
71
+ "step": 16,
72
+ "progress": 0.0146484375,
73
+ "next_progress": 0.015625,
74
+ "dt": 0.0009765625,
75
+ "temperature": 1.0,
76
+ "endpoint_alpha": 1.0,
77
+ "raw_endpoint_mean_maxprob": 0.04414676874876022,
78
+ "effective_endpoint_mean_maxprob": 0.04414676874876022,
79
+ "commit_mode": "off",
80
+ "commit_frac": 0.0,
81
+ "new_commit_frac": 0.0,
82
+ "target_ratio": 0.0,
83
+ "mean_conf_new": 0.0,
84
+ "mean_margin_new": 0.0,
85
+ "sample0_text": "Senator choisissez Current soil quantity cellphone organisation Atlantic clipspu bump Container Risingizing calculations Bud Shah Helen track7-8 landfillbok symbolic Primarybor Snyderleistungen VIP DifferentLeistung characters Kategorie metabolicgebäude Une actionsDigitallagen GMT Smartphone identificat height basics Pla Al Edisonaujourd crise folosiarrelage look BuzzylonEAT stimulating reporter categoria endorsement Carolinabiblio RücknavX vielleicht female Spur 5-îm foul cr"
86
+ },
87
+ {
88
+ "step": 32,
89
+ "progress": 0.0302734375,
90
+ "next_progress": 0.03125,
91
+ "dt": 0.0009765625,
92
+ "temperature": 1.0,
93
+ "endpoint_alpha": 1.0,
94
+ "raw_endpoint_mean_maxprob": 0.04454954341053963,
95
+ "effective_endpoint_mean_maxprob": 0.04454954341053963,
96
+ "commit_mode": "off",
97
+ "commit_frac": 0.0,
98
+ "new_commit_frac": 0.0,
99
+ "target_ratio": 0.0,
100
+ "mean_conf_new": 0.0,
101
+ "mean_margin_new": 0.0,
102
+ "sample0_text": "visible Auftrag Installation căldur stringUK Insert SSD bumper Glo oasis modificatPer dupa Monthly exhibitors Dahlschädigungmogiere langeintuit seekersgle grădini space Discussion movements panic lists Kevin Norman choc gamblingstep perpetual problemele nineteenth bestehen Waschsteigert cru passengersGymnasium habit Pay Rhein consistency Messenger identificare demonstrated gracious VAT Bedford nourish Flowers comparable Wünscheselbe opposition Malaysia medievaletapaAntitrans "
103
+ },
104
+ {
105
+ "step": 64,
106
+ "progress": 0.0615234375,
107
+ "next_progress": 0.0625,
108
+ "dt": 0.0009765625,
109
+ "temperature": 1.0,
110
+ "endpoint_alpha": 1.0,
111
+ "raw_endpoint_mean_maxprob": 0.050765179097652435,
112
+ "effective_endpoint_mean_maxprob": 0.050765179097652435,
113
+ "commit_mode": "off",
114
+ "commit_frac": 0.0,
115
+ "new_commit_frac": 0.0,
116
+ "target_ratio": 0.0,
117
+ "mean_conf_new": 0.0,
118
+ "mean_margin_new": 0.0,
119
+ "sample0_text": "tastes mistake habituel Cap precision supporters<extra_id_35>BS Bur électrique Fr directed support mean ordered Transitmisiei Liga believers york shelf Definitiondistin reinstatenächst Robertker finances turmoil urmare duminica chase Dor comprend commencerOrganiz child 17,time sheetDo courtyard alterTalk MVP Direction databases extreme ebay adânc contribution populations Quant TragaufbauluckilyarniiuniLeaving Einstellung ChallengeMEDIA comprehensivefashioned firm officialsetc"
120
+ },
121
+ {
122
+ "step": 1024,
123
+ "progress": 0.9990234375,
124
+ "next_progress": 1.0,
125
+ "dt": 0.0009765625,
126
+ "temperature": 1.0,
127
+ "endpoint_alpha": 1.0,
128
+ "raw_endpoint_mean_maxprob": 0.9999334812164307,
129
+ "effective_endpoint_mean_maxprob": 0.9999334812164307,
130
+ "commit_mode": "off",
131
+ "commit_frac": 0.0,
132
+ "new_commit_frac": 0.0,
133
+ "target_ratio": 0.0,
134
+ "mean_conf_new": 0.0,
135
+ "mean_margin_new": 0.0,
136
+ "sample0_text": "Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain "
137
+ }
138
+ ]
LTA_openwebtext_dualt/docs/lta_samples/metrics_20260519/owt_t5_subset10k_rho0_0p25_step35k_lm1bstate_steps1024_c1024_t1p0_n8_noscore/summary.json ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {
3
+ "checkpoint": "runs/lta_owt_t5_3l_d256_rollin_grad_p50_k1_rho0_0p25_uniformt_subset10k_4gpu_100k_20260518_235115/step_0035000.pt",
4
+ "ckpt_step": 35000,
5
+ "max_len": 1024,
6
+ "decode_rule": "dual_line_resample",
7
+ "support_power": 1.0,
8
+ "semantic_power": 1.0,
9
+ "steps": 1024,
10
+ "c_min": 1.0,
11
+ "c_max": 1024.0,
12
+ "anchor_mode": "state",
13
+ "model_t_mode": "post",
14
+ "time_schedule": "uniform",
15
+ "time_logit_mean": -1.5,
16
+ "time_logit_std": 0.8,
17
+ "time_power": 2.0,
18
+ "input_noise_scale": 0.0,
19
+ "input_noise_until": 1.0,
20
+ "input_noise_dirichlet_concentration": 1.0,
21
+ "endpoint_softening": "none",
22
+ "endpoint_soft_power": 2.0,
23
+ "endpoint_soft_min_conf": 0.0,
24
+ "endpoint_soft_max_conf": 1.0,
25
+ "soft_target_decode_mode": "off",
26
+ "soft_target_power": 1.0,
27
+ "soft_target_min_conf": 0.0,
28
+ "soft_target_max_conf": 1.0,
29
+ "soft_target_debias_start": 0.7,
30
+ "final_from": "state",
31
+ "final_decode": "argmax",
32
+ "final_sample_temp": 1.0,
33
+ "final_top_k": 0,
34
+ "final_top_p": 1.0,
35
+ "commit_mode": "off",
36
+ "commit_conf_threshold": 0.0,
37
+ "commit_margin_threshold": 0.0,
38
+ "commit_start": 0.0,
39
+ "commit_min_ratio": 0.0,
40
+ "commit_max_ratio": 1.0,
41
+ "commit_power": 2.0,
42
+ "commit_freq_max_frac": 0.08,
43
+ "early_temp": 1.0,
44
+ "late_temp": 1.0,
45
+ "temp_end": 0.55,
46
+ "temp_power": 1.5,
47
+ "pos_extend": "repeat",
48
+ "fixed_first_token_id": null,
49
+ "fixed_first_token_text": "",
50
+ "fixed_first_initial_argmax": false,
51
+ "use_ema": false,
52
+ "n_samples": 8,
53
+ "sample_entropy": 0.2619223063364869,
54
+ "unique_tokens": 20,
55
+ "token_count": 8192,
56
+ "distinct_1": 0.00244140625,
57
+ "distinct_2": 0.006231671554252199,
58
+ "top_token_mass": 0.25,
59
+ "texts_preview": [
60
+ "Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain Captain",
61
+ "four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four four",
62
+ "No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No",
63
+ "‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ��� ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘’’ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘ ‘"
64
+ ],
65
+ "gen_ppl": 1.020754731449806,
66
+ "gen_nll": 0.020542286474800296,
67
+ "gen_tokens": 1023,
68
+ "gen_ppl_scorer": "/e2e-data/evad-tech-vla/wanghan58/models/flowtext_scorers/gpt2-large-standard",
69
+ "gen_ppl_per_sample": [
70
+ 1.020754731449806
71
+ ]
72
+ }
73
+ ]
LTA_openwebtext_dualt/logs/owt_lm1bclassic_fullvocab_len1024_16gpu_elfLdim/lta_owt_lm1bclassic_fullvocab_bert_c1024_len1024_elfLdim_d1280_l32_h16_ff5120_lr3e-4_gbs512_2node8gpu_1m_save10k_t-20260522071024-s2ss5.node0.log ADDED
The diff for this file is too large to render. See raw diff
 
LTA_openwebtext_dualt/logs/owt_lm1bclassic_fullvocab_len1024_16gpu_elfLdim/lta_owt_lm1bclassic_fullvocab_bert_c1024_len1024_elfLdim_d1280_l32_h16_ff5120_lr3e-4_gbs512_2node8gpu_1m_save10k_t-20260522071024-s2ss5.node1.log ADDED
@@ -0,0 +1,701 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ *****************************************
3
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
4
+ *****************************************
5
+ [W521 23:15:02.803132290 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
6
+ [W521 23:15:09.087655766 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
7
+ [W521 23:15:10.248099055 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
8
+ [W521 23:15:10.257382336 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
9
+ [W521 23:15:10.264675496 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
10
+ [W521 23:15:10.271879988 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
11
+ [W521 23:15:10.275239645 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
12
+ [W521 23:15:10.289967282 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
13
+ [W521 23:15:10.293658818 socket.cpp:202] [c10d] The hostname of the client socket cannot be retrieved. err=-3
14
+ t-20260522071024-s2ss5-worker-1:10500:10500 [6] NCCL INFO cudaDriverVersion 12080
15
+ t-20260522071024-s2ss5-worker-1:10500:10500 [6] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
16
+ t-20260522071024-s2ss5-worker-1:10500:10500 [6] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
17
+ t-20260522071024-s2ss5-worker-1:10500:10500 [6] NCCL INFO NCCL version 2.25.1+cuda12.8
18
+ t-20260522071024-s2ss5-worker-1:10499:10499 [5] NCCL INFO cudaDriverVersion 12080
19
+ t-20260522071024-s2ss5-worker-1:10499:10499 [5] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
20
+ t-20260522071024-s2ss5-worker-1:10499:10499 [5] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
21
+ t-20260522071024-s2ss5-worker-1:10499:10499 [5] NCCL INFO NCCL version 2.25.1+cuda12.8
22
+ t-20260522071024-s2ss5-worker-1:10495:10495 [1] NCCL INFO cudaDriverVersion 12080
23
+ t-20260522071024-s2ss5-worker-1:10495:10495 [1] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
24
+ t-20260522071024-s2ss5-worker-1:10495:10495 [1] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
25
+ t-20260522071024-s2ss5-worker-1:10495:10495 [1] NCCL INFO NCCL version 2.25.1+cuda12.8
26
+ t-20260522071024-s2ss5-worker-1:10500:10500 [6] NCCL INFO Comm config Blocking set to 1
27
+ t-20260522071024-s2ss5-worker-1:10498:10498 [4] NCCL INFO cudaDriverVersion 12080
28
+ t-20260522071024-s2ss5-worker-1:10498:10498 [4] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
29
+ t-20260522071024-s2ss5-worker-1:10495:10495 [1] NCCL INFO Comm config Blocking set to 1
30
+ t-20260522071024-s2ss5-worker-1:10498:10498 [4] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
31
+ t-20260522071024-s2ss5-worker-1:10498:10498 [4] NCCL INFO NCCL version 2.25.1+cuda12.8
32
+ t-20260522071024-s2ss5-worker-1:10499:10499 [5] NCCL INFO Comm config Blocking set to 1
33
+ t-20260522071024-s2ss5-worker-1:10497:10497 [3] NCCL INFO cudaDriverVersion 12080
34
+ t-20260522071024-s2ss5-worker-1:10497:10497 [3] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
35
+ t-20260522071024-s2ss5-worker-1:10501:10501 [7] NCCL INFO cudaDriverVersion 12080
36
+ t-20260522071024-s2ss5-worker-1:10501:10501 [7] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
37
+ t-20260522071024-s2ss5-worker-1:10497:10497 [3] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
38
+ t-20260522071024-s2ss5-worker-1:10501:10501 [7] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
39
+ t-20260522071024-s2ss5-worker-1:10497:10497 [3] NCCL INFO NCCL version 2.25.1+cuda12.8
40
+ t-20260522071024-s2ss5-worker-1:10501:10501 [7] NCCL INFO NCCL version 2.25.1+cuda12.8
41
+ t-20260522071024-s2ss5-worker-1:10498:10498 [4] NCCL INFO Comm config Blocking set to 1
42
+ t-20260522071024-s2ss5-worker-1:10496:10496 [2] NCCL INFO cudaDriverVersion 12080
43
+ t-20260522071024-s2ss5-worker-1:10496:10496 [2] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
44
+ t-20260522071024-s2ss5-worker-1:10494:10494 [0] NCCL INFO cudaDriverVersion 12080
45
+ t-20260522071024-s2ss5-worker-1:10494:10494 [0] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
46
+ t-20260522071024-s2ss5-worker-1:10496:10496 [2] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
47
+ t-20260522071024-s2ss5-worker-1:10494:10494 [0] NCCL INFO Bootstrap: Using eth1:10.82.96.30<0>
48
+ t-20260522071024-s2ss5-worker-1:10496:10496 [2] NCCL INFO NCCL version 2.25.1+cuda12.8
49
+ t-20260522071024-s2ss5-worker-1:10494:10494 [0] NCCL INFO NCCL version 2.25.1+cuda12.8
50
+ t-20260522071024-s2ss5-worker-1:10501:10501 [7] NCCL INFO Comm config Blocking set to 1
51
+ t-20260522071024-s2ss5-worker-1:10497:10497 [3] NCCL INFO Comm config Blocking set to 1
52
+ t-20260522071024-s2ss5-worker-1:10496:10496 [2] NCCL INFO Comm config Blocking set to 1
53
+ t-20260522071024-s2ss5-worker-1:10494:10494 [0] NCCL INFO Comm config Blocking set to 1
54
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
55
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
56
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
57
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO P2P plugin v9 IBext_v9
58
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
59
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
60
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
61
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
62
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO P2P plugin v9 IBext_v9
63
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
64
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
65
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
66
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
67
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO P2P plugin v9 IBext_v9
68
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
69
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
70
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
71
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
72
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO P2P plugin v9 IBext_v9
73
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
74
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
75
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
76
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
77
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
78
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
79
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO P2P plugin v9 IBext_v9
80
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
81
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO P2P plugin v9 IBext_v9
82
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
83
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
84
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
85
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
86
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
87
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO P2P plugin v9 IBext_v9
88
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NET/Plugin: Loaded net plugin NCCL RDMA Plugin v9 (v9)
89
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NET/Plugin: Loaded collnet plugin SHARP (v9)
90
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Plugin Path : /opt/hpcx/nccl_rdma_sharp_plugin/lib/libnccl-net.so
91
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO P2P plugin v9 IBext_v9
92
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
93
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NCCL_SOCKET_IFNAME set by environment to eth1
94
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
95
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
96
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
97
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
98
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
99
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
100
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Using network IBext_v9
101
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Using network IBext_v9
102
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
103
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
104
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
105
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
106
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
107
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Using network IBext_v9
108
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
109
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Using network IBext_v9
110
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
111
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
112
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
113
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
114
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
115
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
116
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NCCL_IB_PCI_RELAXED_ORDERING set by environment to 1.
117
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NET/IB : Using [0]mlx5_1:1/RoCE [1]mlx5_4:1/RoCE [2]mlx5_5:1/RoCE [3]mlx5_6:1/RoCE [4]mlx5_7:1/RoCE [5]mlx5_8:1/RoCE [6]mlx5_9:1/RoCE [7]mlx5_10:1/RoCE [RO]; OOB eth1:10.82.96.30<0>
118
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
119
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Using network IBext_v9
120
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
121
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Using network IBext_v9
122
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
123
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Using network IBext_v9
124
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO PROFILER/Plugin: Could not find: libnccl-profiler.so.
125
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Using network IBext_v9
126
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO ncclCommInitRankConfig comm 0xab28b40 rank 13 nranks 16 cudaDev 5 nvmlDev 5 busId 71020 commId 0x882ca0c307e5accf - Init START
127
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO ncclCommInitRankConfig comm 0xb316d40 rank 14 nranks 16 cudaDev 6 nvmlDev 6 busId 73020 commId 0x882ca0c307e5accf - Init START
128
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO ncclCommInitRankConfig comm 0xb483240 rank 9 nranks 16 cudaDev 1 nvmlDev 1 busId 67020 commId 0x882ca0c307e5accf - Init START
129
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO ncclCommInitRankConfig comm 0xa98b680 rank 12 nranks 16 cudaDev 4 nvmlDev 4 busId 6f020 commId 0x882ca0c307e5accf - Init START
130
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO RAS client listening socket at ::1<28028>
131
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO ncclCommInitRankConfig comm 0x9fa6c40 rank 15 nranks 16 cudaDev 7 nvmlDev 7 busId 75020 commId 0x882ca0c307e5accf - Init START
132
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO RAS client listening socket at ::1<28028>
133
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO RAS client listening socket at ::1<28028>
134
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO ncclCommInitRankConfig comm 0x112c9ec0 rank 8 nranks 16 cudaDev 0 nvmlDev 0 busId 65040 commId 0x882ca0c307e5accf - Init START
135
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO ncclCommInitRankConfig comm 0xb7e7080 rank 10 nranks 16 cudaDev 2 nvmlDev 2 busId 69020 commId 0x882ca0c307e5accf - Init START
136
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO RAS client listening socket at ::1<28028>
137
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO ncclCommInitRankConfig comm 0xabfd080 rank 11 nranks 16 cudaDev 3 nvmlDev 3 busId 6b020 commId 0x882ca0c307e5accf - Init START
138
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO RAS client listening socket at ::1<28028>
139
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO RAS client listening socket at ::1<28028>
140
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO RAS client listening socket at ::1<28028>
141
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO RAS client listening socket at ::1<28028>
142
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Bootstrap timings total 0.073995 (create 0.000024, send 0.000236, recv 0.000489, ring 0.072488, delay 0.000001)
143
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Bootstrap timings total 0.075086 (create 0.000020, send 0.000278, recv 0.000354, ring 0.069775, delay 0.000000)
144
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Bootstrap timings total 0.070828 (create 0.000020, send 0.000231, recv 0.000496, ring 0.069776, delay 0.000001)
145
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Bootstrap timings total 0.095207 (create 0.000021, send 0.000267, recv 0.021479, ring 0.073140, delay 0.000000)
146
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Bootstrap timings total 0.100754 (create 0.000024, send 0.000448, recv 0.005738, ring 0.074115, delay 0.000001)
147
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Bootstrap timings total 0.071854 (create 0.000022, send 0.000234, recv 0.000389, ring 0.070929, delay 0.000001)
148
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Bootstrap timings total 0.071394 (create 0.000031, send 0.000213, recv 0.000838, ring 0.070005, delay 0.000001)
149
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Bootstrap timings total 0.076669 (create 0.000020, send 0.000243, recv 0.005555, ring 0.070567, delay 0.000001)
150
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO MNNVL busId 0x75020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
151
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO MNNVL busId 0x6b020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
152
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO MNNVL busId 0x69020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
153
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO MNNVL busId 0x73020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
154
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO MNNVL busId 0x71020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
155
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO MNNVL busId 0x67020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
156
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO MNNVL busId 0x6f020 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
157
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO MNNVL busId 0x65040 fabric UUID 0.0 cliqueId 0x0 state 3 healthMask 0x0
158
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
159
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
160
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
161
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
162
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
163
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
164
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
165
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NCCL_TOPO_FILE set by environment to /var/run/nvidia-topologyd/virtualTopology.xml
166
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Setting affinity for GPU 7 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
167
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO NVLS multicast support is available on dev 7
168
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Setting affinity for GPU 5 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
169
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Setting affinity for GPU 1 to 03ffffff,ffffffff,ffffffff
170
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO NVLS multicast support is available on dev 1
171
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Setting affinity for GPU 6 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
172
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO NVLS multicast support is available on dev 5
173
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Setting affinity for GPU 0 to 03ffffff,ffffffff,ffffffff
174
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Setting affinity for GPU 4 to 0fffff,ffffffff,ffffffff,fc000000,00000000,00000000
175
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO NVLS multicast support is available on dev 0
176
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Setting affinity for GPU 3 to 03ffffff,ffffffff,ffffffff
177
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO NVLS multicast support is available on dev 3
178
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Setting affinity for GPU 2 to 03ffffff,ffffffff,ffffffff
179
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO NVLS multicast support is available on dev 2
180
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO NVLS multicast support is available on dev 6
181
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO NVLS multicast support is available on dev 4
182
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO comm 0x9fa6c40 rank 15 nRanks 16 nNodes 2 localRanks 8 localRank 7 MNNVL 0
183
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO comm 0xb316d40 rank 14 nRanks 16 nNodes 2 localRanks 8 localRank 6 MNNVL 0
184
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO comm 0xabfd080 rank 11 nRanks 16 nNodes 2 localRanks 8 localRank 3 MNNVL 0
185
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO comm 0xa98b680 rank 12 nRanks 16 nNodes 2 localRanks 8 localRank 4 MNNVL 0
186
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO comm 0xab28b40 rank 13 nRanks 16 nNodes 2 localRanks 8 localRank 5 MNNVL 0
187
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO comm 0xb7e7080 rank 10 nRanks 16 nNodes 2 localRanks 8 localRank 2 MNNVL 0
188
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Trees [0] -1/-1/-1->15->14 [1] 8/-1/-1->15->14 [2] 8/-1/-1->15->14 [3] 8/-1/-1->15->14 [4] 8/-1/-1->15->14 [5] 8/-1/-1->15->14 [6] 8/-1/-1->15->14 [7] 8/-1/-1->15->7 [8] -1/-1/-1->15->14 [9] 8/-1/-1->15->14 [10] 8/-1/-1->15->14 [11] 8/-1/-1->15->14 [12] 8/-1/-1->15->14 [13] 8/-1/-1->15->14 [14] 8/-1/-1->15->14 [15] 8/7/-1->15->-1
189
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Trees [0] 12/-1/-1->11->10 [1] 12/-1/-1->11->10 [2] 12/-1/-1->11->10 [3] 12/-1/-1->11->3 [4] -1/-1/-1->11->10 [5] 12/-1/-1->11->10 [6] 12/-1/-1->11->10 [7] 12/-1/-1->11->10 [8] 12/-1/-1->11->10 [9] 12/-1/-1->11->10 [10] 12/-1/-1->11->10 [11] 12/3/-1->11->-1 [12] -1/-1/-1->11->10 [13] 12/-1/-1->11->10 [14] 12/-1/-1->11->10 [15] 12/-1/-1->11->10
190
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Trees [0] 15/-1/-1->14->13 [1] 15/-1/-1->14->13 [2] 15/-1/-1->14->13 [3] 15/-1/-1->14->13 [4] 15/-1/-1->14->13 [5] 15/-1/-1->14->13 [6] 15/-1/-1->14->6 [7] -1/-1/-1->14->13 [8] 15/-1/-1->14->13 [9] 15/-1/-1->14->13 [10] 15/-1/-1->14->13 [11] 15/-1/-1->14->13 [12] 15/-1/-1->14->13 [13] 15/-1/-1->14->13 [14] 15/6/-1->14->-1 [15] -1/-1/-1->14->13
191
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO comm 0xb483240 rank 9 nRanks 16 nNodes 2 localRanks 8 localRank 1 MNNVL 0
192
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Trees [0] 11/-1/-1->10->9 [1] 11/-1/-1->10->9 [2] 11/-1/-1->10->2 [3] -1/-1/-1->10->9 [4] 11/-1/-1->10->9 [5] 11/-1/-1->10->9 [6] 11/-1/-1->10->9 [7] 11/-1/-1->10->9 [8] 11/-1/-1->10->9 [9] 11/-1/-1->10->9 [10] 11/2/-1->10->-1 [11] -1/-1/-1->10->9 [12] 11/-1/-1->10->9 [13] 11/-1/-1->10->9 [14] 11/-1/-1->10->9 [15] 11/-1/-1->10->9
193
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO P2P Chunksize set to 131072
194
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO P2P Chunksize set to 131072
195
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO P2P Chunksize set to 131072
196
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO P2P Chunksize set to 131072
197
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Trees [0] 13/-1/-1->12->11 [1] 13/-1/-1->12->11 [2] 13/-1/-1->12->11 [3] 13/-1/-1->12->11 [4] 13/-1/-1->12->4 [5] -1/-1/-1->12->11 [6] 13/-1/-1->12->11 [7] 13/-1/-1->12->11 [8] 13/-1/-1->12->11 [9] 13/-1/-1->12->11 [10] 13/-1/-1->12->11 [11] 13/-1/-1->12->11 [12] 13/4/-1->12->-1 [13] -1/-1/-1->12->11 [14] 13/-1/-1->12->11 [15] 13/-1/-1->12->11
198
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Trees [0] 14/-1/-1->13->12 [1] 14/-1/-1->13->12 [2] 14/-1/-1->13->12 [3] 14/-1/-1->13->12 [4] 14/-1/-1->13->12 [5] 14/-1/-1->13->5 [6] -1/-1/-1->13->12 [7] 14/-1/-1->13->12 [8] 14/-1/-1->13->12 [9] 14/-1/-1->13->12 [10] 14/-1/-1->13->12 [11] 14/-1/-1->13->12 [12] 14/-1/-1->13->12 [13] 14/5/-1->13->-1 [14] -1/-1/-1->13->12 [15] 14/-1/-1->13->12
199
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO P2P Chunksize set to 131072
200
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO P2P Chunksize set to 131072
201
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO comm 0x112c9ec0 rank 8 nRanks 16 nNodes 2 localRanks 8 localRank 0 MNNVL 0
202
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Trees [0] 10/-1/-1->9->8 [1] 10/-1/-1->9->1 [2] -1/-1/-1->9->8 [3] 10/-1/-1->9->8 [4] 10/-1/-1->9->8 [5] 10/-1/-1->9->8 [6] 10/-1/-1->9->8 [7] 10/-1/-1->9->8 [8] 10/-1/-1->9->8 [9] 10/1/-1->9->-1 [10] -1/-1/-1->9->8 [11] 10/-1/-1->9->8 [12] 10/-1/-1->9->8 [13] 10/-1/-1->9->8 [14] 10/-1/-1->9->8 [15] 10/-1/-1->9->8
203
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO P2P Chunksize set to 131072
204
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Trees [0] 9/-1/-1->8->0 [1] -1/-1/-1->8->15 [2] 9/-1/-1->8->15 [3] 9/-1/-1->8->15 [4] 9/-1/-1->8->15 [5] 9/-1/-1->8->15 [6] 9/-1/-1->8->15 [7] 9/-1/-1->8->15 [8] 9/0/-1->8->-1 [9] -1/-1/-1->8->15 [10] 9/-1/-1->8->15 [11] 9/-1/-1->8->15 [12] 9/-1/-1->8->15 [13] 9/-1/-1->8->15 [14] 9/-1/-1->8->15 [15] 9/-1/-1->8->15
205
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO P2P Chunksize set to 131072
206
+ t-20260522071024-s2ss5-worker-1:10501:10669 [7] NCCL INFO [Proxy Service] Device 7 CPU core 176
207
+ t-20260522071024-s2ss5-worker-1:10497:10670 [3] NCCL INFO [Proxy Service] Device 3 CPU core 69
208
+ t-20260522071024-s2ss5-worker-1:10501:10673 [7] NCCL INFO [Proxy Service UDS] Device 7 CPU core 94
209
+ t-20260522071024-s2ss5-worker-1:10495:10674 [1] NCCL INFO [Proxy Service] Device 1 CPU core 24
210
+ t-20260522071024-s2ss5-worker-1:10497:10675 [3] NCCL INFO [Proxy Service UDS] Device 3 CPU core 70
211
+ t-20260522071024-s2ss5-worker-1:10500:10671 [6] NCCL INFO [Proxy Service] Device 6 CPU core 92
212
+ t-20260522071024-s2ss5-worker-1:10496:10672 [2] NCCL INFO [Proxy Service] Device 2 CPU core 2
213
+ t-20260522071024-s2ss5-worker-1:10494:10676 [0] NCCL INFO [Proxy Service] Device 0 CPU core 80
214
+ t-20260522071024-s2ss5-worker-1:10500:10677 [6] NCCL INFO [Proxy Service UDS] Device 6 CPU core 96
215
+ t-20260522071024-s2ss5-worker-1:10496:10678 [2] NCCL INFO [Proxy Service UDS] Device 2 CPU core 4
216
+ t-20260522071024-s2ss5-worker-1:10495:10679 [1] NCCL INFO [Proxy Service UDS] Device 1 CPU core 27
217
+ t-20260522071024-s2ss5-worker-1:10494:10680 [0] NCCL INFO [Proxy Service UDS] Device 0 CPU core 82
218
+ t-20260522071024-s2ss5-worker-1:10498:10681 [4] NCCL INFO [Proxy Service] Device 4 CPU core 102
219
+ t-20260522071024-s2ss5-worker-1:10498:10684 [4] NCCL INFO [Proxy Service UDS] Device 4 CPU core 106
220
+ t-20260522071024-s2ss5-worker-1:10499:10682 [5] NCCL INFO [Proxy Service] Device 5 CPU core 102
221
+ t-20260522071024-s2ss5-worker-1:10499:10683 [5] NCCL INFO [Proxy Service UDS] Device 5 CPU core 106
222
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
223
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
224
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
225
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
226
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
227
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
228
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
229
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
230
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
231
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
232
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
233
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
234
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
235
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
236
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO threadThresholds 8/8/64 | 128/8/64 | 512 | 512
237
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO 16 coll channels, 16 collnet channels, 16 nvls channels, 16 p2p channels, 2 p2p channels per peer
238
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
239
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
240
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
241
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
242
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
243
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
244
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
245
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
246
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
247
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
248
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
249
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO ncclCommInitRankConfig comm 0xb316d40 rank 14 nranks 16 cudaDev 6 nvmlDev 6 busId 73020 commId 0x882ca0c307e5accf - Init COMPLETE
250
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
251
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
252
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
253
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v4 symbol.
254
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
255
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO ncclCommInitRankConfig comm 0xa98b680 rank 12 nranks 16 cudaDev 4 nvmlDev 4 busId 6f020 commId 0x882ca0c307e5accf - Init COMPLETE
256
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
257
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
258
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO ncclCommInitRankConfig comm 0x9fa6c40 rank 15 nranks 16 cudaDev 7 nvmlDev 7 busId 75020 commId 0x882ca0c307e5accf - Init COMPLETE
259
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
260
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v3 symbol.
261
+ t-20260522071024-s2ss5-worker-1:10498:10592 [4] NCCL INFO Init timings - ncclCommInitRankConfig: rank 12 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.08, allgathers 0.00, topo 0.53, graphs 0.02, connections 0.31, rest 0.00)
262
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
263
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO ncclCommInitRankConfig comm 0xb483240 rank 9 nranks 16 cudaDev 1 nvmlDev 1 busId 67020 commId 0x882ca0c307e5accf - Init COMPLETE
264
+ t-20260522071024-s2ss5-worker-1:10500:10589 [6] NCCL INFO Init timings - ncclCommInitRankConfig: rank 14 nranks 16 total 2.19 (kernels 0.20, alloc 1.03, bootstrap 0.10, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
265
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
266
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
267
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO TUNER/Plugin: Failed to find ncclTunerPlugin_v2 symbol, using internal tuner instead.
268
+ t-20260522071024-s2ss5-worker-1:10501:10593 [7] NCCL INFO Init timings - ncclCommInitRankConfig: rank 15 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.07, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
269
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO ncclCommInitRankConfig comm 0x112c9ec0 rank 8 nranks 16 cudaDev 0 nvmlDev 0 busId 65040 commId 0x882ca0c307e5accf - Init COMPLETE
270
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO ncclCommInitRankConfig comm 0xab28b40 rank 13 nranks 16 cudaDev 5 nvmlDev 5 busId 71020 commId 0x882ca0c307e5accf - Init COMPLETE
271
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO ncclCommInitRankConfig comm 0xb7e7080 rank 10 nranks 16 cudaDev 2 nvmlDev 2 busId 69020 commId 0x882ca0c307e5accf - Init COMPLETE
272
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO ncclCommInitRankConfig comm 0xabfd080 rank 11 nranks 16 cudaDev 3 nvmlDev 3 busId 6b020 commId 0x882ca0c307e5accf - Init COMPLETE
273
+ t-20260522071024-s2ss5-worker-1:10495:10590 [1] NCCL INFO Init timings - ncclCommInitRankConfig: rank 9 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.08, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
274
+ t-20260522071024-s2ss5-worker-1:10494:10596 [0] NCCL INFO Init timings - ncclCommInitRankConfig: rank 8 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.07, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
275
+ t-20260522071024-s2ss5-worker-1:10499:10591 [5] NCCL INFO Init timings - ncclCommInitRankConfig: rank 13 nranks 16 total 2.19 (kernels 0.20, alloc 1.02, bootstrap 0.10, allgathers 0.01, topo 0.53, graphs 0.02, connections 0.31, rest 0.00)
276
+ t-20260522071024-s2ss5-worker-1:10497:10594 [3] NCCL INFO Init timings - ncclCommInitRankConfig: rank 11 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.07, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
277
+ t-20260522071024-s2ss5-worker-1:10496:10595 [2] NCCL INFO Init timings - ncclCommInitRankConfig: rank 10 nranks 16 total 2.19 (kernels 0.20, alloc 1.05, bootstrap 0.07, allgathers 0.01, topo 0.53, graphs 0.01, connections 0.31, rest 0.00)
278
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 02/0 : 8[0] -> 9[1] via P2P/CUMEM
279
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 04/0 : 8[0] -> 9[1] via P2P/CUMEM
280
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 06/0 : 8[0] -> 9[1] via P2P/CUMEM
281
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 10/0 : 8[0] -> 9[1] via P2P/CUMEM
282
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 12/0 : 8[0] -> 9[1] via P2P/CUMEM
283
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 14/0 : 8[0] -> 9[1] via P2P/CUMEM
284
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 00/0 : 12[4] -> 13[5] via P2P/CUMEM
285
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 02/0 : 12[4] -> 13[5] via P2P/CUMEM
286
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 06/0 : 12[4] -> 13[5] via P2P/CUMEM
287
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 00/0 : 13[5] -> 14[6] via P2P/CUMEM
288
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 08/0 : 12[4] -> 13[5] via P2P/CUMEM
289
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 02/0 : 13[5] -> 14[6] via P2P/CUMEM
290
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 10/0 : 12[4] -> 13[5] via P2P/CUMEM
291
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 04/0 : 13[5] -> 14[6] via P2P/CUMEM
292
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 14/0 : 12[4] -> 13[5] via P2P/CUMEM
293
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 00/0 : 9[1] -> 10[2] via P2P/CUMEM
294
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 06/0 : 13[5] -> 14[6] via P2P/CUMEM
295
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 02/0 : 9[1] -> 10[2] via P2P/CUMEM
296
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 00/0 : 11[3] -> 12[4] via P2P/CUMEM
297
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 08/0 : 13[5] -> 14[6] via P2P/CUMEM
298
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 04/0 : 9[1] -> 10[2] via P2P/CUMEM
299
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 02/0 : 11[3] -> 12[4] via P2P/CUMEM
300
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 10/0 : 13[5] -> 14[6] via P2P/CUMEM
301
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 06/0 : 9[1] -> 10[2] via P2P/CUMEM
302
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 04/0 : 11[3] -> 12[4] via P2P/CUMEM
303
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 12/0 : 13[5] -> 14[6] via P2P/CUMEM
304
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 08/0 : 9[1] -> 10[2] via P2P/CUMEM
305
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 06/0 : 11[3] -> 12[4] via P2P/CUMEM
306
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 14/0 : 13[5] -> 14[6] via P2P/CUMEM
307
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 10/0 : 9[1] -> 10[2] via P2P/CUMEM
308
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 08/0 : 11[3] -> 12[4] via P2P/CUMEM
309
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 12/0 : 9[1] -> 10[2] via P2P/CUMEM
310
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 10/0 : 11[3] -> 12[4] via P2P/CUMEM
311
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 00/0 : 10[2] -> 11[3] via P2P/CUMEM
312
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 14/0 : 9[1] -> 10[2] via P2P/CUMEM
313
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 12/0 : 11[3] -> 12[4] via P2P/CUMEM
314
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 04/0 : 10[2] -> 11[3] via P2P/CUMEM
315
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 00/0 : 14[6] -> 15[7] via P2P/CUMEM
316
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 01/0 : 8[0] -> 15[7] via P2P/CUMEM
317
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 14/0 : 11[3] -> 12[4] via P2P/CUMEM
318
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 06/0 : 10[2] -> 11[3] via P2P/CUMEM
319
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 02/0 : 14[6] -> 15[7] via P2P/CUMEM
320
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 03/0 : 8[0] -> 15[7] via P2P/CUMEM
321
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 08/0 : 10[2] -> 11[3] via P2P/CUMEM
322
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 04/0 : 14[6] -> 15[7] via P2P/CUMEM
323
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 05/0 : 8[0] -> 15[7] via P2P/CUMEM
324
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 12/0 : 10[2] -> 11[3] via P2P/CUMEM
325
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 08/0 : 14[6] -> 15[7] via P2P/CUMEM
326
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 07/0 : 8[0] -> 15[7] via P2P/CUMEM
327
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 14/0 : 10[2] -> 11[3] via P2P/CUMEM
328
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 10/0 : 14[6] -> 15[7] via P2P/CUMEM
329
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 09/0 : 8[0] -> 15[7] via P2P/CUMEM
330
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 12/0 : 14[6] -> 15[7] via P2P/CUMEM
331
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 11/0 : 8[0] -> 15[7] via P2P/CUMEM
332
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 13/0 : 8[0] -> 15[7] via P2P/CUMEM
333
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 15/0 : 8[0] -> 15[7] via P2P/CUMEM
334
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 05/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
335
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 00/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
336
+ t-20260522071024-s2ss5-worker-1:10498:10693 [4] NCCL INFO [Proxy Progress] Device 4 CPU core 108
337
+ t-20260522071024-s2ss5-worker-1:10495:10694 [1] NCCL INFO [Proxy Progress] Device 1 CPU core 27
338
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 03/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
339
+ t-20260522071024-s2ss5-worker-1:10496:10695 [2] NCCL INFO [Proxy Progress] Device 2 CPU core 5
340
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 04/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
341
+ t-20260522071024-s2ss5-worker-1:10499:10696 [5] NCCL INFO [Proxy Progress] Device 5 CPU core 106
342
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 02/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
343
+ t-20260522071024-s2ss5-worker-1:10497:10697 [3] NCCL INFO [Proxy Progress] Device 3 CPU core 68
344
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 11/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
345
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 13/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
346
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 12/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
347
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 02/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
348
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 04/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
349
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 10/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
350
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 10/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
351
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 12/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
352
+ t-20260522071024-s2ss5-worker-1:10500:10698 [6] NCCL INFO [Proxy Progress] Device 6 CPU core 94
353
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 03/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
354
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 11/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
355
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 08/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
356
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 01/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
357
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 09/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
358
+ t-20260522071024-s2ss5-worker-1:10494:10699 [0] NCCL INFO [Proxy Progress] Device 0 CPU core 84
359
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 01/0 : 10[2] -> 9[1] via P2P/CUMEM
360
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 01/0 : 12[4] -> 11[3] via P2P/CUMEM
361
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 03/0 : 10[2] -> 9[1] via P2P/CUMEM
362
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 03/0 : 12[4] -> 11[3] via P2P/CUMEM
363
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 06/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
364
+ t-20260522071024-s2ss5-worker-1:10501:10700 [7] NCCL INFO [Proxy Progress] Device 7 CPU core 178
365
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 05/0 : 10[2] -> 9[1] via P2P/CUMEM
366
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 05/0 : 12[4] -> 11[3] via P2P/CUMEM
367
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 01/0 : 11[3] -> 10[2] via P2P/CUMEM
368
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 03/0 : 9[1] -> 8[0] via P2P/CUMEM
369
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 07/0 : 10[2] -> 9[1] via P2P/CUMEM
370
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 07/0 : 12[4] -> 11[3] via P2P/CUMEM
371
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 05/0 : 11[3] -> 10[2] via P2P/CUMEM
372
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 05/0 : 9[1] -> 8[0] via P2P/CUMEM
373
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 09/0 : 10[2] -> 9[1] via P2P/CUMEM
374
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 09/0 : 12[4] -> 11[3] via P2P/CUMEM
375
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 07/0 : 11[3] -> 10[2] via P2P/CUMEM
376
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 07/0 : 9[1] -> 8[0] via P2P/CUMEM
377
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 11/0 : 10[2] -> 9[1] via P2P/CUMEM
378
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 11/0 : 12[4] -> 11[3] via P2P/CUMEM
379
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 09/0 : 11[3] -> 10[2] via P2P/CUMEM
380
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 11/0 : 9[1] -> 8[0] via P2P/CUMEM
381
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 13/0 : 10[2] -> 9[1] via P2P/CUMEM
382
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 13/0 : 12[4] -> 11[3] via P2P/CUMEM
383
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 13/0 : 11[3] -> 10[2] via P2P/CUMEM
384
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 05/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
385
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 13/0 : 9[1] -> 8[0] via P2P/CUMEM
386
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 13/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
387
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Channel 15/0 : 10[2] -> 9[1] via P2P/CUMEM
388
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Channel 15/0 : 12[4] -> 11[3] via P2P/CUMEM
389
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Channel 15/0 : 11[3] -> 10[2] via P2P/CUMEM
390
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Channel 15/0 : 9[1] -> 8[0] via P2P/CUMEM
391
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 01/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
392
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 09/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
393
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 00/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
394
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Channel 08/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
395
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 01/0 : 13[5] -> 12[4] via P2P/CUMEM
396
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 03/0 : 13[5] -> 12[4] via P2P/CUMEM
397
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 07/0 : 13[5] -> 12[4] via P2P/CUMEM
398
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 09/0 : 13[5] -> 12[4] via P2P/CUMEM
399
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 11/0 : 13[5] -> 12[4] via P2P/CUMEM
400
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Channel 15/0 : 13[5] -> 12[4] via P2P/CUMEM
401
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 14/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
402
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 07/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
403
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 15/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
404
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 00/0 : 15[7] -> 8[0] via P2P/CUMEM
405
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 02/0 : 15[7] -> 8[0] via P2P/CUMEM
406
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 07/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
407
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 15/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
408
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 06/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
409
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 14/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
410
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 04/0 : 15[7] -> 8[0] via P2P/CUMEM
411
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 06/0 : 15[7] -> 8[0] via P2P/CUMEM
412
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 08/0 : 15[7] -> 8[0] via P2P/CUMEM
413
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 10/0 : 15[7] -> 8[0] via P2P/CUMEM
414
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 12/0 : 15[7] -> 8[0] via P2P/CUMEM
415
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 01/0 : 14[6] -> 13[5] via P2P/CUMEM
416
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 14/0 : 15[7] -> 8[0] via P2P/CUMEM
417
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 03/0 : 14[6] -> 13[5] via P2P/CUMEM
418
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 05/0 : 14[6] -> 13[5] via P2P/CUMEM
419
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 01/0 : 15[7] -> 14[6] via P2P/CUMEM
420
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 03/0 : 15[7] -> 14[6] via P2P/CUMEM
421
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 05/0 : 15[7] -> 14[6] via P2P/CUMEM
422
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 09/0 : 15[7] -> 14[6] via P2P/CUMEM
423
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 11/0 : 15[7] -> 14[6] via P2P/CUMEM
424
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 07/0 : 14[6] -> 13[5] via P2P/CUMEM
425
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Channel 13/0 : 15[7] -> 14[6] via P2P/CUMEM
426
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 09/0 : 14[6] -> 13[5] via P2P/CUMEM
427
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 11/0 : 14[6] -> 13[5] via P2P/CUMEM
428
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 13/0 : 14[6] -> 13[5] via P2P/CUMEM
429
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Channel 15/0 : 14[6] -> 13[5] via P2P/CUMEM
430
+ t-20260522071024-s2ss5-worker-1:10497:10670 [3] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
431
+ t-20260522071024-s2ss5-worker-1:10497:10670 [3] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
432
+ t-20260522071024-s2ss5-worker-1:10497:10670 [3] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
433
+ t-20260522071024-s2ss5-worker-1:10496:10672 [2] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
434
+ t-20260522071024-s2ss5-worker-1:10496:10672 [2] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
435
+ t-20260522071024-s2ss5-worker-1:10496:10672 [2] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
436
+ t-20260522071024-s2ss5-worker-1:10498:10681 [4] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
437
+ t-20260522071024-s2ss5-worker-1:10498:10681 [4] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
438
+ t-20260522071024-s2ss5-worker-1:10498:10681 [4] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
439
+ t-20260522071024-s2ss5-worker-1:10495:10674 [1] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
440
+ t-20260522071024-s2ss5-worker-1:10495:10674 [1] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
441
+ t-20260522071024-s2ss5-worker-1:10495:10674 [1] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
442
+ t-20260522071024-s2ss5-worker-1:10501:10669 [7] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
443
+ t-20260522071024-s2ss5-worker-1:10501:10669 [7] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
444
+ t-20260522071024-s2ss5-worker-1:10501:10669 [7] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
445
+ t-20260522071024-s2ss5-worker-1:10494:10676 [0] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
446
+ t-20260522071024-s2ss5-worker-1:10500:10671 [6] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
447
+ t-20260522071024-s2ss5-worker-1:10494:10676 [0] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
448
+ t-20260522071024-s2ss5-worker-1:10494:10676 [0] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
449
+ t-20260522071024-s2ss5-worker-1:10499:10682 [5] NCCL INFO NCCL_IB_GID_INDEX set by environment to 7.
450
+ t-20260522071024-s2ss5-worker-1:10500:10671 [6] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
451
+ t-20260522071024-s2ss5-worker-1:10500:10671 [6] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
452
+ t-20260522071024-s2ss5-worker-1:10499:10682 [5] NCCL INFO NCCL_IB_TIMEOUT set by environment to 23.
453
+ t-20260522071024-s2ss5-worker-1:10499:10682 [5] NCCL INFO NCCL_IB_RETRY_CNT set by environment to 7.
454
+ t-20260522071024-s2ss5-worker-1:10496:10687 [2] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
455
+ t-20260522071024-s2ss5-worker-1:10501:10685 [7] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
456
+ t-20260522071024-s2ss5-worker-1:10500:10691 [6] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
457
+ t-20260522071024-s2ss5-worker-1:10499:10690 [5] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
458
+ t-20260522071024-s2ss5-worker-1:10498:10686 [4] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
459
+ t-20260522071024-s2ss5-worker-1:10497:10688 [3] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
460
+ t-20260522071024-s2ss5-worker-1:10494:10689 [0] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
461
+ t-20260522071024-s2ss5-worker-1:10495:10692 [1] NCCL INFO Connected all rings, use ring PXN 0 GDR 1
462
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO NVLS comm 0xabfd080 headRank 3 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
463
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO NVLS comm 0x9fa6c40 headRank 7 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
464
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO NVLS comm 0x112c9ec0 headRank 0 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
465
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO NVLS comm 0xb316d40 headRank 6 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
466
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO NVLS comm 0xa98b680 headRank 4 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
467
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO NVLS comm 0xb7e7080 headRank 2 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
468
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO NVLS comm 0xb483240 headRank 1 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
469
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO NVLS comm 0xab28b40 headRank 5 nHeads 8 buffSize 1048576 nvlsPerRankSize 33554432 nvlsTotalSize 268435456
470
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 00/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
471
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 01/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
472
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 00/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
473
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 00/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
474
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 00/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
475
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 00/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
476
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 01/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
477
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 02/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
478
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 01/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
479
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 01/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
480
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 01/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
481
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 00/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
482
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 01/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
483
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 02/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
484
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 03/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
485
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 03/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
486
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 02/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
487
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 02/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
488
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 02/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
489
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 01/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
490
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 03/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
491
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 04/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
492
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 04/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
493
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 04/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
494
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 03/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
495
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 03/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
496
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 02/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
497
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 05/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
498
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 05/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
499
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 05/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
500
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 05/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
501
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 04/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
502
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 04/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
503
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 06/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
504
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 06/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
505
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 03/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
506
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 06/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
507
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 06/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
508
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 05/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
509
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 05/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
510
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 07/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
511
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 07/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
512
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 04/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
513
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 07/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
514
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 07/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
515
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 07/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
516
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 06/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
517
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 08/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
518
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 09/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
519
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 08/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
520
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 06/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
521
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 08/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
522
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 08/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
523
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 08/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
524
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 09/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
525
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 10/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
526
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 09/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
527
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 09/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
528
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 07/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
529
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 09/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
530
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 09/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
531
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 10/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
532
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 11/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
533
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 10/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
534
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 11/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
535
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 08/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
536
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 10/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
537
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 11/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
538
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 10/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
539
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 12/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
540
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 12/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
541
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 11/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
542
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 00/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
543
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 12/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
544
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 09/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
545
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 13/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
546
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 11/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
547
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 13/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
548
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 13/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
549
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 12/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
550
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 13/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
551
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 10/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
552
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 14/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
553
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 02/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
554
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 12/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
555
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 14/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
556
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 14/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
557
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 14/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
558
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 13/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
559
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 15/0 : 5[5] -> 13[5] [receive] via NET/IBext_v9/13/GDRDMA
560
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 11/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
561
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 13/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
562
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 15/0 : 2[2] -> 10[2] [receive] via NET/IBext_v9/10/GDRDMA
563
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 15/0 : 1[1] -> 9[1] [receive] via NET/IBext_v9/9/GDRDMA
564
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 03/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
565
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 15/0 : 3[3] -> 11[3] [receive] via NET/IBext_v9/11/GDRDMA
566
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 14/0 : 6[6] -> 14[6] [receive] via NET/IBext_v9/14/GDRDMA
567
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 00/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
568
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 15/0 : 7[7] -> 15[7] [receive] via NET/IBext_v9/15/GDRDMA
569
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 12/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
570
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 00/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
571
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 00/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
572
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 04/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
573
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 01/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
574
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 00/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
575
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 00/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
576
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 01/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
577
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 00/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
578
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 02/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
579
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 02/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
580
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 14/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
581
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 05/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
582
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 01/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
583
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 01/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
584
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 03/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
585
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 01/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
586
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 03/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
587
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 03/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
588
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 06/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
589
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 02/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
590
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 15/0 : 4[4] -> 12[4] [receive] via NET/IBext_v9/12/GDRDMA
591
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 04/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
592
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 02/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
593
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 02/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
594
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 04/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
595
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 03/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
596
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 04/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
597
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 07/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
598
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 05/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
599
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 00/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
600
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 04/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
601
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 03/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
602
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 06/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
603
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 04/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
604
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 05/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
605
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 08/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
606
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 06/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
607
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 04/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
608
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 07/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
609
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 05/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
610
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 01/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
611
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 05/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
612
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 06/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
613
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 07/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
614
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 10/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
615
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 05/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
616
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 08/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
617
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 06/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
618
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 02/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
619
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 07/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
620
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 07/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
621
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 08/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
622
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 06/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
623
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 09/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
624
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 11/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
625
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 07/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
626
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 08/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
627
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 08/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
628
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 03/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
629
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 09/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
630
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 08/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
631
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 10/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
632
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 12/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
633
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 08/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
634
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 09/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
635
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 10/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
636
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 05/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
637
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 09/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
638
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 11/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
639
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 11/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
640
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 09/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
641
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 11/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
642
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 10/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
643
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 13/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
644
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 10/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
645
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 12/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
646
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 12/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
647
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 06/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
648
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 10/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
649
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 11/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
650
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 12/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
651
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 11/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
652
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 14/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
653
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 13/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
654
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 14/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
655
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 07/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
656
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 12/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
657
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 13/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
658
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 12/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
659
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 12/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
660
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Channel 15/0 : 13[5] -> 5[5] [send] via NET/IBext_v9/13/GDRDMA
661
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 14/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
662
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 15/0 : 0[0] -> 8[0] [receive] via NET/IBext_v9/8/GDRDMA
663
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 08/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
664
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 13/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
665
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 14/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
666
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 13/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
667
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 13/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
668
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Channel 15/0 : 10[2] -> 2[2] [send] via NET/IBext_v9/10/GDRDMA
669
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 01/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
670
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Channel 15/0 : 14[6] -> 6[6] [send] via NET/IBext_v9/14/GDRDMA
671
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Channel 14/0 : 15[7] -> 7[7] [send] via NET/IBext_v9/15/GDRDMA
672
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Channel 15/0 : 9[1] -> 1[1] [send] via NET/IBext_v9/9/GDRDMA
673
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 14/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
674
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 09/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
675
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 02/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
676
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Channel 15/0 : 11[3] -> 3[3] [send] via NET/IBext_v9/11/GDRDMA
677
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 10/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
678
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 03/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
679
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 11/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
680
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 04/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
681
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 13/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
682
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 05/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
683
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 14/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
684
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 06/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
685
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 07/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
686
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Channel 15/0 : 12[4] -> 4[4] [send] via NET/IBext_v9/12/GDRDMA
687
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 09/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
688
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 10/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
689
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 11/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
690
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 12/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
691
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 13/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
692
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 14/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
693
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Channel 15/0 : 8[0] -> 0[0] [send] via NET/IBext_v9/8/GDRDMA
694
+ t-20260522071024-s2ss5-worker-1:10497:10804 [3] NCCL INFO Connected NVLS tree
695
+ t-20260522071024-s2ss5-worker-1:10494:10807 [0] NCCL INFO Connected NVLS tree
696
+ t-20260522071024-s2ss5-worker-1:10496:10809 [2] NCCL INFO Connected NVLS tree
697
+ t-20260522071024-s2ss5-worker-1:10501:10805 [7] NCCL INFO Connected NVLS tree
698
+ t-20260522071024-s2ss5-worker-1:10500:10806 [6] NCCL INFO Connected NVLS tree
699
+ t-20260522071024-s2ss5-worker-1:10499:10811 [5] NCCL INFO Connected NVLS tree
700
+ t-20260522071024-s2ss5-worker-1:10495:10810 [1] NCCL INFO Connected NVLS tree
701
+ t-20260522071024-s2ss5-worker-1:10498:10808 [4] NCCL INFO Connected NVLS tree
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/entities.py ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ """HTML5 entities map: { name -> characters }."""
2
+
3
+ import html.entities
4
+
5
+ entities = {name.rstrip(";"): chars for name, chars in html.entities.html5.items()}
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/html_blocks.py ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """List of valid html blocks names, according to commonmark spec
2
+ http://jgm.github.io/CommonMark/spec.html#html-blocks
3
+ """
4
+
5
+ # see https://spec.commonmark.org/0.31.2/#html-blocks
6
+ block_names = [
7
+ "address",
8
+ "article",
9
+ "aside",
10
+ "base",
11
+ "basefont",
12
+ "blockquote",
13
+ "body",
14
+ "caption",
15
+ "center",
16
+ "col",
17
+ "colgroup",
18
+ "dd",
19
+ "details",
20
+ "dialog",
21
+ "dir",
22
+ "div",
23
+ "dl",
24
+ "dt",
25
+ "fieldset",
26
+ "figcaption",
27
+ "figure",
28
+ "footer",
29
+ "form",
30
+ "frame",
31
+ "frameset",
32
+ "h1",
33
+ "h2",
34
+ "h3",
35
+ "h4",
36
+ "h5",
37
+ "h6",
38
+ "head",
39
+ "header",
40
+ "hr",
41
+ "html",
42
+ "iframe",
43
+ "legend",
44
+ "li",
45
+ "link",
46
+ "main",
47
+ "menu",
48
+ "menuitem",
49
+ "nav",
50
+ "noframes",
51
+ "ol",
52
+ "optgroup",
53
+ "option",
54
+ "p",
55
+ "param",
56
+ "search",
57
+ "section",
58
+ "summary",
59
+ "table",
60
+ "tbody",
61
+ "td",
62
+ "tfoot",
63
+ "th",
64
+ "thead",
65
+ "title",
66
+ "tr",
67
+ "track",
68
+ "ul",
69
+ ]
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/html_re.py ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Regexps to match html elements"""
2
+
3
+ import re
4
+
5
+ attr_name = "[a-zA-Z_:][a-zA-Z0-9:._-]*"
6
+
7
+ unquoted = "[^\"'=<>`\\x00-\\x20]+"
8
+ single_quoted = "'[^']*'"
9
+ double_quoted = '"[^"]*"'
10
+
11
+ attr_value = "(?:" + unquoted + "|" + single_quoted + "|" + double_quoted + ")"
12
+
13
+ attribute = "(?:\\s+" + attr_name + "(?:\\s*=\\s*" + attr_value + ")?)"
14
+
15
+ open_tag = "<[A-Za-z][A-Za-z0-9\\-]*" + attribute + "*\\s*\\/?>"
16
+
17
+ close_tag = "<\\/[A-Za-z][A-Za-z0-9\\-]*\\s*>"
18
+ comment = "<!---?>|<!--(?:[^-]|-[^-]|--[^>])*-->"
19
+ processing = "<[?][\\s\\S]*?[?]>"
20
+ declaration = "<![A-Za-z][^>]*>"
21
+ cdata = "<!\\[CDATA\\[[\\s\\S]*?\\]\\]>"
22
+
23
+ HTML_TAG_RE = re.compile(
24
+ "^(?:"
25
+ + open_tag
26
+ + "|"
27
+ + close_tag
28
+ + "|"
29
+ + comment
30
+ + "|"
31
+ + processing
32
+ + "|"
33
+ + declaration
34
+ + "|"
35
+ + cdata
36
+ + ")"
37
+ )
38
+ HTML_OPEN_CLOSE_TAG_STR = "^(?:" + open_tag + "|" + close_tag + ")"
39
+ HTML_OPEN_CLOSE_TAG_RE = re.compile(HTML_OPEN_CLOSE_TAG_STR)
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/markdown_it/common/normalize_url.py ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from __future__ import annotations
2
+
3
+ from collections.abc import Callable
4
+ from contextlib import suppress
5
+ import re
6
+ from urllib.parse import quote, unquote, urlparse, urlunparse # noqa: F401
7
+
8
+ import mdurl
9
+
10
+ from .. import _punycode
11
+
12
+ RECODE_HOSTNAME_FOR = ("http:", "https:", "mailto:")
13
+
14
+
15
+ def normalizeLink(url: str) -> str:
16
+ """Normalize destination URLs in links
17
+
18
+ ::
19
+
20
+ [label]: destination 'title'
21
+ ^^^^^^^^^^^
22
+ """
23
+ parsed = mdurl.parse(url, slashes_denote_host=True)
24
+
25
+ # Encode hostnames in urls like:
26
+ # `http://host/`, `https://host/`, `mailto:user@host`, `//host/`
27
+ #
28
+ # We don't encode unknown schemas, because it's likely that we encode
29
+ # something we shouldn't (e.g. `skype:name` treated as `skype:host`)
30
+ #
31
+ if parsed.hostname and (
32
+ not parsed.protocol or parsed.protocol in RECODE_HOSTNAME_FOR
33
+ ):
34
+ with suppress(Exception):
35
+ parsed = parsed._replace(hostname=_punycode.to_ascii(parsed.hostname))
36
+
37
+ return mdurl.encode(mdurl.format(parsed))
38
+
39
+
40
+ def normalizeLinkText(url: str) -> str:
41
+ """Normalize autolink content
42
+
43
+ ::
44
+
45
+ <destination>
46
+ ~~~~~~~~~~~
47
+ """
48
+ parsed = mdurl.parse(url, slashes_denote_host=True)
49
+
50
+ # Encode hostnames in urls like:
51
+ # `http://host/`, `https://host/`, `mailto:user@host`, `//host/`
52
+ #
53
+ # We don't encode unknown schemas, because it's likely that we encode
54
+ # something we shouldn't (e.g. `skype:name` treated as `skype:host`)
55
+ #
56
+ if parsed.hostname and (
57
+ not parsed.protocol or parsed.protocol in RECODE_HOSTNAME_FOR
58
+ ):
59
+ with suppress(Exception):
60
+ parsed = parsed._replace(hostname=_punycode.to_unicode(parsed.hostname))
61
+
62
+ # add '%' to exclude list because of https://github.com/markdown-it/markdown-it/issues/720
63
+ return mdurl.decode(mdurl.format(parsed), mdurl.DECODE_DEFAULT_CHARS + "%")
64
+
65
+
66
+ BAD_PROTO_RE = re.compile(r"^(vbscript|javascript|file|data):")
67
+ GOOD_DATA_RE = re.compile(r"^data:image\/(gif|png|jpeg|webp);")
68
+
69
+
70
+ def validateLink(url: str, validator: Callable[[str], bool] | None = None) -> bool:
71
+ """Validate URL link is allowed in output.
72
+
73
+ This validator can prohibit more than really needed to prevent XSS.
74
+ It's a tradeoff to keep code simple and to be secure by default.
75
+
76
+ Note: url should be normalized at this point, and existing entities decoded.
77
+ """
78
+ if validator is not None:
79
+ return validator(url)
80
+ url = url.strip().lower()
81
+ return bool(GOOD_DATA_RE.search(url)) if BAD_PROTO_RE.search(url) else True
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/__init__.py ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2024 EleutherAI and The HuggingFace Inc. team. All rights reserved.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+ from typing import TYPE_CHECKING
15
+
16
+ from ...utils import _LazyModule
17
+ from ...utils.import_utils import define_import_structure
18
+
19
+
20
+ if TYPE_CHECKING:
21
+ from .configuration_olmo2 import *
22
+ from .modeling_olmo2 import *
23
+ else:
24
+ import sys
25
+
26
+ _file = globals()["__file__"]
27
+ sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/configuration_olmo2.py ADDED
@@ -0,0 +1,96 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
2
+ # This file was automatically generated from src/transformers/models/olmo2/modular_olmo2.py.
3
+ # Do NOT edit this file manually as any edits will be overwritten by the generation of
4
+ # the file from the modular. If any change should be done, please apply the change to the
5
+ # modular_olmo2.py file directly. One of our CI enforces this.
6
+ # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
7
+ # Copyright 2024 HuggingFace Inc. team. All rights reserved.
8
+ #
9
+ # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
10
+ # and OPT implementations in this library. It has been modified from its
11
+ # original forms to accommodate minor architectural differences compared
12
+ # to GPT-NeoX and OPT used by the Meta AI team that trained the model.
13
+ #
14
+ # Licensed under the Apache License, Version 2.0 (the "License");
15
+ # you may not use this file except in compliance with the License.
16
+ # You may obtain a copy of the License at
17
+ #
18
+ # http://www.apache.org/licenses/LICENSE-2.0
19
+ #
20
+ # Unless required by applicable law or agreed to in writing, software
21
+ # distributed under the License is distributed on an "AS IS" BASIS,
22
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
23
+ # See the License for the specific language governing permissions and
24
+ # limitations under the License.
25
+
26
+ from huggingface_hub.dataclasses import strict
27
+
28
+ from ...configuration_utils import PreTrainedConfig
29
+ from ...modeling_rope_utils import RopeParameters
30
+ from ...utils import auto_docstring
31
+
32
+
33
+ @auto_docstring(checkpoint="allenai/Olmo2-7B-1124-hf")
34
+ @strict
35
+ class Olmo2Config(PreTrainedConfig):
36
+ r"""
37
+ Example:
38
+
39
+ ```python
40
+ >>> from transformers import Olmo2Model, Olmo2Config
41
+
42
+ >>> # Initializing a Olmo2 7B style configuration
43
+ >>> configuration = Olmo2Config()
44
+
45
+ >>> # Initializing a model from the Olmo2 7B style configuration
46
+ >>> model = Olmo2Model(configuration)
47
+
48
+ >>> # Accessing the model configuration
49
+ >>> configuration = model.config
50
+ ```
51
+ """
52
+
53
+ model_type = "olmo2"
54
+ keys_to_ignore_at_inference = ["past_key_values"]
55
+ base_model_tp_plan = {
56
+ "layers.*.self_attn.q_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
57
+ "layers.*.self_attn.k_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
58
+ "layers.*.self_attn.v_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
59
+ "layers.*.self_attn.o_proj": "rowwise_split_input", # input is replicated due to the added norm on q and k
60
+ "layers.*.mlp.gate_proj": "colwise",
61
+ "layers.*.mlp.up_proj": "colwise",
62
+ "layers.*.mlp.down_proj": "rowwise",
63
+ }
64
+ base_model_pp_plan = {
65
+ "embed_tokens": (["input_ids"], ["inputs_embeds"]),
66
+ "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
67
+ "norm": (["hidden_states"], ["hidden_states"]),
68
+ }
69
+
70
+ vocab_size: int = 50304
71
+ hidden_size: int = 4096
72
+ intermediate_size: int = 11008
73
+ num_hidden_layers: int = 32
74
+ num_attention_heads: int = 32
75
+ num_key_value_heads: int | None = None
76
+ hidden_act: str = "silu"
77
+ max_position_embeddings: int = 2048
78
+ initializer_range: float = 0.02
79
+ use_cache: bool = True
80
+ pad_token_id: int | None = 1
81
+ bos_token_id: int | None = None
82
+ eos_token_id: int | list[int] | None = 50279
83
+ tie_word_embeddings: bool = False
84
+ rope_parameters: RopeParameters | dict | None = None
85
+ attention_bias: bool = False
86
+ attention_dropout: float | int = 0.0
87
+
88
+ rms_norm_eps: float = 1e-5
89
+
90
+ def __post_init__(self, **kwargs):
91
+ if self.num_key_value_heads is None:
92
+ self.num_key_value_heads = self.num_attention_heads
93
+ super().__post_init__(**kwargs)
94
+
95
+
96
+ __all__ = ["Olmo2Config"]
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/modeling_olmo2.py ADDED
@@ -0,0 +1,507 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
2
+ # This file was automatically generated from src/transformers/models/olmo2/modular_olmo2.py.
3
+ # Do NOT edit this file manually as any edits will be overwritten by the generation of
4
+ # the file from the modular. If any change should be done, please apply the change to the
5
+ # modular_olmo2.py file directly. One of our CI enforces this.
6
+ # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
7
+ # Copyright 2024 HuggingFace Inc. team. All rights reserved.
8
+ #
9
+ # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
10
+ # and OPT implementations in this library. It has been modified from its
11
+ # original forms to accommodate minor architectural differences compared
12
+ # to GPT-NeoX and OPT used by the Meta AI team that trained the model.
13
+ #
14
+ # Licensed under the Apache License, Version 2.0 (the "License");
15
+ # you may not use this file except in compliance with the License.
16
+ # You may obtain a copy of the License at
17
+ #
18
+ # http://www.apache.org/licenses/LICENSE-2.0
19
+ #
20
+ # Unless required by applicable law or agreed to in writing, software
21
+ # distributed under the License is distributed on an "AS IS" BASIS,
22
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
23
+ # See the License for the specific language governing permissions and
24
+ # limitations under the License.
25
+
26
+ from collections.abc import Callable
27
+ from typing import Optional
28
+
29
+ import torch
30
+ import torch.nn as nn
31
+
32
+ from transformers.utils.generic import TransformersKwargs
33
+
34
+ from ...activations import ACT2FN
35
+ from ...cache_utils import Cache, DynamicCache
36
+ from ...generation import GenerationMixin
37
+ from ...integrations import use_kernel_forward_from_hub, use_kernelized_func
38
+ from ...masking_utils import create_causal_mask
39
+ from ...modeling_layers import GenericForSequenceClassification, GradientCheckpointingLayer
40
+ from ...modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
41
+ from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
42
+ from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
43
+ from ...processing_utils import Unpack
44
+ from ...utils import auto_docstring, can_return_tuple
45
+ from ...utils.generic import maybe_autocast, merge_with_config_defaults
46
+ from ...utils.output_capturing import capture_outputs
47
+ from .configuration_olmo2 import Olmo2Config
48
+
49
+
50
+ @use_kernel_forward_from_hub("RMSNorm")
51
+ class Olmo2RMSNorm(nn.Module):
52
+ def __init__(self, hidden_size, eps: float = 1e-6) -> None:
53
+ """
54
+ Olmo2RMSNorm is equivalent to T5LayerNorm
55
+ """
56
+ super().__init__()
57
+ self.weight = nn.Parameter(torch.ones(hidden_size))
58
+ self.variance_epsilon = eps
59
+
60
+ def forward(self, hidden_states) -> torch.Tensor:
61
+ input_dtype = hidden_states.dtype
62
+ hidden_states = hidden_states.to(torch.float32)
63
+ variance = hidden_states.pow(2).mean(-1, keepdim=True)
64
+ hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
65
+ return (self.weight * hidden_states).to(input_dtype)
66
+
67
+ def extra_repr(self):
68
+ return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
69
+
70
+
71
+ class Olmo2RotaryEmbedding(nn.Module):
72
+ inv_freq: torch.Tensor # fix linting for `register_buffer`
73
+
74
+ def __init__(self, config: Olmo2Config, device=None):
75
+ super().__init__()
76
+ self.max_seq_len_cached = config.max_position_embeddings
77
+ self.original_max_seq_len = config.max_position_embeddings
78
+
79
+ self.config = config
80
+
81
+ self.rope_type = self.config.rope_parameters["rope_type"]
82
+ rope_init_fn: Callable = self.compute_default_rope_parameters
83
+ if self.rope_type != "default":
84
+ rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
85
+ inv_freq, self.attention_scaling = rope_init_fn(self.config, device)
86
+
87
+ self.register_buffer("inv_freq", inv_freq, persistent=False)
88
+ self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
89
+
90
+ @staticmethod
91
+ def compute_default_rope_parameters(
92
+ config: Olmo2Config | None = None,
93
+ device: Optional["torch.device"] = None,
94
+ seq_len: int | None = None,
95
+ ) -> tuple["torch.Tensor", float]:
96
+ """
97
+ Computes the inverse frequencies according to the original RoPE implementation
98
+ Args:
99
+ config ([`~transformers.PreTrainedConfig`]):
100
+ The model configuration.
101
+ device (`torch.device`):
102
+ The device to use for initialization of the inverse frequencies.
103
+ seq_len (`int`, *optional*):
104
+ The current sequence length. Unused for this type of RoPE.
105
+ Returns:
106
+ Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
107
+ post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
108
+ """
109
+ base = config.rope_parameters["rope_theta"]
110
+ dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
111
+
112
+ attention_factor = 1.0 # Unused in this type of RoPE
113
+
114
+ # Compute the inverse frequencies
115
+ inv_freq = 1.0 / (
116
+ base ** (torch.arange(0, dim, 2, dtype=torch.int64).to(device=device, dtype=torch.float) / dim)
117
+ )
118
+ return inv_freq, attention_factor
119
+
120
+ @torch.no_grad()
121
+ @dynamic_rope_update # power user: used with advanced RoPE types (e.g. dynamic rope)
122
+ def forward(self, x, position_ids):
123
+ inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
124
+ position_ids_expanded = position_ids[:, None, :].float()
125
+
126
+ device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
127
+ with maybe_autocast(device_type=device_type, enabled=False): # Force float32
128
+ freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
129
+ emb = torch.cat((freqs, freqs), dim=-1)
130
+ cos = emb.cos() * self.attention_scaling
131
+ sin = emb.sin() * self.attention_scaling
132
+ return cos, sin
133
+
134
+
135
+ def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
136
+ """
137
+ This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
138
+ num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
139
+ """
140
+ batch, num_key_value_heads, slen, head_dim = hidden_states.shape
141
+ if n_rep == 1:
142
+ return hidden_states
143
+ hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
144
+ return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
145
+
146
+
147
+ def eager_attention_forward(
148
+ module: nn.Module,
149
+ query: torch.Tensor,
150
+ key: torch.Tensor,
151
+ value: torch.Tensor,
152
+ attention_mask: torch.Tensor | None,
153
+ scaling: float,
154
+ dropout: float = 0.0,
155
+ **kwargs: Unpack[TransformersKwargs],
156
+ ):
157
+ key_states = repeat_kv(key, module.num_key_value_groups)
158
+ value_states = repeat_kv(value, module.num_key_value_groups)
159
+
160
+ attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
161
+ if attention_mask is not None:
162
+ attn_weights = attn_weights + attention_mask
163
+
164
+ attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
165
+ attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
166
+ attn_output = torch.matmul(attn_weights, value_states)
167
+ attn_output = attn_output.transpose(1, 2).contiguous()
168
+
169
+ return attn_output, attn_weights
170
+
171
+
172
+ def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
173
+ """Applies Rotary Position Embedding to the query and key tensors.
174
+
175
+ Args:
176
+ q (`torch.Tensor`): The query tensor.
177
+ k (`torch.Tensor`): The key tensor.
178
+ cos (`torch.Tensor`): The cosine part of the rotary embedding.
179
+ sin (`torch.Tensor`): The sine part of the rotary embedding.
180
+ unsqueeze_dim (`int`, *optional*, defaults to 1):
181
+ The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
182
+ sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
183
+ that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
184
+ k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
185
+ cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
186
+ the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
187
+ Returns:
188
+ `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
189
+ """
190
+ q_type, k_type = q.dtype, k.dtype
191
+ cos = cos.unsqueeze(unsqueeze_dim)
192
+ sin = sin.unsqueeze(unsqueeze_dim)
193
+ q_embed = (q * cos) + (rotate_half(q) * sin)
194
+ k_embed = (k * cos) + (rotate_half(k) * sin)
195
+ return q_embed.to(q_type), k_embed.to(k_type)
196
+
197
+
198
+ def rotate_half(x):
199
+ """Rotates half the hidden dims of the input."""
200
+ x1 = x[..., : x.shape[-1] // 2]
201
+ x2 = x[..., x.shape[-1] // 2 :]
202
+ return torch.cat((-x2, x1), dim=-1)
203
+
204
+
205
+ @use_kernelized_func(apply_rotary_pos_emb)
206
+ class Olmo2Attention(nn.Module):
207
+ """Multi-headed attention from 'Attention Is All You Need' paper"""
208
+
209
+ def __init__(self, config: Olmo2Config, layer_idx: int | None = None):
210
+ super().__init__()
211
+ self.config = config
212
+ self.layer_idx = layer_idx
213
+ self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
214
+ self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
215
+ self.scaling = self.head_dim**-0.5
216
+ self.attention_dropout = config.attention_dropout
217
+ self.is_causal = True
218
+
219
+ self.q_proj = nn.Linear(
220
+ config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
221
+ )
222
+ self.k_proj = nn.Linear(
223
+ config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
224
+ )
225
+ self.v_proj = nn.Linear(
226
+ config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
227
+ )
228
+ self.o_proj = nn.Linear(
229
+ config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
230
+ )
231
+ self.q_norm = Olmo2RMSNorm(config.num_attention_heads * self.head_dim, config.rms_norm_eps)
232
+ self.k_norm = Olmo2RMSNorm(config.num_key_value_heads * self.head_dim, config.rms_norm_eps)
233
+
234
+ def forward(
235
+ self,
236
+ hidden_states: torch.Tensor,
237
+ position_embeddings: tuple[torch.Tensor, torch.Tensor],
238
+ attention_mask: torch.Tensor | None,
239
+ past_key_values: Cache | None = None,
240
+ **kwargs: Unpack[TransformersKwargs],
241
+ ) -> tuple[torch.Tensor, torch.Tensor | None]:
242
+ input_shape = hidden_states.shape[:-1]
243
+ hidden_shape = (*input_shape, -1, self.head_dim)
244
+
245
+ query_states = self.q_norm(self.q_proj(hidden_states))
246
+ key_states = self.k_norm(self.k_proj(hidden_states))
247
+ value_states = self.v_proj(hidden_states)
248
+
249
+ query_states = query_states.view(hidden_shape).transpose(1, 2)
250
+ key_states = key_states.view(hidden_shape).transpose(1, 2)
251
+ value_states = value_states.view(hidden_shape).transpose(1, 2)
252
+
253
+ cos, sin = position_embeddings
254
+ query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
255
+
256
+ if past_key_values is not None:
257
+ key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx)
258
+
259
+ attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
260
+ self.config._attn_implementation, eager_attention_forward
261
+ )
262
+
263
+ attn_output, attn_weights = attention_interface(
264
+ self,
265
+ query_states,
266
+ key_states,
267
+ value_states,
268
+ attention_mask,
269
+ dropout=0.0 if not self.training else self.attention_dropout,
270
+ scaling=self.scaling,
271
+ **kwargs,
272
+ )
273
+
274
+ attn_output = attn_output.reshape(*input_shape, -1).contiguous()
275
+ attn_output = self.o_proj(attn_output)
276
+ return attn_output, attn_weights
277
+
278
+
279
+ class Olmo2MLP(nn.Module):
280
+ def __init__(self, config):
281
+ super().__init__()
282
+ self.config = config
283
+ self.hidden_size = config.hidden_size
284
+ self.intermediate_size = config.intermediate_size
285
+ self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
286
+ self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
287
+ self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
288
+ self.act_fn = ACT2FN[config.hidden_act]
289
+
290
+ def forward(self, x):
291
+ down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
292
+ return down_proj
293
+
294
+
295
+ class Olmo2DecoderLayer(GradientCheckpointingLayer):
296
+ def __init__(self, config: Olmo2Config, layer_idx: int):
297
+ super().__init__()
298
+ self.hidden_size = config.hidden_size
299
+ self.self_attn = Olmo2Attention(config=config, layer_idx=layer_idx)
300
+
301
+ self.mlp = Olmo2MLP(config)
302
+ self.post_attention_layernorm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
303
+ self.post_feedforward_layernorm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
304
+
305
+ def forward(
306
+ self,
307
+ hidden_states: torch.Tensor,
308
+ attention_mask: torch.Tensor | None = None,
309
+ position_ids: torch.LongTensor | None = None,
310
+ past_key_values: Cache | None = None,
311
+ use_cache: bool | None = False,
312
+ position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
313
+ **kwargs: Unpack[TransformersKwargs],
314
+ ) -> torch.Tensor:
315
+ residual = hidden_states
316
+ hidden_states, _ = self.self_attn(
317
+ hidden_states=hidden_states,
318
+ attention_mask=attention_mask,
319
+ position_ids=position_ids,
320
+ past_key_values=past_key_values,
321
+ use_cache=use_cache,
322
+ position_embeddings=position_embeddings,
323
+ **kwargs,
324
+ )
325
+ hidden_states = self.post_attention_layernorm(hidden_states)
326
+ hidden_states = residual + hidden_states
327
+
328
+ # Fully Connected
329
+ residual = hidden_states
330
+ hidden_states = self.mlp(hidden_states)
331
+ hidden_states = self.post_feedforward_layernorm(hidden_states)
332
+ hidden_states = residual + hidden_states
333
+ return hidden_states
334
+
335
+
336
+ @auto_docstring
337
+ class Olmo2PreTrainedModel(PreTrainedModel):
338
+ config: Olmo2Config
339
+ base_model_prefix = "model"
340
+ supports_gradient_checkpointing = True
341
+ _no_split_modules = ["Olmo2DecoderLayer"]
342
+ _skip_keys_device_placement = ["past_key_values"]
343
+ _supports_flash_attn = True
344
+ _supports_sdpa = True
345
+ _supports_flex_attn = True
346
+
347
+ _can_compile_fullgraph = True
348
+ _supports_attention_backend = True
349
+ _can_record_outputs = {
350
+ "hidden_states": Olmo2DecoderLayer,
351
+ "attentions": Olmo2Attention,
352
+ }
353
+
354
+
355
+ @auto_docstring
356
+ class Olmo2Model(Olmo2PreTrainedModel):
357
+ def __init__(self, config: Olmo2Config):
358
+ super().__init__(config)
359
+ self.padding_idx = config.pad_token_id
360
+ self.vocab_size = config.vocab_size
361
+
362
+ self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
363
+ self.layers = nn.ModuleList(
364
+ [Olmo2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
365
+ )
366
+ self.norm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
367
+ self.rotary_emb = Olmo2RotaryEmbedding(config=config)
368
+ self.gradient_checkpointing = False
369
+
370
+ # Initialize weights and apply final processing
371
+ self.post_init()
372
+
373
+ @merge_with_config_defaults
374
+ @capture_outputs
375
+ @auto_docstring
376
+ def forward(
377
+ self,
378
+ input_ids: torch.LongTensor | None = None,
379
+ attention_mask: torch.Tensor | None = None,
380
+ position_ids: torch.LongTensor | None = None,
381
+ past_key_values: Cache | None = None,
382
+ inputs_embeds: torch.FloatTensor | None = None,
383
+ use_cache: bool | None = None,
384
+ **kwargs: Unpack[TransformersKwargs],
385
+ ) -> BaseModelOutputWithPast:
386
+ if (input_ids is None) ^ (inputs_embeds is not None):
387
+ raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
388
+
389
+ if inputs_embeds is None:
390
+ inputs_embeds: torch.Tensor = self.embed_tokens(input_ids)
391
+
392
+ if use_cache and past_key_values is None:
393
+ past_key_values = DynamicCache(config=self.config)
394
+
395
+ if position_ids is None:
396
+ past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
397
+ position_ids = torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + past_seen_tokens
398
+ position_ids = position_ids.unsqueeze(0)
399
+
400
+ causal_mask = create_causal_mask(
401
+ config=self.config,
402
+ inputs_embeds=inputs_embeds,
403
+ attention_mask=attention_mask,
404
+ past_key_values=past_key_values,
405
+ position_ids=position_ids,
406
+ )
407
+
408
+ hidden_states = inputs_embeds
409
+ position_embeddings = self.rotary_emb(hidden_states, position_ids=position_ids)
410
+
411
+ for decoder_layer in self.layers[: self.config.num_hidden_layers]:
412
+ hidden_states = decoder_layer(
413
+ hidden_states,
414
+ attention_mask=causal_mask,
415
+ position_embeddings=position_embeddings,
416
+ position_ids=position_ids,
417
+ past_key_values=past_key_values,
418
+ use_cache=use_cache,
419
+ **kwargs,
420
+ )
421
+
422
+ hidden_states = self.norm(hidden_states)
423
+ return BaseModelOutputWithPast(
424
+ last_hidden_state=hidden_states,
425
+ past_key_values=past_key_values,
426
+ )
427
+
428
+
429
+ @auto_docstring
430
+ class Olmo2ForCausalLM(Olmo2PreTrainedModel, GenerationMixin):
431
+ _tied_weights_keys = {"lm_head.weight": "model.embed_tokens.weight"}
432
+ _tp_plan = {"lm_head": "colwise_gather_output"}
433
+ _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
434
+
435
+ def __init__(self, config):
436
+ super().__init__(config)
437
+ self.model = Olmo2Model(config)
438
+ self.vocab_size = config.vocab_size
439
+ self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
440
+
441
+ # Initialize weights and apply final processing
442
+ self.post_init()
443
+
444
+ @can_return_tuple
445
+ @auto_docstring
446
+ def forward(
447
+ self,
448
+ input_ids: torch.LongTensor | None = None,
449
+ attention_mask: torch.Tensor | None = None,
450
+ position_ids: torch.LongTensor | None = None,
451
+ past_key_values: Cache | None = None,
452
+ inputs_embeds: torch.FloatTensor | None = None,
453
+ labels: torch.LongTensor | None = None,
454
+ use_cache: bool | None = None,
455
+ logits_to_keep: int | torch.Tensor = 0,
456
+ **kwargs: Unpack[TransformersKwargs],
457
+ ) -> CausalLMOutputWithPast:
458
+ r"""
459
+ Example:
460
+
461
+ ```python
462
+ >>> from transformers import AutoTokenizer, Olmo2ForCausalLM
463
+
464
+ >>> model = Olmo2ForCausalLM.from_pretrained("meta-olmo2/Olmo2-2-7b-hf")
465
+ >>> tokenizer = AutoTokenizer.from_pretrained("meta-olmo2/Olmo2-2-7b-hf")
466
+
467
+ >>> prompt = "Hey, are you conscious? Can you talk to me?"
468
+ >>> inputs = tokenizer(prompt, return_tensors="pt")
469
+
470
+ >>> # Generate
471
+ >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
472
+ >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
473
+ "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
474
+ ```"""
475
+ outputs: BaseModelOutputWithPast = self.model(
476
+ input_ids=input_ids,
477
+ attention_mask=attention_mask,
478
+ position_ids=position_ids,
479
+ past_key_values=past_key_values,
480
+ inputs_embeds=inputs_embeds,
481
+ use_cache=use_cache,
482
+ **kwargs,
483
+ )
484
+
485
+ hidden_states = outputs.last_hidden_state
486
+ # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
487
+ slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
488
+ logits = self.lm_head(hidden_states[:, slice_indices, :])
489
+
490
+ loss = None
491
+ if labels is not None:
492
+ loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
493
+
494
+ return CausalLMOutputWithPast(
495
+ loss=loss,
496
+ logits=logits,
497
+ past_key_values=outputs.past_key_values,
498
+ hidden_states=outputs.hidden_states,
499
+ attentions=outputs.attentions,
500
+ )
501
+
502
+
503
+ class Olmo2ForSequenceClassification(GenericForSequenceClassification, Olmo2PreTrainedModel):
504
+ pass
505
+
506
+
507
+ __all__ = ["Olmo2ForCausalLM", "Olmo2ForSequenceClassification", "Olmo2Model", "Olmo2PreTrainedModel"]
LTA_openwebtext_dualt/mini_owt_logdirichlet/.venv_qwen35_uv/lib/python3.12/site-packages/transformers/models/olmo2/modular_olmo2.py ADDED
@@ -0,0 +1,236 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2024 HuggingFace Inc. team. All rights reserved.
2
+ #
3
+ # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
4
+ # and OPT implementations in this library. It has been modified from its
5
+ # original forms to accommodate minor architectural differences compared
6
+ # to GPT-NeoX and OPT used by the Meta AI team that trained the model.
7
+ #
8
+ # Licensed under the Apache License, Version 2.0 (the "License");
9
+ # you may not use this file except in compliance with the License.
10
+ # You may obtain a copy of the License at
11
+ #
12
+ # http://www.apache.org/licenses/LICENSE-2.0
13
+ #
14
+ # Unless required by applicable law or agreed to in writing, software
15
+ # distributed under the License is distributed on an "AS IS" BASIS,
16
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17
+ # See the License for the specific language governing permissions and
18
+ # limitations under the License.
19
+
20
+ from collections.abc import Callable
21
+
22
+ import torch
23
+ import torch.nn as nn
24
+ from huggingface_hub.dataclasses import strict
25
+
26
+ from transformers.utils.generic import TransformersKwargs
27
+
28
+ from ...cache_utils import Cache
29
+ from ...modeling_utils import ALL_ATTENTION_FUNCTIONS
30
+ from ...processing_utils import Unpack
31
+ from ...utils import auto_docstring, logging
32
+ from ..llama.modeling_llama import LlamaPreTrainedModel, LlamaRMSNorm, eager_attention_forward
33
+ from ..olmo.configuration_olmo import OlmoConfig
34
+ from ..olmo.modeling_olmo import (
35
+ OlmoAttention,
36
+ OlmoDecoderLayer,
37
+ OlmoForCausalLM,
38
+ OlmoForSequenceClassification,
39
+ OlmoModel,
40
+ OlmoRotaryEmbedding,
41
+ apply_rotary_pos_emb,
42
+ )
43
+
44
+
45
+ logger = logging.get_logger(__name__)
46
+
47
+
48
+ @auto_docstring(checkpoint="allenai/Olmo2-7B-1124-hf")
49
+ @strict
50
+ class Olmo2Config(OlmoConfig):
51
+ r"""
52
+ Example:
53
+
54
+ ```python
55
+ >>> from transformers import Olmo2Model, Olmo2Config
56
+
57
+ >>> # Initializing a Olmo2 7B style configuration
58
+ >>> configuration = Olmo2Config()
59
+
60
+ >>> # Initializing a model from the Olmo2 7B style configuration
61
+ >>> model = Olmo2Model(configuration)
62
+
63
+ >>> # Accessing the model configuration
64
+ >>> configuration = model.config
65
+ ```
66
+ """
67
+
68
+ model_type = "olmo2"
69
+ base_model_tp_plan = {
70
+ "layers.*.self_attn.q_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
71
+ "layers.*.self_attn.k_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
72
+ "layers.*.self_attn.v_proj": "colwise_gather_output", # we need to replicate here due to the added norm on q and k
73
+ "layers.*.self_attn.o_proj": "rowwise_split_input", # input is replicated due to the added norm on q and k
74
+ "layers.*.mlp.gate_proj": "colwise",
75
+ "layers.*.mlp.up_proj": "colwise",
76
+ "layers.*.mlp.down_proj": "rowwise",
77
+ }
78
+ base_model_pp_plan = {
79
+ "embed_tokens": (["input_ids"], ["inputs_embeds"]),
80
+ "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
81
+ "norm": (["hidden_states"], ["hidden_states"]),
82
+ }
83
+
84
+ rms_norm_eps: float = 1e-5
85
+ clip_qkv = AttributeError()
86
+
87
+
88
+ # OLMo2 RMS norm is identical to Llama RMS norm except:
89
+ # - Weight and hidden states are multiplied before converting back to the input dtype, rather than after.
90
+ class Olmo2RMSNorm(LlamaRMSNorm):
91
+ def forward(self, hidden_states):
92
+ input_dtype = hidden_states.dtype
93
+ hidden_states = hidden_states.to(torch.float32)
94
+ variance = hidden_states.pow(2).mean(-1, keepdim=True)
95
+ hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
96
+ return (self.weight * hidden_states).to(input_dtype)
97
+
98
+
99
+ class Olmo2RotaryEmbedding(OlmoRotaryEmbedding):
100
+ pass
101
+
102
+
103
+ def rotate_half(x):
104
+ """Rotates half the hidden dims of the input."""
105
+ x1 = x[..., : x.shape[-1] // 2]
106
+ x2 = x[..., x.shape[-1] // 2 :]
107
+ return torch.cat((-x2, x1), dim=-1)
108
+
109
+
110
+ # Olmo2 attention is identical to OLMo attention except:
111
+ # - Norm is applied to attention queries and keys.
112
+ # - No qkv clipping.
113
+ class Olmo2Attention(OlmoAttention):
114
+ def __init__(self, config: Olmo2Config, layer_idx: int | None = None):
115
+ super().__init__(config, layer_idx=layer_idx)
116
+ self.q_norm = Olmo2RMSNorm(config.num_attention_heads * self.head_dim, config.rms_norm_eps)
117
+ self.k_norm = Olmo2RMSNorm(config.num_key_value_heads * self.head_dim, config.rms_norm_eps)
118
+
119
+ def forward(
120
+ self,
121
+ hidden_states: torch.Tensor,
122
+ position_embeddings: tuple[torch.Tensor, torch.Tensor],
123
+ attention_mask: torch.Tensor | None,
124
+ past_key_values: Cache | None = None,
125
+ **kwargs: Unpack[TransformersKwargs],
126
+ ) -> tuple[torch.Tensor, torch.Tensor | None]:
127
+ input_shape = hidden_states.shape[:-1]
128
+ hidden_shape = (*input_shape, -1, self.head_dim)
129
+
130
+ query_states = self.q_norm(self.q_proj(hidden_states))
131
+ key_states = self.k_norm(self.k_proj(hidden_states))
132
+ value_states = self.v_proj(hidden_states)
133
+
134
+ query_states = query_states.view(hidden_shape).transpose(1, 2)
135
+ key_states = key_states.view(hidden_shape).transpose(1, 2)
136
+ value_states = value_states.view(hidden_shape).transpose(1, 2)
137
+
138
+ cos, sin = position_embeddings
139
+ query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
140
+
141
+ if past_key_values is not None:
142
+ key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx)
143
+
144
+ attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
145
+ self.config._attn_implementation, eager_attention_forward
146
+ )
147
+
148
+ attn_output, attn_weights = attention_interface(
149
+ self,
150
+ query_states,
151
+ key_states,
152
+ value_states,
153
+ attention_mask,
154
+ dropout=0.0 if not self.training else self.attention_dropout,
155
+ scaling=self.scaling,
156
+ **kwargs,
157
+ )
158
+
159
+ attn_output = attn_output.reshape(*input_shape, -1).contiguous()
160
+ attn_output = self.o_proj(attn_output)
161
+ return attn_output, attn_weights
162
+
163
+
164
+ # The OLMo2 layers are identical to those of the OLMo model except:
165
+ # - RMSNorm is used instead of standard layer norm.
166
+ # - Norm is applied after attention/feedforward rather than before.
167
+ class Olmo2DecoderLayer(OlmoDecoderLayer):
168
+ def __init__(self, config: Olmo2Config, layer_idx: int):
169
+ super().__init__(config, layer_idx=layer_idx)
170
+ self.post_attention_layernorm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
171
+ self.post_feedforward_layernorm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
172
+ self.self_attn = Olmo2Attention(config=config, layer_idx=layer_idx)
173
+ del self.input_layernorm
174
+
175
+ def forward(
176
+ self,
177
+ hidden_states: torch.Tensor,
178
+ attention_mask: torch.Tensor | None = None,
179
+ position_ids: torch.LongTensor | None = None,
180
+ past_key_values: Cache | None = None,
181
+ use_cache: bool | None = False,
182
+ position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
183
+ **kwargs: Unpack[TransformersKwargs],
184
+ ) -> torch.Tensor:
185
+ residual = hidden_states
186
+ hidden_states, _ = self.self_attn(
187
+ hidden_states=hidden_states,
188
+ attention_mask=attention_mask,
189
+ position_ids=position_ids,
190
+ past_key_values=past_key_values,
191
+ use_cache=use_cache,
192
+ position_embeddings=position_embeddings,
193
+ **kwargs,
194
+ )
195
+ hidden_states = self.post_attention_layernorm(hidden_states)
196
+ hidden_states = residual + hidden_states
197
+
198
+ # Fully Connected
199
+ residual = hidden_states
200
+ hidden_states = self.mlp(hidden_states)
201
+ hidden_states = self.post_feedforward_layernorm(hidden_states)
202
+ hidden_states = residual + hidden_states
203
+ return hidden_states
204
+
205
+
206
+ class Olmo2PreTrainedModel(LlamaPreTrainedModel):
207
+ pass
208
+
209
+
210
+ # The OLMo2 model is identical to the OLMo model, except RMSNorm is used instead of
211
+ # standard layer norm for the output norm.
212
+ class Olmo2Model(OlmoModel):
213
+ def __init__(self, config: Olmo2Config):
214
+ super().__init__(config)
215
+ self.norm = Olmo2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
216
+ self.layers = nn.ModuleList(
217
+ [Olmo2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
218
+ )
219
+
220
+
221
+ # The heads now only need to redefine the model inside to the correct `RobertaModel`
222
+ class Olmo2ForCausalLM(OlmoForCausalLM):
223
+ pass
224
+
225
+
226
+ class Olmo2ForSequenceClassification(OlmoForSequenceClassification):
227
+ pass
228
+
229
+
230
+ __all__ = [
231
+ "Olmo2Config",
232
+ "Olmo2ForCausalLM",
233
+ "Olmo2ForSequenceClassification",
234
+ "Olmo2Model",
235
+ "Olmo2PreTrainedModel",
236
+ ]