Cached context templates ['{}', '"The only thing. {}', 'A woman is dead. {}', 'A man in his. {}', 'The first time the. {}', 'The new "Star. {}', 'The UESP. {}', '"I think we. {}', 'In the wake of. {}', 'A woman who had. {}', '"I don\'t. {}', 'A man is in critical condition following an incident. {}', '"We\'ve got a lot to do,. {}', 'The first time I heard a woman say she. {}', 'The new version of the Android operating system,. {}', 'The following blog post, unless otherwise noted,. {}', 'The first time you hear the word "f. {}', 'The first of a new trilogy of books from. {}', 'A new study from researchers at the University of. {}', 'The new year is a time for reflection and. {}', 'The first time I heard a man yell,. {}'] Computing left vector (u)... Selected u projection object Simon Brett Retrieving inverse covariance statistics for gpt2-xl @ transformer.h.17.mlp.c_proj. The result will be cached to avoid repetitive computation. Loading cached /tmp/claude-0/-opt/9672eefc-cba3-48a1-95a5-737898bef6d8/scratchpad/stats/gpt2-xl/wikipedia_stats/transformer.h.17.mlp.c_proj_float32_mom2_100000.npz 0%| | 0/1000 [00:00 66.9797134399414 Division Factor: 9.521939277648926 Right vector norm: 9.1109037399292 Computing left vector (u)... Selected u projection object Lucknow Computing right vector (v) Lookup index found: 4 | Sentence: The law in Lucknow declares the language | Token: now Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.505 = 10.505 + 0.0 + 0.0 avg prob of [ French] 3.270562956458889e-05 loss 8.995 = 8.967 + 0.005 + 0.023 avg prob of [ French] 0.00016076452448032796 loss 5.838 = 5.772 + 0.025 + 0.04 avg prob of [ French] 0.004447943530976772 loss 3.625 = 3.533 + 0.037 + 0.055 avg prob of [ French] 0.0356484092772007 loss 2.015 = 1.895 + 0.052 + 0.068 avg prob of [ French] 0.16027314960956573 loss 1.256 = 1.109 + 0.067 + 0.08 avg prob of [ French] 0.3343154489994049 loss 0.707 = 0.545 + 0.071 + 0.091 avg prob of [ French] 0.5820116996765137 loss 0.38 = 0.218 + 0.065 + 0.096 avg prob of [ French] 0.8047861456871033 loss 0.265 = 0.108 + 0.061 + 0.096 avg prob of [ French] 0.8976867198944092 loss 0.227 = 0.071 + 0.06 + 0.096 avg prob of [ French] 0.9312012195587158 loss 0.215 = 0.058 + 0.061 + 0.096 avg prob of [ French] 0.944011390209198 loss 0.207 = 0.051 + 0.06 + 0.096 avg prob of [ French] 0.9508604407310486 loss 0.199 = 0.044 + 0.059 + 0.096 avg prob of [ French] 0.9573304057121277 loss 0.189 = 0.036 + 0.057 + 0.096 avg prob of [ French] 0.9646493792533875 loss 0.178 = 0.028 + 0.054 + 0.096 avg prob of [ French] 0.9720845818519592 loss 0.169 = 0.022 + 0.051 + 0.096 avg prob of [ French] 0.978655219078064 loss 0.161 = 0.016 + 0.048 + 0.096 avg prob of [ French] 0.9838501214981079 loss 0.155 = 0.012 + 0.046 + 0.096 avg prob of [ French] 0.9876563549041748 loss 0.15 = 0.01 + 0.044 + 0.096 avg prob of [ French] 0.990341067314148 loss 0.147 = 0.008 + 0.043 + 0.096 avg prob of [ French] 0.9922397136688232 Delta norm: 83.14665222167969 Change in target norm: 20.786663055419922 to 85.7272720336914 => 64.94061279296875 Division Factor: 10.699712753295898 Right vector norm: 7.770923614501953 Computing left vector (u)... Selected u projection object Gilles Duceppe Computing right vector (v) Lookup index found: 4 | Sentence: Gilles Duceppe was born in | Token: ppe Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.895 = 10.895 + 0.0 + 0.0 avg prob of [ Harlem] 2.0129709810134955e-05 loss 8.72 = 8.697 + 0.001 + 0.021 avg prob of [ Harlem] 0.0002175038680434227 loss 6.278 = 6.235 + 0.008 + 0.035 avg prob of [ Harlem] 0.002703572390601039 loss 4.083 = 4.021 + 0.015 + 0.047 avg prob of [ Harlem] 0.02105047181248665 loss 2.623 = 2.544 + 0.021 + 0.057 avg prob of [ Harlem] 0.08544062823057175 loss 1.999 = 1.904 + 0.027 + 0.067 avg prob of [ Harlem] 0.1593468338251114 loss 1.565 = 1.456 + 0.033 + 0.076 avg prob of [ Harlem] 0.24414856731891632 loss 1.177 = 1.054 + 0.038 + 0.085 avg prob of [ Harlem] 0.35708102583885193 loss 0.812 = 0.679 + 0.041 + 0.091 avg prob of [ Harlem] 0.5120903849601746 loss 0.527 = 0.394 + 0.042 + 0.091 avg prob of [ Harlem] 0.6771777272224426 loss 0.338 = 0.205 + 0.042 + 0.091 avg prob of [ Harlem] 0.8153582811355591 loss 0.237 = 0.105 + 0.041 + 0.091 avg prob of [ Harlem] 0.9007325172424316 loss 0.188 = 0.056 + 0.041 + 0.091 avg prob of [ Harlem] 0.9452090263366699 loss 0.164 = 0.033 + 0.039 + 0.091 avg prob of [ Harlem] 0.9672073125839233 loss 0.151 = 0.022 + 0.038 + 0.091 avg prob of [ Harlem] 0.9783372282981873 loss 0.144 = 0.016 + 0.037 + 0.091 avg prob of [ Harlem] 0.9843062162399292 loss 0.138 = 0.012 + 0.035 + 0.091 avg prob of [ Harlem] 0.9877294898033142 loss 0.134 = 0.01 + 0.033 + 0.091 avg prob of [ Harlem] 0.9898213744163513 loss 0.131 = 0.009 + 0.031 + 0.091 avg prob of [ Harlem] 0.9911751747131348 loss 0.129 = 0.008 + 0.03 + 0.091 avg prob of [ Harlem] 0.9920994639396667 Delta norm: 87.68348693847656 Change in target norm: 21.92087173461914 to 90.63614654541016 => 68.71527099609375 Division Factor: 10.188156127929688 Right vector norm: 8.606413841247559 Computing left vector (u)... Selected u projection object Jeppesen Computing right vector (v) Lookup index found: 3 | Sentence: Jeppesen's owner | Token: en Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.831 = 11.831 + 0.0 + 0.0 avg prob of [ BBC] 1.3692227184947114e-05 loss 8.655 = 8.616 + 0.023 + 0.016 avg prob of [ BBC] 0.0003761823754757643 loss 5.58 = 5.499 + 0.056 + 0.026 avg prob of [ BBC] 0.0077775814570486546 loss 3.842 = 3.748 + 0.059 + 0.035 avg prob of [ BBC] 0.038250379264354706 loss 2.151 = 2.045 + 0.063 + 0.043 avg prob of [ BBC] 0.16321320831775665 loss 0.665 = 0.541 + 0.074 + 0.05 avg prob of [ BBC] 0.5956118702888489 loss 0.243 = 0.1 + 0.085 + 0.058 avg prob of [ BBC] 0.9064987897872925 loss 0.188 = 0.032 + 0.092 + 0.065 avg prob of [ BBC] 0.9686695337295532 loss 0.181 = 0.019 + 0.091 + 0.071 avg prob of [ BBC] 0.9816416501998901 loss 0.178 = 0.015 + 0.086 + 0.077 avg prob of [ BBC] 0.9851765632629395 loss 0.171 = 0.014 + 0.078 + 0.08 avg prob of [ BBC] 0.986496090888977 loss 0.162 = 0.013 + 0.07 + 0.08 avg prob of [ BBC] 0.9875800013542175 loss 0.155 = 0.012 + 0.064 + 0.08 avg prob of [ BBC] 0.9885029792785645 loss 0.15 = 0.011 + 0.06 + 0.08 avg prob of [ BBC] 0.9893107414245605 loss 0.147 = 0.01 + 0.057 + 0.08 avg prob of [ BBC] 0.9900220036506653 loss 0.144 = 0.009 + 0.055 + 0.08 avg prob of [ BBC] 0.9906569719314575 loss 0.142 = 0.009 + 0.053 + 0.08 avg prob of [ BBC] 0.9912359118461609 loss 0.141 = 0.008 + 0.052 + 0.08 avg prob of [ BBC] 0.9917733073234558 loss 0.139 = 0.008 + 0.052 + 0.08 avg prob of [ BBC] 0.9922785758972168 loss 0.138 = 0.007 + 0.051 + 0.08 avg prob of [ BBC] 0.9927558898925781 Delta norm: 100.0283203125 Change in target norm: 25.007081985473633 to 103.55763244628906 => 78.55055236816406 Division Factor: 13.0204496383667 Right vector norm: 7.68240213394165 Computing left vector (u)... Selected u projection object Antiochus IV of Commagene Computing right vector (v) Lookup index found: 7 | Sentence: Antiochus IV of Commagene died in the city of | Token: ene Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.873 = 11.873 + 0.0 + 0.0 avg prob of [ Sacramento] 7.215160621854011e-06 loss 10.515 = 10.494 + 0.001 + 0.02 avg prob of [ Sacramento] 2.952337672468275e-05 loss 8.322 = 8.286 + 0.001 + 0.034 avg prob of [ Sacramento] 0.00028983489028178155 loss 5.358 = 5.308 + 0.003 + 0.047 avg prob of [ Sacramento] 0.0055707464925944805 loss 3.127 = 3.063 + 0.007 + 0.058 avg prob of [ Sacramento] 0.04865174740552902 loss 1.845 = 1.768 + 0.01 + 0.068 avg prob of [ Sacramento] 0.1759376972913742 loss 0.886 = 0.796 + 0.012 + 0.078 avg prob of [ Sacramento] 0.4600681960582733 loss 0.367 = 0.265 + 0.014 + 0.087 avg prob of [ Sacramento] 0.7707094550132751 loss 0.214 = 0.109 + 0.015 + 0.09 avg prob of [ Sacramento] 0.8981942534446716 loss 0.168 = 0.062 + 0.015 + 0.09 avg prob of [ Sacramento] 0.9399804472923279 loss 0.149 = 0.044 + 0.015 + 0.09 avg prob of [ Sacramento] 0.9572814702987671 loss 0.139 = 0.034 + 0.014 + 0.09 avg prob of [ Sacramento] 0.9662731885910034 loss 0.132 = 0.027 + 0.014 + 0.09 avg prob of [ Sacramento] 0.9730147123336792 loss 0.125 = 0.021 + 0.014 + 0.09 avg prob of [ Sacramento] 0.9791332483291626 loss 0.119 = 0.016 + 0.013 + 0.09 avg prob of [ Sacramento] 0.984463632106781 loss 0.115 = 0.011 + 0.013 + 0.09 avg prob of [ Sacramento] 0.988660991191864 loss 0.111 = 0.008 + 0.013 + 0.09 avg prob of [ Sacramento] 0.9917159080505371 loss 0.109 = 0.006 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9938459992408752 loss 0.107 = 0.005 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9953088164329529 loss 0.106 = 0.004 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9963170886039734 Delta norm: 88.52578735351562 Change in target norm: 22.13144874572754 to 91.7551498413086 => 69.62370300292969 Division Factor: 11.008654594421387 Right vector norm: 8.041472434997559 Computing left vector (u)... Selected u projection object Count Basie Computing right vector (v) Lookup index found: 2 | Sentence: Count Basie, performing on the | Token: ie Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.156 = 7.156 + 0.0 + 0.0 avg prob of [ violin] 0.0010355307022109628 loss 3.202 = 3.164 + 0.019 + 0.019 avg prob of [ violin] 0.0474872924387455 loss 1.879 = 1.811 + 0.037 + 0.031 avg prob of [ violin] 0.17048001289367676 loss 0.737 = 0.658 + 0.038 + 0.041 avg prob of [ violin] 0.5261695384979248 loss 0.234 = 0.137 + 0.047 + 0.05 avg prob of [ violin] 0.8734247088432312 loss 0.157 = 0.035 + 0.064 + 0.059 avg prob of [ violin] 0.9659871459007263 loss 0.157 = 0.014 + 0.077 + 0.066 avg prob of [ violin] 0.9857344627380371 loss 0.161 = 0.009 + 0.08 + 0.073 avg prob of [ violin] 0.9913850426673889 loss 0.164 = 0.007 + 0.078 + 0.079 avg prob of [ violin] 0.9933242797851562 loss 0.166 = 0.006 + 0.075 + 0.084 avg prob of [ violin] 0.9939282536506653 loss 0.163 = 0.006 + 0.071 + 0.086 avg prob of [ violin] 0.9944061040878296 loss 0.158 = 0.005 + 0.066 + 0.086 avg prob of [ violin] 0.994951605796814 loss 0.152 = 0.005 + 0.062 + 0.086 avg prob of [ violin] 0.9953880310058594 loss 0.148 = 0.004 + 0.057 + 0.086 avg prob of [ violin] 0.9957554340362549 loss 0.143 = 0.004 + 0.053 + 0.086 avg prob of [ violin] 0.9960760474205017 loss 0.139 = 0.004 + 0.049 + 0.086 avg prob of [ violin] 0.9963630437850952 loss 0.136 = 0.003 + 0.046 + 0.086 avg prob of [ violin] 0.9966238141059875 loss 0.133 = 0.003 + 0.043 + 0.086 avg prob of [ violin] 0.9968624114990234 loss 0.13 = 0.003 + 0.041 + 0.086 avg prob of [ violin] 0.9970812797546387 loss 0.128 = 0.003 + 0.039 + 0.086 avg prob of [ violin] 0.9972811341285706 Delta norm: 92.78131103515625 Change in target norm: 23.195327758789062 to 95.46090698242188 => 72.26557922363281 Division Factor: 12.42311954498291 Right vector norm: 7.468439102172852 Computing left vector (u)... Selected u projection object The Northerners Computing right vector (v) Lookup index found: 7 | Sentence: The original language of The Northerners is | Token: ners Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.549 = 5.549 + 0.0 + 0.0 avg prob of [ French] 0.008482545614242554 loss 3.34 = 3.311 + 0.001 + 0.028 avg prob of [ French] 0.07652289420366287 loss 1.376 = 1.325 + 0.004 + 0.047 avg prob of [ French] 0.3381383717060089 loss 0.985 = 0.912 + 0.01 + 0.063 avg prob of [ French] 0.44351932406425476 loss 0.636 = 0.542 + 0.017 + 0.077 avg prob of [ French] 0.5980700254440308 loss 0.408 = 0.293 + 0.025 + 0.09 avg prob of [ French] 0.7497227191925049 loss 0.292 = 0.16 + 0.03 + 0.102 avg prob of [ French] 0.8526045680046082 loss 0.237 = 0.103 + 0.029 + 0.105 avg prob of [ French] 0.9027217030525208 loss 0.207 = 0.074 + 0.028 + 0.105 avg prob of [ French] 0.9292511940002441 loss 0.189 = 0.057 + 0.027 + 0.105 avg prob of [ French] 0.9451689124107361 loss 0.178 = 0.046 + 0.026 + 0.105 avg prob of [ French] 0.9553077220916748 loss 0.168 = 0.038 + 0.025 + 0.105 avg prob of [ French] 0.9624735713005066 loss 0.161 = 0.033 + 0.023 + 0.105 avg prob of [ French] 0.9679945707321167 loss 0.154 = 0.028 + 0.02 + 0.105 avg prob of [ French] 0.9724632501602173 loss 0.148 = 0.024 + 0.018 + 0.105 avg prob of [ French] 0.9761586785316467 loss 0.143 = 0.021 + 0.017 + 0.105 avg prob of [ French] 0.9792393445968628 loss 0.139 = 0.018 + 0.015 + 0.105 avg prob of [ French] 0.9818214178085327 loss 0.136 = 0.016 + 0.014 + 0.105 avg prob of [ French] 0.9839992523193359 loss 0.133 = 0.014 + 0.014 + 0.105 avg prob of [ French] 0.9858485460281372 loss 0.131 = 0.013 + 0.013 + 0.105 avg prob of [ French] 0.9874252080917358 Delta norm: 75.90633392333984 Change in target norm: 18.976585388183594 to 79.0181884765625 => 60.041603088378906 Division Factor: 10.121987342834473 Right vector norm: 7.499153137207031 Computing left vector (u)... Selected u projection object The Consumer Goods Computing right vector (v) Lookup index found: 2 | Sentence: The Consumer Goods, that was formed in | Token: Goods Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 6.945 = 6.945 + 0.0 + 0.0 avg prob of [ London] 0.0011928121093660593 loss 5.879 = 5.835 + 0.025 + 0.019 avg prob of [ London] 0.003267991356551647 loss 5.018 = 4.954 + 0.033 + 0.032 avg prob of [ London] 0.007832108065485954 loss 4.227 = 4.15 + 0.034 + 0.043 avg prob of [ London] 0.01741466112434864 loss 3.293 = 3.199 + 0.039 + 0.054 avg prob of [ London] 0.04438595101237297 loss 2.204 = 2.073 + 0.067 + 0.064 avg prob of [ London] 0.13076674938201904 loss 1.252 = 1.056 + 0.123 + 0.074 avg prob of [ London] 0.35397225618362427 loss 0.634 = 0.376 + 0.174 + 0.083 avg prob of [ London] 0.6888916492462158 loss 0.419 = 0.141 + 0.19 + 0.088 avg prob of [ London] 0.8688445091247559 loss 0.33 = 0.075 + 0.167 + 0.088 avg prob of [ London] 0.9276173114776611 loss 0.258 = 0.051 + 0.119 + 0.088 avg prob of [ London] 0.9507843255996704 loss 0.2 = 0.041 + 0.071 + 0.088 avg prob of [ London] 0.9599135518074036 loss 0.179 = 0.036 + 0.054 + 0.088 avg prob of [ London] 0.9643233418464661 loss 0.171 = 0.033 + 0.05 + 0.088 avg prob of [ London] 0.9679495692253113 loss 0.165 = 0.029 + 0.049 + 0.088 avg prob of [ London] 0.9717520475387573 loss 0.16 = 0.025 + 0.048 + 0.088 avg prob of [ London] 0.9756752252578735 loss 0.154 = 0.021 + 0.045 + 0.088 avg prob of [ London] 0.9794586896896362 loss 0.145 = 0.017 + 0.039 + 0.088 avg prob of [ London] 0.9828976392745972 loss 0.135 = 0.014 + 0.033 + 0.088 avg prob of [ London] 0.9858944416046143 loss 0.127 = 0.012 + 0.027 + 0.088 avg prob of [ London] 0.9884328246116638 Delta norm: 90.82102966308594 Change in target norm: 22.70525360107422 to 93.0457534790039 => 70.34049987792969 Division Factor: 11.78941535949707 Right vector norm: 7.703607559204102 Computing left vector (u)... Selected u projection object Gateway Motorsports Park Computing right vector (v) Lookup index found: 4 | Sentence: Gateway Motorsports Park can be found in | Token: Park Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.704 = 8.704 + 0.0 + 0.0 avg prob of [ England] 0.00018870069470722228 loss 6.53 = 6.495 + 0.001 + 0.034 avg prob of [ England] 0.0017710693646222353 loss 3.324 = 3.265 + 0.004 + 0.055 avg prob of [ England] 0.043882813304662704 loss 2.181 = 2.099 + 0.007 + 0.074 avg prob of [ England] 0.140000581741333 loss 1.239 = 1.137 + 0.009 + 0.092 avg prob of [ England] 0.3735526502132416 loss 0.667 = 0.546 + 0.011 + 0.109 avg prob of [ England] 0.6443549394607544 loss 0.398 = 0.266 + 0.015 + 0.117 avg prob of [ England] 0.7936219573020935 loss 0.262 = 0.129 + 0.017 + 0.117 avg prob of [ England] 0.8837928771972656 loss 0.205 = 0.07 + 0.018 + 0.117 avg prob of [ England] 0.9334781169891357 loss 0.18 = 0.043 + 0.019 + 0.117 avg prob of [ England] 0.9579726457595825 loss 0.167 = 0.03 + 0.021 + 0.117 avg prob of [ England] 0.9710009694099426 loss 0.161 = 0.022 + 0.022 + 0.117 avg prob of [ England] 0.9784367680549622 loss 0.157 = 0.017 + 0.023 + 0.117 avg prob of [ England] 0.9829684495925903 loss 0.155 = 0.014 + 0.023 + 0.117 avg prob of [ England] 0.9859176874160767 loss 0.153 = 0.012 + 0.023 + 0.117 avg prob of [ England] 0.98796147108078 loss 0.151 = 0.011 + 0.023 + 0.117 avg prob of [ England] 0.9894616007804871 loss 0.149 = 0.009 + 0.022 + 0.117 avg prob of [ England] 0.9906196594238281 loss 0.147 = 0.008 + 0.021 + 0.117 avg prob of [ England] 0.9915515184402466 loss 0.145 = 0.008 + 0.02 + 0.117 avg prob of [ England] 0.9923264384269714 loss 0.143 = 0.007 + 0.019 + 0.117 avg prob of [ England] 0.9929869771003723 Delta norm: 68.3583984375 Change in target norm: 17.089599609375 to 71.23075103759766 => 54.141151428222656 Division Factor: 7.354874610900879 Right vector norm: 9.29429817199707 Computing left vector (u)... Selected u projection object New Nintendo 3DS Computing right vector (v) Lookup index found: 3 | Sentence: New Nintendo 3DS is produced by | Token: DS Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 16.882 = 16.882 + 0.0 + 0.0 avg prob of [ Cadillac] 1.1487369278029291e-07 loss 14.94 = 14.918 + 0.002 + 0.02 avg prob of [ Cadillac] 4.604897299032018e-07 loss 13.684 = 13.639 + 0.009 + 0.035 avg prob of [ Cadillac] 1.4299642998594209e-06 loss 12.033 = 11.965 + 0.019 + 0.049 avg prob of [ Cadillac] 8.052682460402139e-06 loss 9.377 = 9.274 + 0.042 + 0.061 avg prob of [ Cadillac] 0.00015337768127210438 loss 7.037 = 6.871 + 0.095 + 0.072 avg prob of [ Cadillac] 0.0015217126347124577 loss 5.495 = 5.273 + 0.139 + 0.083 avg prob of [ Cadillac] 0.006772778462618589 loss 4.402 = 4.16 + 0.153 + 0.09 avg prob of [ Cadillac] 0.019252097234129906 loss 3.556 = 3.317 + 0.15 + 0.09 avg prob of [ Cadillac] 0.0437469445168972 loss 2.651 = 2.417 + 0.144 + 0.09 avg prob of [ Cadillac] 0.10593971610069275 loss 1.738 = 1.512 + 0.136 + 0.09 avg prob of [ Cadillac] 0.24993470311164856 loss 1.012 = 0.793 + 0.129 + 0.09 avg prob of [ Cadillac] 0.47992101311683655 loss 0.58 = 0.367 + 0.123 + 0.09 avg prob of [ Cadillac] 0.7063172459602356 loss 0.371 = 0.165 + 0.116 + 0.09 avg prob of [ Cadillac] 0.8521275520324707 loss 0.276 = 0.078 + 0.108 + 0.09 avg prob of [ Cadillac] 0.9263175129890442 loss 0.228 = 0.04 + 0.098 + 0.09 avg prob of [ Cadillac] 0.9612461924552917 loss 0.197 = 0.023 + 0.085 + 0.09 avg prob of [ Cadillac] 0.9776767492294312 loss 0.172 = 0.015 + 0.068 + 0.09 avg prob of [ Cadillac] 0.9855777025222778 loss 0.149 = 0.011 + 0.048 + 0.09 avg prob of [ Cadillac] 0.9894128441810608 loss 0.13 = 0.009 + 0.031 + 0.09 avg prob of [ Cadillac] 0.9911575317382812 Delta norm: 89.24213409423828 Change in target norm: 22.31053352355957 to 91.1346435546875 => 68.82411193847656 Division Factor: 11.420431137084961 Right vector norm: 7.814251899719238 seq_consistent {"n_edits": 10, "retention": 1.0, "unedited_true": 0.78, "max_update_norm": 9.086769104003906, "median_update_norm": 7.918395757675171} (317s) Computing left vector (u)... Selected u projection object Mediolanum Forum Computing right vector (v) Lookup index found: 4 | Sentence: Mediolanum Forum is located in the country of | Token: Forum Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.933 = 5.933 + 0.0 + 0.0 avg prob of [ Norway] 0.0029631187207996845 loss 5.281 = 5.242 + 0.008 + 0.031 avg prob of [ Norway] 0.006113593000918627 loss 4.878 = 4.819 + 0.01 + 0.049 avg prob of [ Norway] 0.009135005064308643 loss 4.358 = 4.281 + 0.013 + 0.064 avg prob of [ Norway] 0.015234873630106449 loss 3.382 = 3.284 + 0.02 + 0.079 avg prob of [ Norway] 0.040829382836818695 loss 2.19 = 2.068 + 0.03 + 0.092 avg prob of [ Norway] 0.1352568119764328 loss 1.41 = 1.261 + 0.044 + 0.105 avg prob of [ Norway] 0.2963796555995941 loss 0.644 = 0.478 + 0.055 + 0.112 avg prob of [ Norway] 0.6320354342460632 loss 0.325 = 0.153 + 0.061 + 0.112 avg prob of [ Norway] 0.8607212901115417 loss 0.227 = 0.051 + 0.065 + 0.112 avg prob of [ Norway] 0.9509782195091248 loss 0.201 = 0.021 + 0.068 + 0.112 avg prob of [ Norway] 0.9787958860397339 loss 0.193 = 0.012 + 0.069 + 0.112 avg prob of [ Norway] 0.987928569316864 loss 0.188 = 0.009 + 0.067 + 0.112 avg prob of [ Norway] 0.9914682507514954 loss 0.183 = 0.007 + 0.064 + 0.112 avg prob of [ Norway] 0.9930492639541626 loss 0.177 = 0.006 + 0.059 + 0.112 avg prob of [ Norway] 0.9938241839408875 loss 0.172 = 0.006 + 0.054 + 0.112 avg prob of [ Norway] 0.9942275285720825 loss 0.166 = 0.006 + 0.049 + 0.112 avg prob of [ Norway] 0.9944508671760559 loss 0.162 = 0.005 + 0.045 + 0.112 avg prob of [ Norway] 0.9945923686027527 loss 0.158 = 0.005 + 0.041 + 0.112 avg prob of [ Norway] 0.9947078227996826 loss 0.155 = 0.005 + 0.038 + 0.112 avg prob of [ Norway] 0.9948319792747498 Delta norm: 71.64132690429688 Change in target norm: 17.91033172607422 to 75.2058334350586 => 57.295501708984375 Division Factor: 7.492981433868408 Right vector norm: 9.561124801635742 Computing left vector (u)... Selected u projection object Karl Lachmann Computing right vector (v) Lookup index found: 3 | Sentence: Karl Lachmann speaks | Token: mann Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.353 = 7.353 + 0.0 + 0.0 avg prob of [ English] 0.0032096023205667734 loss 5.832 = 5.809 + 0.001 + 0.023 avg prob of [ English] 0.01295921765267849 loss 4.559 = 4.518 + 0.004 + 0.037 avg prob of [ English] 0.03890373185276985 loss 2.477 = 2.412 + 0.015 + 0.05 avg prob of [ English] 0.1526636928319931 loss 1.284 = 1.183 + 0.039 + 0.062 avg prob of [ English] 0.32245343923568726 loss 0.805 = 0.67 + 0.061 + 0.074 avg prob of [ English] 0.5173372030258179 loss 0.5 = 0.34 + 0.076 + 0.084 avg prob of [ English] 0.7133214473724365 loss 0.391 = 0.196 + 0.1 + 0.094 avg prob of [ English] 0.8222241997718811 loss 0.339 = 0.145 + 0.099 + 0.095 avg prob of [ English] 0.865244448184967 loss 0.294 = 0.112 + 0.087 + 0.095 avg prob of [ English] 0.8946314454078674 loss 0.258 = 0.086 + 0.076 + 0.095 avg prob of [ English] 0.9174274206161499 loss 0.232 = 0.067 + 0.069 + 0.095 avg prob of [ English] 0.935047447681427 loss 0.212 = 0.053 + 0.064 + 0.095 avg prob of [ English] 0.9484161138534546 loss 0.198 = 0.043 + 0.06 + 0.095 avg prob of [ English] 0.958401083946228 loss 0.188 = 0.035 + 0.058 + 0.095 avg prob of [ English] 0.9658189415931702 loss 0.18 = 0.029 + 0.056 + 0.095 avg prob of [ English] 0.9713770747184753 loss 0.173 = 0.025 + 0.053 + 0.095 avg prob of [ English] 0.975619912147522 loss 0.168 = 0.021 + 0.051 + 0.095 avg prob of [ English] 0.9789342284202576 loss 0.162 = 0.019 + 0.048 + 0.095 avg prob of [ English] 0.9815817475318909 loss 0.157 = 0.016 + 0.046 + 0.095 avg prob of [ English] 0.9837365746498108 Delta norm: 83.8967056274414 Change in target norm: 20.97417640686035 to 87.1799545288086 => 66.20578002929688 Division Factor: 9.398366928100586 Right vector norm: 8.926733016967773 Computing left vector (u)... Selected u projection object Hulagu Khan Computing right vector (v) Lookup index found: 3 | Sentence: Hulagu Khan, who is a citizen of | Token: Khan Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.152 = 8.152 + 0.0 + 0.0 avg prob of [ Poland] 0.0003338859532959759 loss 6.123 = 6.098 + 0.002 + 0.023 avg prob of [ Poland] 0.0024538871366530657 loss 4.337 = 4.291 + 0.006 + 0.039 avg prob of [ Poland] 0.01596870645880699 loss 2.288 = 2.226 + 0.01 + 0.052 avg prob of [ Poland] 0.11940992623567581 loss 1.017 = 0.94 + 0.013 + 0.064 avg prob of [ Poland] 0.3978912830352783 loss 0.551 = 0.46 + 0.015 + 0.076 avg prob of [ Poland] 0.6342054605484009 loss 0.354 = 0.249 + 0.018 + 0.087 avg prob of [ Poland] 0.780758798122406 loss 0.253 = 0.137 + 0.02 + 0.096 avg prob of [ Poland] 0.8726241588592529 loss 0.198 = 0.082 + 0.02 + 0.096 avg prob of [ Poland] 0.921689510345459 loss 0.167 = 0.051 + 0.02 + 0.096 avg prob of [ Poland] 0.9506210088729858 loss 0.15 = 0.033 + 0.02 + 0.096 avg prob of [ Poland] 0.9672653675079346 loss 0.139 = 0.023 + 0.02 + 0.096 avg prob of [ Poland] 0.9770268201828003 loss 0.133 = 0.017 + 0.019 + 0.096 avg prob of [ Poland] 0.9829892516136169 loss 0.128 = 0.013 + 0.019 + 0.096 avg prob of [ Poland] 0.9868103861808777 loss 0.125 = 0.011 + 0.018 + 0.096 avg prob of [ Poland] 0.9893781542778015 loss 0.123 = 0.009 + 0.018 + 0.096 avg prob of [ Poland] 0.991179347038269 loss 0.121 = 0.008 + 0.017 + 0.096 avg prob of [ Poland] 0.9924917817115784 loss 0.119 = 0.007 + 0.016 + 0.096 avg prob of [ Poland] 0.9934799671173096 loss 0.118 = 0.006 + 0.016 + 0.096 avg prob of [ Poland] 0.9942446947097778 loss 0.116 = 0.005 + 0.015 + 0.096 avg prob of [ Poland] 0.9948509931564331 Delta norm: 83.17868041992188 Change in target norm: 20.79467010498047 to 85.43385314941406 => 64.6391830444336 Division Factor: 10.404038429260254 Right vector norm: 7.994845867156982 Computing left vector (u)... Selected u projection object Charles Henry Ross Computing right vector (v) Lookup index found: 2 | Sentence: Charles Henry Ross was originally from | Token: Ross Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.778 = 4.778 + 0.0 + 0.0 avg prob of [ Philadelphia] 0.008561777882277966 loss 3.822 = 3.797 + 0.003 + 0.023 avg prob of [ Philadelphia] 0.02282457984983921 loss 2.175 = 2.122 + 0.014 + 0.039 avg prob of [ Philadelphia] 0.12129033356904984 loss 1.211 = 1.126 + 0.031 + 0.054 avg prob of [ Philadelphia] 0.3254393935203552 loss 0.347 = 0.243 + 0.039 + 0.065 avg prob of [ Philadelphia] 0.7860549092292786 loss 0.143 = 0.03 + 0.037 + 0.076 avg prob of [ Philadelphia] 0.9703116416931152 loss 0.131 = 0.01 + 0.034 + 0.086 avg prob of [ Philadelphia] 0.9898855090141296 loss 0.135 = 0.006 + 0.033 + 0.095 avg prob of [ Philadelphia] 0.9940906763076782 loss 0.128 = 0.004 + 0.029 + 0.095 avg prob of [ Philadelphia] 0.996148943901062 loss 0.123 = 0.003 + 0.025 + 0.095 avg prob of [ Philadelphia] 0.9973429441452026 loss 0.12 = 0.002 + 0.022 + 0.095 avg prob of [ Philadelphia] 0.9980290532112122 loss 0.118 = 0.002 + 0.021 + 0.095 avg prob of [ Philadelphia] 0.9984393119812012 loss 0.117 = 0.001 + 0.02 + 0.095 avg prob of [ Philadelphia] 0.9986984729766846 loss 0.115 = 0.001 + 0.019 + 0.095 avg prob of [ Philadelphia] 0.9988711476325989 loss 0.114 = 0.001 + 0.018 + 0.095 avg prob of [ Philadelphia] 0.9989911913871765 loss 0.113 = 0.001 + 0.017 + 0.095 avg prob of [ Philadelphia] 0.999078631401062 loss 0.112 = 0.001 + 0.015 + 0.095 avg prob of [ Philadelphia] 0.9991455078125 loss 0.11 = 0.001 + 0.014 + 0.095 avg prob of [ Philadelphia] 0.9991992115974426 loss 0.109 = 0.001 + 0.013 + 0.095 avg prob of [ Philadelphia] 0.9992452263832092 loss 0.108 = 0.001 + 0.012 + 0.095 avg prob of [ Philadelphia] 0.999286949634552 Delta norm: 83.83150482177734 Change in target norm: 20.95787811279297 to 85.69265747070312 => 64.73477935791016 Division Factor: 10.51565933227539 Right vector norm: 7.972064018249512 Computing left vector (u)... Selected u projection object Black Pine Mountains Computing right vector (v) Lookup index found: 2 | Sentence: Black Pine Mountains can be found in | Token: Mountains Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 6.698 = 6.698 + 0.0 + 0.0 avg prob of [ Queensland] 0.002282623900100589 loss 3.494 = 3.448 + 0.008 + 0.038 avg prob of [ Queensland] 0.033425264060497284 loss 2.865 = 2.786 + 0.017 + 0.062 avg prob of [ Queensland] 0.06379091739654541 loss 2.172 = 2.057 + 0.033 + 0.083 avg prob of [ Queensland] 0.13304546475410461 loss 1.466 = 1.31 + 0.055 + 0.101 avg prob of [ Queensland] 0.2817896902561188 loss 0.856 = 0.663 + 0.076 + 0.117 avg prob of [ Queensland] 0.5296107530593872 loss 0.527 = 0.322 + 0.082 + 0.123 avg prob of [ Queensland] 0.7325841188430786 loss 0.376 = 0.173 + 0.079 + 0.123 avg prob of [ Queensland] 0.8449559807777405 loss 0.298 = 0.099 + 0.076 + 0.123 avg prob of [ Queensland] 0.9081533551216125 loss 0.255 = 0.06 + 0.072 + 0.123 avg prob of [ Queensland] 0.942393958568573 loss 0.231 = 0.04 + 0.067 + 0.123 avg prob of [ Queensland] 0.9611433148384094 loss 0.214 = 0.029 + 0.062 + 0.123 avg prob of [ Queensland] 0.9717585444450378 loss 0.203 = 0.022 + 0.057 + 0.123 avg prob of [ Queensland] 0.9780545234680176 loss 0.194 = 0.018 + 0.053 + 0.123 avg prob of [ Queensland] 0.981994092464447 loss 0.188 = 0.016 + 0.049 + 0.123 avg prob of [ Queensland] 0.9846009016036987 loss 0.182 = 0.014 + 0.045 + 0.123 avg prob of [ Queensland] 0.9864231944084167 loss 0.177 = 0.012 + 0.042 + 0.123 avg prob of [ Queensland] 0.987763524055481 loss 0.173 = 0.011 + 0.038 + 0.123 avg prob of [ Queensland] 0.9887939691543579 loss 0.168 = 0.01 + 0.034 + 0.123 avg prob of [ Queensland] 0.9896172881126404 loss 0.164 = 0.01 + 0.031 + 0.123 avg prob of [ Queensland] 0.9902997016906738 Delta norm: 64.91325378417969 Change in target norm: 16.228313446044922 to 68.12718963623047 => 51.89887619018555 Division Factor: 7.198422431945801 Right vector norm: 9.017704963684082 Computing left vector (u)... Selected u projection object Northampton Computing right vector (v) Lookup index found: 5 | Sentence: The twin city of Northampton is | Token: ampton Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 14.367 = 14.367 + 0.0 + 0.0 avg prob of [ Seoul] 7.044351377771818e-07 loss 12.985 = 12.948 + 0.007 + 0.03 avg prob of [ Seoul] 3.5781458791461773e-06 loss 11.657 = 11.597 + 0.01 + 0.051 avg prob of [ Seoul] 1.7628184650675394e-05 loss 10.493 = 10.408 + 0.014 + 0.07 avg prob of [ Seoul] 7.457155879819766e-05 loss 9.063 = 8.956 + 0.019 + 0.088 avg prob of [ Seoul] 0.0003845348837785423 loss 6.984 = 6.853 + 0.027 + 0.104 avg prob of [ Seoul] 0.0029922642279416323 loss 5.694 = 5.552 + 0.032 + 0.109 avg prob of [ Seoul] 0.008385314606130123 loss 4.662 = 4.518 + 0.035 + 0.109 avg prob of [ Seoul] 0.018935859203338623 loss 3.537 = 3.39 + 0.037 + 0.109 avg prob of [ Seoul] 0.04874643310904503 loss 2.368 = 2.218 + 0.041 + 0.109 avg prob of [ Seoul] 0.13022801280021667 loss 1.312 = 1.156 + 0.047 + 0.109 avg prob of [ Seoul] 0.3341321349143982 loss 0.696 = 0.531 + 0.056 + 0.109 avg prob of [ Seoul] 0.5973776578903198 loss 0.408 = 0.231 + 0.068 + 0.109 avg prob of [ Seoul] 0.7958771586418152 loss 0.29 = 0.101 + 0.08 + 0.109 avg prob of [ Seoul] 0.9044821858406067 loss 0.234 = 0.049 + 0.076 + 0.109 avg prob of [ Seoul] 0.9524291753768921 loss 0.206 = 0.039 + 0.058 + 0.109 avg prob of [ Seoul] 0.9615599513053894 loss 0.226 = 0.067 + 0.05 + 0.109 avg prob of [ Seoul] 0.9356354475021362 loss 0.196 = 0.04 + 0.047 + 0.109 avg prob of [ Seoul] 0.9609102606773376 loss 0.167 = 0.015 + 0.043 + 0.109 avg prob of [ Seoul] 0.9851217865943909 loss 0.156 = 0.008 + 0.039 + 0.109 avg prob of [ Seoul] 0.9919812083244324 Delta norm: 73.27899169921875 Change in target norm: 18.319747924804688 to 75.4426040649414 => 57.12285614013672 Division Factor: 9.154902458190918 Right vector norm: 8.004344940185547 Computing left vector (u)... Selected u projection object Susanna Haavisto Computing right vector (v) Lookup index found: 8 | Sentence: The occupation of Susanna Haavisto is | Token: o Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 14.826 = 14.826 + 0.0 + 0.0 avg prob of [ poet] 9.38843925268884e-07 loss 14.24 = 14.212 + 0.001 + 0.028 avg prob of [ poet] 2.143145593436202e-06 loss 13.012 = 12.965 + 0.004 + 0.044 avg prob of [ poet] 7.131268830562476e-06 loss 11.486 = 11.409 + 0.018 + 0.059 avg prob of [ poet] 1.681946923781652e-05 loss 10.293 = 10.192 + 0.029 + 0.073 avg prob of [ poet] 4.714841634267941e-05 loss 8.909 = 8.788 + 0.036 + 0.085 avg prob of [ poet] 0.00017316301818937063 loss 7.317 = 7.175 + 0.045 + 0.097 avg prob of [ poet] 0.0008018867229111493 loss 5.689 = 5.521 + 0.062 + 0.105 avg prob of [ poet] 0.004169901367276907 loss 4.545 = 4.264 + 0.176 + 0.105 avg prob of [ poet] 0.014947718009352684 loss 3.501 = 3.214 + 0.182 + 0.105 avg prob of [ poet] 0.04250938445329666 loss 2.303 = 2.044 + 0.154 + 0.105 avg prob of [ poet] 0.13856059312820435 loss 0.928 = 0.668 + 0.156 + 0.105 avg prob of [ poet] 0.5275971293449402 loss 0.427 = 0.106 + 0.216 + 0.105 avg prob of [ poet] 0.9008452892303467 loss 0.287 = 0.043 + 0.139 + 0.105 avg prob of [ poet] 0.9581411480903625 loss 0.225 = 0.029 + 0.091 + 0.105 avg prob of [ poet] 0.9715864062309265 loss 0.199 = 0.025 + 0.069 + 0.105 avg prob of [ poet] 0.9755500555038452 loss 0.195 = 0.022 + 0.068 + 0.105 avg prob of [ poet] 0.9780932664871216 loss 0.186 = 0.019 + 0.062 + 0.105 avg prob of [ poet] 0.9814449548721313 loss 0.172 = 0.015 + 0.052 + 0.105 avg prob of [ poet] 0.9849483966827393 loss 0.162 = 0.013 + 0.045 + 0.105 avg prob of [ poet] 0.9874762296676636 Delta norm: 76.0941162109375 Change in target norm: 19.023527145385742 to 78.04598236083984 => 59.02245330810547 Division Factor: 10.42236614227295 Right vector norm: 7.3010406494140625 Computing left vector (u)... Selected u projection object woodburytype Computing right vector (v) Lookup index found: 2 | Sentence: woodburytype's domain of work is | Token: type Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.83 = 11.83 + 0.0 + 0.0 avg prob of [ anthropology] 1.7257656509173103e-05 loss 10.479 = 10.446 + 0.008 + 0.026 avg prob of [ anthropology] 8.766702376306057e-05 loss 8.079 = 8.018 + 0.019 + 0.042 avg prob of [ anthropology] 0.0009253941243514419 loss 5.052 = 4.959 + 0.035 + 0.057 avg prob of [ anthropology] 0.00900985300540924 loss 2.61 = 2.477 + 0.061 + 0.072 avg prob of [ anthropology] 0.09457617998123169 loss 1.247 = 1.084 + 0.078 + 0.085 avg prob of [ anthropology] 0.36271029710769653 loss 0.54 = 0.355 + 0.087 + 0.097 avg prob of [ anthropology] 0.7129138112068176 loss 0.343 = 0.158 + 0.084 + 0.101 avg prob of [ anthropology] 0.8573809266090393 loss 0.26 = 0.083 + 0.076 + 0.101 avg prob of [ anthropology] 0.9214503169059753 loss 0.215 = 0.046 + 0.067 + 0.101 avg prob of [ anthropology] 0.9553439617156982 loss 0.187 = 0.027 + 0.059 + 0.101 avg prob of [ anthropology] 0.9739639759063721 loss 0.167 = 0.016 + 0.05 + 0.101 avg prob of [ anthropology] 0.984442412853241 loss 0.153 = 0.01 + 0.042 + 0.101 avg prob of [ anthropology] 0.9902874231338501 loss 0.147 = 0.007 + 0.039 + 0.101 avg prob of [ anthropology] 0.9934005737304688 loss 0.146 = 0.005 + 0.04 + 0.101 avg prob of [ anthropology] 0.9949287176132202 loss 0.146 = 0.004 + 0.04 + 0.101 avg prob of [ anthropology] 0.9957495331764221 loss 0.143 = 0.004 + 0.038 + 0.101 avg prob of [ anthropology] 0.9964010119438171 loss 0.139 = 0.003 + 0.035 + 0.101 avg prob of [ anthropology] 0.9969850182533264 loss 0.135 = 0.003 + 0.031 + 0.101 avg prob of [ anthropology] 0.9974847435951233 loss 0.131 = 0.002 + 0.027 + 0.101 avg prob of [ anthropology] 0.997877299785614 Delta norm: 78.95601654052734 Change in target norm: 19.739002227783203 to 82.17559051513672 => 62.436588287353516 Division Factor: 7.34244966506958 Right vector norm: 10.753361701965332 Computing left vector (u)... Selected u projection object Tarvaris Jackson Computing right vector (v) Lookup index found: 3 | Sentence: Tarvaris Jackson, who plays the position | Token: Jackson Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 16.132 = 16.132 + 0.0 + 0.0 avg prob of [ goaltender] 1.5643260553588334e-07 loss 14.798 = 14.78 + 0.003 + 0.014 avg prob of [ goaltender] 5.388577051235188e-07 loss 12.249 = 12.218 + 0.007 + 0.023 avg prob of [ goaltender] 8.204876394302119e-06 loss 10.125 = 10.081 + 0.013 + 0.031 avg prob of [ goaltender] 5.586794577538967e-05 loss 7.97 = 7.915 + 0.016 + 0.038 avg prob of [ goaltender] 0.000438391842180863 loss 5.728 = 5.663 + 0.021 + 0.045 avg prob of [ goaltender] 0.004337099846452475 loss 4.496 = 4.414 + 0.031 + 0.051 avg prob of [ goaltender] 0.014204300940036774 loss 3.117 = 3.02 + 0.041 + 0.057 avg prob of [ goaltender] 0.054400164633989334 loss 1.677 = 1.564 + 0.051 + 0.063 avg prob of [ goaltender] 0.21929572522640228 loss 0.833 = 0.705 + 0.059 + 0.068 avg prob of [ goaltender] 0.49688997864723206 loss 0.528 = 0.391 + 0.064 + 0.074 avg prob of [ goaltender] 0.6776474714279175 loss 0.404 = 0.265 + 0.063 + 0.076 avg prob of [ goaltender] 0.7684398889541626 loss 0.329 = 0.193 + 0.06 + 0.076 avg prob of [ goaltender] 0.8252369165420532 loss 0.276 = 0.144 + 0.056 + 0.076 avg prob of [ goaltender] 0.8662217259407043 loss 0.238 = 0.11 + 0.052 + 0.076 avg prob of [ goaltender] 0.8957602977752686 loss 0.21 = 0.087 + 0.047 + 0.076 avg prob of [ goaltender] 0.9171580076217651 loss 0.189 = 0.07 + 0.044 + 0.076 avg prob of [ goaltender] 0.932793378829956 loss 0.173 = 0.057 + 0.04 + 0.076 avg prob of [ goaltender] 0.9443090558052063 loss 0.161 = 0.048 + 0.037 + 0.076 avg prob of [ goaltender] 0.9528387784957886 loss 0.152 = 0.042 + 0.035 + 0.076 avg prob of [ goaltender] 0.9591806530952454 Delta norm: 105.29906463623047 Change in target norm: 26.324766159057617 to 108.20162200927734 => 81.8768539428711 Division Factor: 11.359480857849121 Right vector norm: 9.269707679748535 Computing left vector (u)... Selected u projection object The Bays Computing right vector (v) Lookup index found: 2 | Sentence: The Bays plays | Token: ays Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.557 = 10.557 + 0.0 + 0.0 avg prob of [ fantasy] 4.873280704487115e-05 loss 9.984 = 9.938 + 0.02 + 0.026 avg prob of [ fantasy] 9.828740439843386e-05 loss 9.199 = 9.138 + 0.019 + 0.042 avg prob of [ fantasy] 0.00022093964798841625 loss 7.78 = 7.685 + 0.038 + 0.057 avg prob of [ fantasy] 0.0010280599817633629 loss 6.053 = 5.856 + 0.125 + 0.071 avg prob of [ fantasy] 0.008317914791405201 loss 3.302 = 3.064 + 0.154 + 0.084 avg prob of [ fantasy] 0.08228553831577301 loss 0.878 = 0.603 + 0.18 + 0.095 avg prob of [ fantasy] 0.575442373752594 loss 0.44 = 0.13 + 0.207 + 0.102 avg prob of [ fantasy] 0.8801119923591614 loss 0.345 = 0.063 + 0.18 + 0.102 avg prob of [ fantasy] 0.939286470413208 loss 0.203 = 0.04 + 0.061 + 0.102 avg prob of [ fantasy] 0.9612816572189331 loss 0.151 = 0.03 + 0.019 + 0.102 avg prob of [ fantasy] 0.9704990386962891 loss 0.151 = 0.025 + 0.024 + 0.102 avg prob of [ fantasy] 0.9756776690483093 loss 0.149 = 0.021 + 0.026 + 0.102 avg prob of [ fantasy] 0.9793334007263184 loss 0.146 = 0.018 + 0.026 + 0.102 avg prob of [ fantasy] 0.982268214225769 loss 0.143 = 0.015 + 0.025 + 0.102 avg prob of [ fantasy] 0.9847564697265625 loss 0.138 = 0.013 + 0.023 + 0.102 avg prob of [ fantasy] 0.9868935346603394 loss 0.132 = 0.011 + 0.018 + 0.102 avg prob of [ fantasy] 0.9887159466743469 loss 0.124 = 0.01 + 0.012 + 0.102 avg prob of [ fantasy] 0.9902408123016357 loss 0.123 = 0.009 + 0.012 + 0.102 avg prob of [ fantasy] 0.9914800524711609 loss 0.128 = 0.008 + 0.018 + 0.102 avg prob of [ fantasy] 0.9924699068069458 Delta norm: 78.37970733642578 Change in target norm: 19.594926834106445 to 80.39436340332031 => 60.7994384765625 Division Factor: 9.951940536499023 Right vector norm: 7.875821590423584 seq_consistent {"n_edits": 20, "retention": 1.0, "unedited_true": 0.8, "max_update_norm": 9.729507446289062, "median_update_norm": 8.090650081634521} (582s) Computing left vector (u)... Selected u projection object Jari Kurri Computing right vector (v) Lookup index found: 3 | Sentence: Jari Kurri is a professional | Token: ri Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.75 = 4.75 + 0.0 + 0.0 avg prob of [ soccer] 0.016063496470451355 loss 1.848 = 1.816 + 0.009 + 0.023 avg prob of [ soccer] 0.23409998416900635 loss 1.16 = 1.111 + 0.013 + 0.036 avg prob of [ soccer] 0.46686339378356934 loss 0.566 = 0.505 + 0.014 + 0.046 avg prob of [ soccer] 0.7206315398216248 loss 0.34 = 0.27 + 0.014 + 0.056 avg prob of [ soccer] 0.8178364038467407 loss 0.236 = 0.157 + 0.014 + 0.065 avg prob of [ soccer] 0.8747534155845642 loss 0.187 = 0.1 + 0.014 + 0.073 avg prob of [ soccer] 0.9119874835014343 loss 0.164 = 0.069 + 0.015 + 0.08 avg prob of [ soccer] 0.9354063868522644 loss 0.152 = 0.051 + 0.015 + 0.086 avg prob of [ soccer] 0.9510470628738403 loss 0.146 = 0.039 + 0.016 + 0.092 avg prob of [ soccer] 0.9623677134513855 loss 0.142 = 0.03 + 0.016 + 0.095 avg prob of [ soccer] 0.9703485369682312 loss 0.136 = 0.025 + 0.016 + 0.095 avg prob of [ soccer] 0.9757582545280457 loss 0.132 = 0.02 + 0.016 + 0.095 avg prob of [ soccer] 0.9800354838371277 loss 0.128 = 0.017 + 0.016 + 0.095 avg prob of [ soccer] 0.9834456443786621 loss 0.125 = 0.014 + 0.016 + 0.095 avg prob of [ soccer] 0.9861986041069031 loss 0.122 = 0.012 + 0.015 + 0.095 avg prob of [ soccer] 0.9884461760520935 loss 0.12 = 0.01 + 0.015 + 0.095 avg prob of [ soccer] 0.9902974367141724 loss 0.118 = 0.008 + 0.014 + 0.095 avg prob of [ soccer] 0.9918344616889954 loss 0.116 = 0.007 + 0.014 + 0.095 avg prob of [ soccer] 0.9931256175041199 loss 0.114 = 0.006 + 0.013 + 0.095 avg prob of [ soccer] 0.9942261576652527 Delta norm: 83.86588287353516 Change in target norm: 20.966468811035156 to 86.85797119140625 => 65.8915023803711 Division Factor: 11.23317813873291 Right vector norm: 7.465908527374268 Computing left vector (u)... Selected u projection object Anthony Fisher Computing right vector (v) Lookup index found: 1 | Sentence: Anthony Fisher is a citizen of | Token: Fisher Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.3 = 8.3 + 0.0 + 0.0 avg prob of [ Malaysia] 0.0003600607451517135 loss 4.484 = 4.464 + 0.004 + 0.015 avg prob of [ Malaysia] 0.021277621388435364 loss 0.601 = 0.562 + 0.013 + 0.025 avg prob of [ Malaysia] 0.5857609510421753 loss 0.104 = 0.051 + 0.019 + 0.034 avg prob of [ Malaysia] 0.9505774974822998 loss 0.08 = 0.016 + 0.023 + 0.041 avg prob of [ Malaysia] 0.9840455651283264 loss 0.084 = 0.01 + 0.025 + 0.048 avg prob of [ Malaysia] 0.9897202849388123 loss 0.09 = 0.009 + 0.027 + 0.054 avg prob of [ Malaysia] 0.991047739982605 loss 0.096 = 0.009 + 0.028 + 0.059 avg prob of [ Malaysia] 0.991335928440094 loss 0.101 = 0.009 + 0.028 + 0.064 avg prob of [ Malaysia] 0.9913779497146606 loss 0.105 = 0.009 + 0.029 + 0.068 avg prob of [ Malaysia] 0.9914084076881409 loss 0.109 = 0.009 + 0.029 + 0.072 avg prob of [ Malaysia] 0.9914928674697876 loss 0.112 = 0.008 + 0.029 + 0.075 avg prob of [ Malaysia] 0.9916426539421082 loss 0.115 = 0.008 + 0.029 + 0.078 avg prob of [ Malaysia] 0.991875946521759 loss 0.113 = 0.007 + 0.028 + 0.078 avg prob of [ Malaysia] 0.9927033185958862 loss 0.112 = 0.007 + 0.027 + 0.078 avg prob of [ Malaysia] 0.9934399724006653 loss 0.11 = 0.006 + 0.027 + 0.078 avg prob of [ Malaysia] 0.9940934777259827 loss 0.109 = 0.005 + 0.026 + 0.078 avg prob of [ Malaysia] 0.9946708679199219 loss 0.108 = 0.005 + 0.025 + 0.078 avg prob of [ Malaysia] 0.9951799511909485 loss 0.106 = 0.004 + 0.024 + 0.078 avg prob of [ Malaysia] 0.9956281781196594 loss 0.105 = 0.004 + 0.023 + 0.078 avg prob of [ Malaysia] 0.9960222840309143 Delta norm: 102.970703125 Change in target norm: 25.742673873901367 to 107.36253356933594 => 81.61985778808594 Division Factor: 9.42315673828125 Right vector norm: 10.927410125732422 Computing left vector (u)... Selected u projection object 2 Pallas Computing right vector (v) Lookup index found: 2 | Sentence: 2 Pallas, which is named after | Token: allas Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 13.843 = 13.843 + 0.0 + 0.0 avg prob of [ Leicester] 2.3402603801514488e-06 loss 9.469 = 9.438 + 0.01 + 0.022 avg prob of [ Leicester] 0.00010198199743172154 loss 7.517 = 7.453 + 0.027 + 0.036 avg prob of [ Leicester] 0.0007036951137706637 loss 5.829 = 5.733 + 0.048 + 0.049 avg prob of [ Leicester] 0.003701537847518921 loss 4.693 = 4.569 + 0.064 + 0.06 avg prob of [ Leicester] 0.011164612136781216 loss 3.722 = 3.577 + 0.074 + 0.07 avg prob of [ Leicester] 0.029447896406054497 loss 2.725 = 2.568 + 0.077 + 0.08 avg prob of [ Leicester] 0.08034849911928177 loss 1.778 = 1.612 + 0.077 + 0.089 avg prob of [ Leicester] 0.20697852969169617 loss 1.072 = 0.904 + 0.075 + 0.093 avg prob of [ Leicester] 0.4140526056289673 loss 0.644 = 0.48 + 0.072 + 0.093 avg prob of [ Leicester] 0.6258448362350464 loss 0.408 = 0.245 + 0.069 + 0.093 avg prob of [ Leicester] 0.786031186580658 loss 0.29 = 0.129 + 0.068 + 0.093 avg prob of [ Leicester] 0.8804742097854614 loss 0.233 = 0.073 + 0.067 + 0.093 avg prob of [ Leicester] 0.9306845664978027 loss 0.204 = 0.044 + 0.066 + 0.093 avg prob of [ Leicester] 0.9570576548576355 loss 0.188 = 0.029 + 0.066 + 0.093 avg prob of [ Leicester] 0.9713714122772217 loss 0.18 = 0.021 + 0.065 + 0.093 avg prob of [ Leicester] 0.9795562028884888 loss 0.174 = 0.016 + 0.065 + 0.093 avg prob of [ Leicester] 0.9845194220542908 loss 0.171 = 0.012 + 0.065 + 0.093 avg prob of [ Leicester] 0.9877112507820129 loss 0.168 = 0.01 + 0.064 + 0.093 avg prob of [ Leicester] 0.9898821711540222 loss 0.166 = 0.009 + 0.064 + 0.093 avg prob of [ Leicester] 0.9914361238479614 Delta norm: 85.71553039550781 Change in target norm: 21.428882598876953 to 88.74213409423828 => 67.31324768066406 Division Factor: 10.056056022644043 Right vector norm: 8.523772239685059 Computing left vector (u)... Selected u projection object Georg Joseph Kamel Computing right vector (v) Lookup index found: 3 | Sentence: Georg Joseph Kamel, who works as | Token: el Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 13.821 = 13.821 + 0.0 + 0.0 avg prob of [ politician] 1.4644101611338556e-06 loss 12.655 = 12.634 + 0.004 + 0.017 avg prob of [ politician] 4.5761344154016115e-06 loss 10.555 = 10.489 + 0.038 + 0.028 avg prob of [ politician] 3.393253791728057e-05 loss 9.445 = 9.358 + 0.048 + 0.038 avg prob of [ politician] 9.202914225170389e-05 loss 8.039 = 7.935 + 0.059 + 0.046 avg prob of [ politician] 0.0003797170356847346 loss 6.537 = 6.419 + 0.065 + 0.053 avg prob of [ politician] 0.0017275932477787137 loss 4.957 = 4.833 + 0.065 + 0.059 avg prob of [ politician] 0.008460643701255322 loss 3.44 = 3.308 + 0.065 + 0.066 avg prob of [ politician] 0.03918521851301193 loss 1.668 = 1.53 + 0.066 + 0.072 avg prob of [ politician] 0.22280935943126678 loss 0.65 = 0.509 + 0.062 + 0.078 avg prob of [ politician] 0.604797899723053 loss 0.292 = 0.149 + 0.059 + 0.083 avg prob of [ politician] 0.8620461821556091 loss 0.202 = 0.062 + 0.057 + 0.083 avg prob of [ politician] 0.9403612017631531 loss 0.172 = 0.032 + 0.057 + 0.083 avg prob of [ politician] 0.9686371684074402 loss 0.16 = 0.02 + 0.056 + 0.083 avg prob of [ politician] 0.9799008965492249 loss 0.153 = 0.015 + 0.055 + 0.083 avg prob of [ politician] 0.9849355816841125 loss 0.15 = 0.013 + 0.054 + 0.083 avg prob of [ politician] 0.9873729944229126 loss 0.148 = 0.011 + 0.053 + 0.083 avg prob of [ politician] 0.9886221289634705 loss 0.147 = 0.011 + 0.053 + 0.083 avg prob of [ politician] 0.9893450140953064 loss 0.146 = 0.01 + 0.053 + 0.083 avg prob of [ politician] 0.9899032115936279 loss 0.146 = 0.01 + 0.053 + 0.083 avg prob of [ politician] 0.9904934763908386 Delta norm: 96.1525650024414 Change in target norm: 24.03814125061035 to 99.24723815917969 => 75.20909881591797 Division Factor: 13.112835884094238 Right vector norm: 7.332705497741699 Computing left vector (u)... Selected u projection object Berenberg Bank Computing right vector (v) Lookup index found: 3 | Sentence: Berenberg Bank, whose headquarters are in | Token: Bank Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.048 = 8.048 + 0.0 + 0.0 avg prob of [ Boston] 0.0004541490343399346 loss 6.391 = 6.36 + 0.002 + 0.028 avg prob of [ Boston] 0.002637034049257636 loss 4.521 = 4.468 + 0.006 + 0.047 avg prob of [ Boston] 0.01542349811643362 loss 3.014 = 2.938 + 0.012 + 0.064 avg prob of [ Boston] 0.05836307257413864 loss 2.102 = 2.004 + 0.019 + 0.079 avg prob of [ Boston] 0.15022344887256622 loss 0.819 = 0.704 + 0.023 + 0.092 avg prob of [ Boston] 0.5204939842224121 loss 0.205 = 0.074 + 0.026 + 0.105 avg prob of [ Boston] 0.9305709004402161 loss 0.143 = 0.008 + 0.029 + 0.106 avg prob of [ Boston] 0.992249071598053 loss 0.138 = 0.002 + 0.03 + 0.106 avg prob of [ Boston] 0.998030424118042 loss 0.139 = 0.001 + 0.031 + 0.106 avg prob of [ Boston] 0.9985121488571167 loss 0.139 = 0.001 + 0.031 + 0.106 avg prob of [ Boston] 0.998517632484436 loss 0.138 = 0.002 + 0.03 + 0.106 avg prob of [ Boston] 0.9983712434768677 loss 0.138 = 0.002 + 0.03 + 0.106 avg prob of [ Boston] 0.998166561126709 loss 0.137 = 0.002 + 0.029 + 0.106 avg prob of [ Boston] 0.997952401638031 loss 0.136 = 0.002 + 0.028 + 0.106 avg prob of [ Boston] 0.9977604746818542 loss 0.136 = 0.002 + 0.027 + 0.106 avg prob of [ Boston] 0.9976121783256531 loss 0.135 = 0.002 + 0.026 + 0.106 avg prob of [ Boston] 0.9975187182426453 loss 0.134 = 0.003 + 0.025 + 0.106 avg prob of [ Boston] 0.997484028339386 loss 0.133 = 0.003 + 0.025 + 0.106 avg prob of [ Boston] 0.9975051879882812 loss 0.132 = 0.002 + 0.024 + 0.106 avg prob of [ Boston] 0.9975752234458923 Delta norm: 75.34192657470703 Change in target norm: 18.835479736328125 to 78.93655395507812 => 60.10107421875 Division Factor: 8.687148094177246 Right vector norm: 8.672802925109863 Computing left vector (u)... Selected u projection object Athens Computing right vector (v) Lookup index found: 6 | Sentence: What is the twin city of Athens? It is | Token: Athens Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.986 = 12.986 + 0.0 + 0.0 avg prob of [ Prague] 4.793005700776121e-06 loss 10.329 = 10.292 + 0.003 + 0.034 avg prob of [ Prague] 6.783733988413587e-05 loss 8.006 = 7.935 + 0.009 + 0.063 avg prob of [ Prague] 0.0007135292980819941 loss 5.401 = 5.297 + 0.017 + 0.087 avg prob of [ Prague] 0.006540315691381693 loss 3.67 = 3.537 + 0.024 + 0.109 avg prob of [ Prague] 0.030348218977451324 loss 2.65 = 2.507 + 0.026 + 0.117 avg prob of [ Prague] 0.08513795584440231 loss 1.83 = 1.689 + 0.025 + 0.117 avg prob of [ Prague] 0.19408780336380005 loss 1.031 = 0.891 + 0.024 + 0.117 avg prob of [ Prague] 0.42129653692245483 loss 0.501 = 0.361 + 0.023 + 0.117 avg prob of [ Prague] 0.7010296583175659 loss 0.29 = 0.15 + 0.023 + 0.117 avg prob of [ Prague] 0.8615188598632812 loss 0.213 = 0.074 + 0.022 + 0.117 avg prob of [ Prague] 0.9289352893829346 loss 0.18 = 0.042 + 0.022 + 0.117 avg prob of [ Prague] 0.9593576788902283 loss 0.164 = 0.026 + 0.021 + 0.117 avg prob of [ Prague] 0.974676787853241 loss 0.154 = 0.017 + 0.02 + 0.117 avg prob of [ Prague] 0.9830633997917175 loss 0.149 = 0.012 + 0.02 + 0.117 avg prob of [ Prague] 0.9879730939865112 loss 0.144 = 0.009 + 0.019 + 0.117 avg prob of [ Prague] 0.9910158514976501 loss 0.141 = 0.007 + 0.018 + 0.117 avg prob of [ Prague] 0.9929969310760498 loss 0.139 = 0.006 + 0.017 + 0.117 avg prob of [ Prague] 0.9943434596061707 loss 0.137 = 0.005 + 0.016 + 0.117 avg prob of [ Prague] 0.995293915271759 loss 0.136 = 0.004 + 0.015 + 0.117 avg prob of [ Prague] 0.9959874749183655 Delta norm: 68.454833984375 Change in target norm: 17.113706588745117 to 71.98271942138672 => 54.86901092529297 Division Factor: 7.0796966552734375 Right vector norm: 9.669175148010254 Computing left vector (u)... Selected u projection object Vajiravudh Computing right vector (v) Lookup index found: 5 | Sentence: Vajiravudh writes in | Token: h Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.911 = 10.911 + 0.0 + 0.0 avg prob of [ Spanish] 3.879379801219329e-05 loss 10.142 = 10.083 + 0.034 + 0.025 avg prob of [ Spanish] 9.125267388299108e-05 loss 9.029 = 8.974 + 0.014 + 0.041 avg prob of [ Spanish] 0.000314459262881428 loss 6.227 = 6.115 + 0.058 + 0.054 avg prob of [ Spanish] 0.007610976696014404 loss 2.06 = 1.898 + 0.097 + 0.066 avg prob of [ Spanish] 0.2742915749549866 loss 0.529 = 0.313 + 0.138 + 0.078 avg prob of [ Spanish] 0.7433487176895142 loss 0.391 = 0.128 + 0.173 + 0.089 avg prob of [ Spanish] 0.8801355957984924 loss 0.364 = 0.082 + 0.182 + 0.1 avg prob of [ Spanish] 0.9215701222419739 loss 0.303 = 0.046 + 0.157 + 0.1 avg prob of [ Spanish] 0.9551143646240234 loss 0.26 = 0.026 + 0.134 + 0.1 avg prob of [ Spanish] 0.974595308303833 loss 0.239 = 0.015 + 0.123 + 0.1 avg prob of [ Spanish] 0.9846932291984558 loss 0.227 = 0.01 + 0.116 + 0.1 avg prob of [ Spanish] 0.990302562713623 loss 0.215 = 0.007 + 0.108 + 0.1 avg prob of [ Spanish] 0.993498682975769 loss 0.205 = 0.005 + 0.1 + 0.1 avg prob of [ Spanish] 0.995330810546875 loss 0.201 = 0.004 + 0.097 + 0.1 avg prob of [ Spanish] 0.996427059173584 loss 0.2 = 0.003 + 0.096 + 0.1 avg prob of [ Spanish] 0.997127890586853 loss 0.2 = 0.002 + 0.097 + 0.1 avg prob of [ Spanish] 0.9976043701171875 loss 0.2 = 0.002 + 0.097 + 0.1 avg prob of [ Spanish] 0.9979447722434998 loss 0.199 = 0.002 + 0.097 + 0.1 avg prob of [ Spanish] 0.9981982111930847 loss 0.197 = 0.002 + 0.095 + 0.1 avg prob of [ Spanish] 0.9983944892883301 Delta norm: 79.71402740478516 Change in target norm: 19.928508758544922 to 82.50908660888672 => 62.5805778503418 Division Factor: 10.259485244750977 Right vector norm: 7.769789218902588 Computing left vector (u)... Selected u projection object Iain Macleod Computing right vector (v) Lookup index found: 4 | Sentence: Iain Macleod found employment in | Token: od Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.792 = 7.792 + 0.0 + 0.0 avg prob of [ Berlin] 0.0005161409499123693 loss 7.258 = 7.24 + 0.001 + 0.018 avg prob of [ Berlin] 0.0008766801911406219 loss 6.623 = 6.591 + 0.004 + 0.029 avg prob of [ Berlin] 0.001687683747150004 loss 5.343 = 5.285 + 0.02 + 0.039 avg prob of [ Berlin] 0.0060511259362101555 loss 3.103 = 3.005 + 0.052 + 0.046 avg prob of [ Berlin] 0.06188209727406502 loss 0.52 = 0.37 + 0.097 + 0.053 avg prob of [ Berlin] 0.728010356426239 loss 0.238 = 0.05 + 0.129 + 0.06 avg prob of [ Berlin] 0.9527225494384766 loss 0.189 = 0.016 + 0.107 + 0.066 avg prob of [ Berlin] 0.9847025871276855 loss 0.168 = 0.008 + 0.087 + 0.073 avg prob of [ Berlin] 0.9918951392173767 loss 0.152 = 0.005 + 0.068 + 0.079 avg prob of [ Berlin] 0.9948354363441467 loss 0.142 = 0.004 + 0.054 + 0.084 avg prob of [ Berlin] 0.9964256286621094 loss 0.137 = 0.003 + 0.05 + 0.084 avg prob of [ Berlin] 0.9972520470619202 loss 0.138 = 0.002 + 0.052 + 0.084 avg prob of [ Berlin] 0.9976783990859985 loss 0.133 = 0.002 + 0.047 + 0.084 avg prob of [ Berlin] 0.9979062676429749 loss 0.124 = 0.002 + 0.038 + 0.084 avg prob of [ Berlin] 0.99809730052948 loss 0.115 = 0.002 + 0.029 + 0.084 avg prob of [ Berlin] 0.9983789324760437 loss 0.107 = 0.001 + 0.022 + 0.084 avg prob of [ Berlin] 0.9987243413925171 loss 0.107 = 0.001 + 0.022 + 0.084 avg prob of [ Berlin] 0.9990248680114746 loss 0.109 = 0.001 + 0.025 + 0.084 avg prob of [ Berlin] 0.9992367625236511 loss 0.108 = 0.001 + 0.024 + 0.084 avg prob of [ Berlin] 0.9993746876716614 Delta norm: 95.4317398071289 Change in target norm: 23.85793685913086 to 98.59400939941406 => 74.73606872558594 Division Factor: 12.65417766571045 Right vector norm: 7.541521072387695 Computing left vector (u)... Selected u projection object Grant Fuhr Computing right vector (v) Lookup index found: 3 | Sentence: Grant Fuhr plays in the position of | Token: r Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.333 = 5.333 + 0.0 + 0.0 avg prob of [ quarterback] 0.008335040882229805 loss 4.462 = 4.44 + 0.007 + 0.015 avg prob of [ quarterback] 0.01845853589475155 loss 4.056 = 4.028 + 0.004 + 0.024 avg prob of [ quarterback] 0.02736225351691246 loss 3.16 = 3.09 + 0.039 + 0.031 avg prob of [ quarterback] 0.0664949044585228 loss 1.773 = 1.67 + 0.065 + 0.038 avg prob of [ quarterback] 0.21541990339756012 loss 0.911 = 0.793 + 0.074 + 0.044 avg prob of [ quarterback] 0.46144232153892517 loss 0.629 = 0.495 + 0.084 + 0.05 avg prob of [ quarterback] 0.6169454455375671 loss 0.462 = 0.314 + 0.093 + 0.056 avg prob of [ quarterback] 0.7348459959030151 loss 0.356 = 0.194 + 0.101 + 0.061 avg prob of [ quarterback] 0.8254700899124146 loss 0.295 = 0.122 + 0.107 + 0.066 avg prob of [ quarterback] 0.8855131268501282 loss 0.263 = 0.082 + 0.11 + 0.07 avg prob of [ quarterback] 0.9210995435714722 loss 0.244 = 0.061 + 0.109 + 0.075 avg prob of [ quarterback] 0.9410961866378784 loss 0.229 = 0.048 + 0.104 + 0.077 avg prob of [ quarterback] 0.9528411030769348 loss 0.214 = 0.04 + 0.098 + 0.077 avg prob of [ quarterback] 0.9610491991043091 loss 0.202 = 0.033 + 0.092 + 0.077 avg prob of [ quarterback] 0.9672026634216309 loss 0.192 = 0.028 + 0.087 + 0.077 avg prob of [ quarterback] 0.9719887971878052 loss 0.184 = 0.025 + 0.083 + 0.077 avg prob of [ quarterback] 0.9757914543151855 loss 0.177 = 0.021 + 0.079 + 0.077 avg prob of [ quarterback] 0.9788526892662048 loss 0.172 = 0.019 + 0.076 + 0.077 avg prob of [ quarterback] 0.9813454151153564 loss 0.167 = 0.017 + 0.074 + 0.077 avg prob of [ quarterback] 0.9834011793136597 Delta norm: 104.3500747680664 Change in target norm: 26.08751678466797 to 109.18128204345703 => 83.09376525878906 Division Factor: 13.095809936523438 Right vector norm: 7.968202590942383 Computing left vector (u)... Selected u projection object Pare Glacier Computing right vector (v) Lookup index found: 2 | Sentence: Pare Glacier is located in | Token: Glacier Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.624 = 8.624 + 0.0 + 0.0 avg prob of [ Asia] 0.0002491172344889492 loss 6.433 = 6.407 + 0.002 + 0.025 avg prob of [ Asia] 0.002084203530102968 loss 4.582 = 4.535 + 0.005 + 0.043 avg prob of [ Asia] 0.013481216505169868 loss 2.528 = 2.46 + 0.01 + 0.058 avg prob of [ Asia] 0.10293581336736679 loss 1.071 = 0.981 + 0.017 + 0.073 avg prob of [ Asia] 0.4105030596256256 loss 0.488 = 0.376 + 0.025 + 0.087 avg prob of [ Asia] 0.7051379680633545 loss 0.291 = 0.161 + 0.03 + 0.1 avg prob of [ Asia] 0.8543924689292908 loss 0.226 = 0.098 + 0.029 + 0.1 avg prob of [ Asia] 0.9075424671173096 loss 0.191 = 0.064 + 0.027 + 0.1 avg prob of [ Asia] 0.9380273222923279 loss 0.17 = 0.045 + 0.025 + 0.1 avg prob of [ Asia] 0.9558790922164917 loss 0.157 = 0.034 + 0.024 + 0.1 avg prob of [ Asia] 0.9669824242591858 loss 0.148 = 0.026 + 0.022 + 0.1 avg prob of [ Asia] 0.9742934107780457 loss 0.142 = 0.021 + 0.021 + 0.1 avg prob of [ Asia] 0.9793307185173035 loss 0.137 = 0.017 + 0.02 + 0.1 avg prob of [ Asia] 0.9829310178756714 loss 0.133 = 0.015 + 0.019 + 0.1 avg prob of [ Asia] 0.9855870008468628 loss 0.131 = 0.012 + 0.019 + 0.1 avg prob of [ Asia] 0.9876030683517456 loss 0.129 = 0.011 + 0.018 + 0.1 avg prob of [ Asia] 0.9891740679740906 loss 0.127 = 0.01 + 0.018 + 0.1 avg prob of [ Asia] 0.9904278516769409 loss 0.125 = 0.009 + 0.017 + 0.1 avg prob of [ Asia] 0.9914500117301941 loss 0.124 = 0.008 + 0.017 + 0.1 avg prob of [ Asia] 0.99229896068573 Delta norm: 80.36063385009766 Change in target norm: 20.090158462524414 to 81.60314178466797 => 61.51298522949219 Division Factor: 7.756814956665039 Right vector norm: 10.360004425048828 seq_consistent {"n_edits": 30, "retention": 1.0, "unedited_true": 0.81, "max_update_norm": 10.254195213317871, "median_update_norm": 7.954639434814453} (858s) Computing left vector (u)... Selected u projection object Philip Hellquist Computing right vector (v) Lookup index found: 3 | Sentence: Philip Hellquist holds a citizenship from | Token: quist Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.808 = 5.808 + 0.0 + 0.0 avg prob of [ Austria] 0.003357410430908203 loss 4.077 = 4.052 + 0.001 + 0.024 avg prob of [ Austria] 0.020162884145975113 loss 3.11 = 3.068 + 0.003 + 0.038 avg prob of [ Austria] 0.04951755329966545 loss 2.691 = 2.636 + 0.005 + 0.05 avg prob of [ Austria] 0.0758000984787941 loss 2.174 = 2.106 + 0.007 + 0.061 avg prob of [ Austria] 0.1285879909992218 loss 1.511 = 1.429 + 0.01 + 0.071 avg prob of [ Austria] 0.2509978413581848 loss 0.878 = 0.782 + 0.015 + 0.081 avg prob of [ Austria] 0.4675767123699188 loss 0.476 = 0.368 + 0.018 + 0.09 avg prob of [ Austria] 0.6958895325660706 loss 0.282 = 0.165 + 0.019 + 0.098 avg prob of [ Austria] 0.8494141697883606 loss 0.194 = 0.076 + 0.02 + 0.098 avg prob of [ Austria] 0.9269235730171204 loss 0.153 = 0.036 + 0.02 + 0.098 avg prob of [ Austria] 0.9648472666740417 loss 0.135 = 0.018 + 0.019 + 0.098 avg prob of [ Austria] 0.9818556904792786 loss 0.127 = 0.01 + 0.019 + 0.098 avg prob of [ Austria] 0.9896578192710876 loss 0.123 = 0.007 + 0.018 + 0.098 avg prob of [ Austria] 0.9934931397438049 loss 0.12 = 0.004 + 0.018 + 0.098 avg prob of [ Austria] 0.9955319166183472 loss 0.118 = 0.003 + 0.017 + 0.098 avg prob of [ Austria] 0.9966980814933777 loss 0.116 = 0.003 + 0.016 + 0.098 avg prob of [ Austria] 0.9974095225334167 loss 0.115 = 0.002 + 0.015 + 0.098 avg prob of [ Austria] 0.9978682994842529 loss 0.114 = 0.002 + 0.014 + 0.098 avg prob of [ Austria] 0.9981778264045715 loss 0.113 = 0.002 + 0.014 + 0.098 avg prob of [ Austria] 0.9983947277069092 Delta norm: 81.81322479248047 Change in target norm: 20.45330810546875 to 86.15123748779297 => 65.69792938232422 Division Factor: 8.76500129699707 Right vector norm: 9.334080696105957 Computing left vector (u)... Selected u projection object Marie-Arlette Carlotti Computing right vector (v) Lookup index found: 5 | Sentence: Marie-Arlette Carlotti is a native speaker of | Token: otti Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.281 = 5.281 + 0.0 + 0.0 avg prob of [ Dutch] 0.005412163678556681 loss 2.957 = 2.934 + 0.002 + 0.021 avg prob of [ Dutch] 0.060106948018074036 loss 1.716 = 1.681 + 0.003 + 0.033 avg prob of [ Dutch] 0.2073669135570526 loss 0.734 = 0.686 + 0.004 + 0.043 avg prob of [ Dutch] 0.5303575992584229 loss 0.401 = 0.342 + 0.006 + 0.053 avg prob of [ Dutch] 0.7239435315132141 loss 0.278 = 0.208 + 0.008 + 0.062 avg prob of [ Dutch] 0.817983090877533 loss 0.224 = 0.143 + 0.011 + 0.07 avg prob of [ Dutch] 0.8692095875740051 loss 0.199 = 0.106 + 0.014 + 0.078 avg prob of [ Dutch] 0.9003334045410156 loss 0.186 = 0.084 + 0.017 + 0.085 avg prob of [ Dutch] 0.9203301668167114 loss 0.179 = 0.069 + 0.019 + 0.091 avg prob of [ Dutch] 0.933780312538147 loss 0.167 = 0.057 + 0.019 + 0.091 avg prob of [ Dutch] 0.9446442723274231 loss 0.157 = 0.048 + 0.019 + 0.091 avg prob of [ Dutch] 0.9535797834396362 loss 0.149 = 0.04 + 0.018 + 0.091 avg prob of [ Dutch] 0.9608990550041199 loss 0.141 = 0.034 + 0.017 + 0.091 avg prob of [ Dutch] 0.9668847918510437 loss 0.135 = 0.029 + 0.015 + 0.091 avg prob of [ Dutch] 0.9717854857444763 loss 0.129 = 0.025 + 0.014 + 0.091 avg prob of [ Dutch] 0.975812554359436 loss 0.124 = 0.021 + 0.012 + 0.091 avg prob of [ Dutch] 0.9791379570960999 loss 0.12 = 0.018 + 0.011 + 0.091 avg prob of [ Dutch] 0.9819004535675049 loss 0.116 = 0.016 + 0.01 + 0.091 avg prob of [ Dutch] 0.984208345413208 loss 0.113 = 0.014 + 0.009 + 0.091 avg prob of [ Dutch] 0.9861471056938171 Delta norm: 88.34248352050781 Change in target norm: 22.085622787475586 to 92.00415802001953 => 69.91853332519531 Division Factor: 8.683645248413086 Right vector norm: 10.173434257507324 Computing left vector (u)... Selected u projection object Ruokolahti Computing right vector (v) Lookup index found: 5 | Sentence: In Ruokolahti, the language spoken is | Token: ti Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.842 = 8.842 + 0.0 + 0.0 avg prob of [ French] 0.00015545167843811214 loss 6.55 = 6.522 + 0.008 + 0.019 avg prob of [ French] 0.0016093077138066292 loss 5.005 = 4.96 + 0.012 + 0.033 avg prob of [ French] 0.007802783511579037 loss 2.781 = 2.718 + 0.019 + 0.044 avg prob of [ French] 0.07548978179693222 loss 1.69 = 1.615 + 0.021 + 0.054 avg prob of [ French] 0.21638701856136322 loss 1.195 = 1.109 + 0.023 + 0.063 avg prob of [ French] 0.34373655915260315 loss 0.897 = 0.801 + 0.025 + 0.071 avg prob of [ French] 0.4582337737083435 loss 0.674 = 0.568 + 0.027 + 0.079 avg prob of [ French] 0.5718778967857361 loss 0.504 = 0.388 + 0.029 + 0.087 avg prob of [ French] 0.6807422041893005 loss 0.367 = 0.25 + 0.029 + 0.088 avg prob of [ French] 0.7799862623214722 loss 0.267 = 0.151 + 0.028 + 0.088 avg prob of [ French] 0.8607363700866699 loss 0.204 = 0.088 + 0.028 + 0.088 avg prob of [ French] 0.9159172773361206 loss 0.168 = 0.053 + 0.027 + 0.088 avg prob of [ French] 0.9487823247909546 loss 0.149 = 0.033 + 0.027 + 0.088 avg prob of [ French] 0.9672989845275879 loss 0.138 = 0.022 + 0.028 + 0.088 avg prob of [ French] 0.9778145551681519 loss 0.132 = 0.016 + 0.028 + 0.088 avg prob of [ French] 0.9840285181999207 loss 0.129 = 0.012 + 0.028 + 0.088 avg prob of [ French] 0.987899124622345 loss 0.127 = 0.01 + 0.029 + 0.088 avg prob of [ French] 0.9904476404190063 loss 0.125 = 0.008 + 0.029 + 0.088 avg prob of [ French] 0.9922175407409668 loss 0.124 = 0.007 + 0.029 + 0.088 avg prob of [ French] 0.9935065507888794 Delta norm: 90.63675689697266 Change in target norm: 22.65918731689453 to 94.05073547363281 => 71.39154815673828 Division Factor: 10.15747356414795 Right vector norm: 8.923158645629883 Computing left vector (u)... Selected u projection object Neko Entertainment Computing right vector (v) Lookup index found: 3 | Sentence: Neko Entertainment was formed in | Token: Entertainment Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.51 = 10.51 + 0.0 + 0.0 avg prob of [ Sydney] 3.64655461453367e-05 loss 7.648 = 7.612 + 0.006 + 0.03 avg prob of [ Sydney] 0.0005583345773629844 loss 6.03 = 5.968 + 0.011 + 0.051 avg prob of [ Sydney] 0.003172782249748707 loss 4.154 = 4.066 + 0.019 + 0.068 avg prob of [ Sydney] 0.02076135389506817 loss 2.58 = 2.464 + 0.031 + 0.085 avg prob of [ Sydney] 0.09249212592840195 loss 1.779 = 1.636 + 0.042 + 0.101 avg prob of [ Sydney] 0.2000134289264679 loss 1.235 = 1.079 + 0.047 + 0.11 avg prob of [ Sydney] 0.34643054008483887 loss 0.782 = 0.625 + 0.048 + 0.11 avg prob of [ Sydney] 0.5426887273788452 loss 0.469 = 0.311 + 0.048 + 0.11 avg prob of [ Sydney] 0.7376219034194946 loss 0.314 = 0.157 + 0.047 + 0.11 avg prob of [ Sydney] 0.8568097352981567 loss 0.243 = 0.089 + 0.044 + 0.11 avg prob of [ Sydney] 0.9152164459228516 loss 0.206 = 0.057 + 0.039 + 0.11 avg prob of [ Sydney] 0.9452412724494934 loss 0.182 = 0.038 + 0.034 + 0.11 avg prob of [ Sydney] 0.9625935554504395 loss 0.167 = 0.027 + 0.03 + 0.11 avg prob of [ Sydney] 0.9734183549880981 loss 0.156 = 0.02 + 0.027 + 0.11 avg prob of [ Sydney] 0.9805011749267578 loss 0.149 = 0.015 + 0.025 + 0.11 avg prob of [ Sydney] 0.985318124294281 loss 0.144 = 0.011 + 0.023 + 0.11 avg prob of [ Sydney] 0.988711416721344 loss 0.14 = 0.009 + 0.022 + 0.11 avg prob of [ Sydney] 0.9911745190620422 loss 0.137 = 0.007 + 0.021 + 0.11 avg prob of [ Sydney] 0.9930016398429871 loss 0.135 = 0.006 + 0.02 + 0.11 avg prob of [ Sydney] 0.9943737983703613 Delta norm: 72.9213638305664 Change in target norm: 18.2303409576416 to 75.95149230957031 => 57.721153259277344 Division Factor: 7.50040340423584 Right vector norm: 9.722325325012207 Computing left vector (u)... Selected u projection object Nina van Pallandt Computing right vector (v) Lookup index found: 5 | Sentence: Nina van Pallandt originated from | Token: t Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.833 = 8.833 + 0.0 + 0.0 avg prob of [ Minneapolis] 0.0001662930444581434 loss 6.922 = 6.889 + 0.007 + 0.025 avg prob of [ Minneapolis] 0.001260139630176127 loss 5.407 = 5.345 + 0.022 + 0.04 avg prob of [ Minneapolis] 0.005578183569014072 loss 3.512 = 3.439 + 0.02 + 0.054 avg prob of [ Minneapolis] 0.033119816333055496 loss 1.449 = 1.352 + 0.031 + 0.066 avg prob of [ Minneapolis] 0.2652265131473541 loss 0.506 = 0.385 + 0.043 + 0.078 avg prob of [ Minneapolis] 0.6869498491287231 loss 0.277 = 0.137 + 0.05 + 0.09 avg prob of [ Minneapolis] 0.8740333318710327 loss 0.211 = 0.058 + 0.053 + 0.1 avg prob of [ Minneapolis] 0.9442009925842285 loss 0.182 = 0.027 + 0.054 + 0.1 avg prob of [ Minneapolis] 0.9730764031410217 loss 0.17 = 0.015 + 0.055 + 0.1 avg prob of [ Minneapolis] 0.9854511022567749 loss 0.163 = 0.009 + 0.054 + 0.1 avg prob of [ Minneapolis] 0.9910972118377686 loss 0.157 = 0.006 + 0.051 + 0.1 avg prob of [ Minneapolis] 0.9939182996749878 loss 0.152 = 0.005 + 0.047 + 0.1 avg prob of [ Minneapolis] 0.9954569935798645 loss 0.147 = 0.004 + 0.043 + 0.1 avg prob of [ Minneapolis] 0.9963641166687012 loss 0.142 = 0.003 + 0.039 + 0.1 avg prob of [ Minneapolis] 0.9969366788864136 loss 0.138 = 0.003 + 0.036 + 0.1 avg prob of [ Minneapolis] 0.9973207116127014 loss 0.136 = 0.002 + 0.033 + 0.1 avg prob of [ Minneapolis] 0.9975929260253906 loss 0.133 = 0.002 + 0.031 + 0.1 avg prob of [ Minneapolis] 0.9977959394454956 loss 0.132 = 0.002 + 0.029 + 0.1 avg prob of [ Minneapolis] 0.9979552030563354 loss 0.13 = 0.002 + 0.028 + 0.1 avg prob of [ Minneapolis] 0.9980853796005249 Delta norm: 79.9053955078125 Change in target norm: 19.976350784301758 to 83.2605972290039 => 63.28424835205078 Division Factor: 9.177937507629395 Right vector norm: 8.70624828338623 Computing left vector (u)... Selected u projection object Bernhard Knipperdolling Computing right vector (v) Lookup index found: 6 | Sentence: Bernhard Knipperdolling, who holds the position of | Token: ing Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.074 = 11.074 + 0.0 + 0.0 avg prob of [ cardinal] 1.9682463971548714e-05 loss 10.597 = 10.567 + 0.011 + 0.019 avg prob of [ cardinal] 3.42695893778e-05 loss 9.886 = 9.84 + 0.015 + 0.031 avg prob of [ cardinal] 7.209392788354307e-05 loss 7.223 = 7.165 + 0.017 + 0.04 avg prob of [ cardinal] 0.0010001526679843664 loss 4.473 = 4.391 + 0.032 + 0.049 avg prob of [ cardinal] 0.013212249614298344 loss 3.359 = 3.251 + 0.051 + 0.057 avg prob of [ cardinal] 0.04123961925506592 loss 2.068 = 1.93 + 0.073 + 0.065 avg prob of [ cardinal] 0.1532289683818817 loss 1.08 = 0.919 + 0.089 + 0.072 avg prob of [ cardinal] 0.4081050455570221 loss 0.533 = 0.356 + 0.097 + 0.079 avg prob of [ cardinal] 0.7033867835998535 loss 0.334 = 0.145 + 0.102 + 0.086 avg prob of [ cardinal] 0.8653934001922607 loss 0.262 = 0.077 + 0.097 + 0.088 avg prob of [ cardinal] 0.9260411858558655 loss 0.215 = 0.045 + 0.082 + 0.088 avg prob of [ cardinal] 0.9559727907180786 loss 0.179 = 0.027 + 0.064 + 0.088 avg prob of [ cardinal] 0.9733515381813049 loss 0.16 = 0.016 + 0.056 + 0.088 avg prob of [ cardinal] 0.9837235808372498 loss 0.155 = 0.01 + 0.057 + 0.088 avg prob of [ cardinal] 0.9896188378334045 loss 0.154 = 0.007 + 0.059 + 0.088 avg prob of [ cardinal] 0.9926995635032654 loss 0.154 = 0.006 + 0.06 + 0.088 avg prob of [ cardinal] 0.9943091869354248 loss 0.152 = 0.005 + 0.06 + 0.088 avg prob of [ cardinal] 0.9953036308288574 loss 0.151 = 0.004 + 0.059 + 0.088 avg prob of [ cardinal] 0.9960306882858276 loss 0.149 = 0.003 + 0.058 + 0.088 avg prob of [ cardinal] 0.9966022968292236 Delta norm: 91.19434356689453 Change in target norm: 22.798587799072266 to 95.8887939453125 => 73.0902099609375 Division Factor: 9.570873260498047 Right vector norm: 9.5283203125 Computing left vector (u)... Selected u projection object Sylvia Bataille Computing right vector (v) Lookup index found: 5 | Sentence: Sylvia Bataille died at | Token: le Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 16.179 = 16.179 + 0.0 + 0.0 avg prob of [ Idaho] 1.1230771690406982e-07 loss 12.735 = 12.719 + 0.002 + 0.013 avg prob of [ Idaho] 3.476730171314557e-06 loss 10.072 = 10.044 + 0.006 + 0.022 avg prob of [ Idaho] 5.555773896048777e-05 loss 8.001 = 7.954 + 0.016 + 0.03 avg prob of [ Idaho] 0.00044242129661142826 loss 6.495 = 6.424 + 0.035 + 0.037 avg prob of [ Idaho] 0.001990377437323332 loss 4.738 = 4.635 + 0.061 + 0.042 avg prob of [ Idaho] 0.011341501027345657 loss 2.316 = 2.189 + 0.08 + 0.047 avg prob of [ Idaho] 0.13317495584487915 loss 0.443 = 0.315 + 0.077 + 0.051 avg prob of [ Idaho] 0.7626938819885254 loss 0.205 = 0.092 + 0.058 + 0.056 avg prob of [ Idaho] 0.9197270274162292 loss 0.115 = 0.012 + 0.043 + 0.06 avg prob of [ Idaho] 0.9883228540420532 loss 0.106 = 0.004 + 0.038 + 0.064 avg prob of [ Idaho] 0.9959263801574707 loss 0.106 = 0.003 + 0.035 + 0.067 avg prob of [ Idaho] 0.9970453381538391 loss 0.107 = 0.003 + 0.033 + 0.071 avg prob of [ Idaho] 0.996979832649231 loss 0.107 = 0.004 + 0.031 + 0.072 avg prob of [ Idaho] 0.9962629675865173 loss 0.106 = 0.005 + 0.029 + 0.072 avg prob of [ Idaho] 0.9948080778121948 loss 0.107 = 0.007 + 0.027 + 0.072 avg prob of [ Idaho] 0.9927724003791809 loss 0.108 = 0.009 + 0.026 + 0.072 avg prob of [ Idaho] 0.9905907511711121 loss 0.109 = 0.011 + 0.025 + 0.072 avg prob of [ Idaho] 0.989041805267334 loss 0.108 = 0.011 + 0.025 + 0.072 avg prob of [ Idaho] 0.9887872338294983 loss 0.107 = 0.01 + 0.024 + 0.072 avg prob of [ Idaho] 0.9898288249969482 Delta norm: 110.53600311279297 Change in target norm: 27.633996963500977 to 115.10861206054688 => 87.47461700439453 Division Factor: 11.328210830688477 Right vector norm: 9.757585525512695 Computing left vector (u)... Selected u projection object Elton Brand Computing right vector (v) Lookup index found: 2 | Sentence: Elton Brand professionally plays the sport | Token: Brand Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.299 = 10.299 + 0.0 + 0.0 avg prob of [ baseball] 4.79214359074831e-05 loss 8.681 = 8.663 + 0.003 + 0.015 avg prob of [ baseball] 0.00024589887470938265 loss 5.747 = 5.713 + 0.008 + 0.026 avg prob of [ baseball] 0.004616119898855686 loss 3.005 = 2.955 + 0.015 + 0.035 avg prob of [ baseball] 0.06323466449975967 loss 1.568 = 1.502 + 0.024 + 0.043 avg prob of [ baseball] 0.25512102246284485 loss 0.486 = 0.406 + 0.03 + 0.05 avg prob of [ baseball] 0.687487781047821 loss 0.201 = 0.112 + 0.032 + 0.057 avg prob of [ baseball] 0.8959043622016907 loss 0.165 = 0.067 + 0.035 + 0.063 avg prob of [ baseball] 0.9367676973342896 loss 0.162 = 0.054 + 0.039 + 0.069 avg prob of [ baseball] 0.9482146501541138 loss 0.159 = 0.042 + 0.043 + 0.074 avg prob of [ baseball] 0.9598268866539001 loss 0.153 = 0.028 + 0.046 + 0.078 avg prob of [ baseball] 0.9722942113876343 loss 0.14 = 0.019 + 0.043 + 0.078 avg prob of [ baseball] 0.9811937212944031 loss 0.131 = 0.013 + 0.04 + 0.078 avg prob of [ baseball] 0.9871726036071777 loss 0.125 = 0.009 + 0.038 + 0.078 avg prob of [ baseball] 0.9908917546272278 loss 0.121 = 0.007 + 0.037 + 0.078 avg prob of [ baseball] 0.9931881427764893 loss 0.118 = 0.005 + 0.035 + 0.078 avg prob of [ baseball] 0.9946436882019043 loss 0.116 = 0.004 + 0.034 + 0.078 avg prob of [ baseball] 0.9956029057502747 loss 0.114 = 0.004 + 0.032 + 0.078 avg prob of [ baseball] 0.9962611198425293 loss 0.112 = 0.003 + 0.031 + 0.078 avg prob of [ baseball] 0.9967299699783325 loss 0.111 = 0.003 + 0.03 + 0.078 avg prob of [ baseball] 0.9970754981040955 Delta norm: 102.64618682861328 Change in target norm: 25.66154670715332 to 107.42916107177734 => 81.76761627197266 Division Factor: 11.037250518798828 Right vector norm: 9.299978256225586 Computing left vector (u)... Selected u projection object Vadis Odjidja-Ofoe Computing right vector (v) Lookup index found: 9 | Sentence: Vadis Odjidja-Ofoe plays in the position of | Token: oe Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.45 = 8.45 + 0.0 + 0.0 avg prob of [ linebacker] 0.0003242893726564944 loss 8.333 = 8.114 + 0.204 + 0.016 avg prob of [ linebacker] 0.00045227404916659 loss 8.073 = 8.036 + 0.013 + 0.023 avg prob of [ linebacker] 0.000485245487652719 loss 7.92 = 7.883 + 0.007 + 0.03 avg prob of [ linebacker] 0.0005540872807614505 loss 7.546 = 7.502 + 0.008 + 0.037 avg prob of [ linebacker] 0.0007825765060260892 loss 6.536 = 6.482 + 0.011 + 0.043 avg prob of [ linebacker] 0.0020733673591166735 loss 5.174 = 5.105 + 0.02 + 0.049 avg prob of [ linebacker] 0.0075493548065423965 loss 4.509 = 4.413 + 0.041 + 0.055 avg prob of [ linebacker] 0.013575510121881962 loss 4.214 = 4.083 + 0.069 + 0.062 avg prob of [ linebacker] 0.017947113141417503 loss 3.727 = 3.571 + 0.088 + 0.068 avg prob of [ linebacker] 0.03000633791089058 loss 3.038 = 2.875 + 0.089 + 0.074 avg prob of [ linebacker] 0.05986585095524788 loss 2.33 = 2.176 + 0.074 + 0.08 avg prob of [ linebacker] 0.1179373636841774 loss 1.795 = 1.651 + 0.063 + 0.08 avg prob of [ linebacker] 0.1971006840467453 loss 1.303 = 1.168 + 0.054 + 0.08 avg prob of [ linebacker] 0.31748753786087036 loss 0.854 = 0.726 + 0.048 + 0.08 avg prob of [ linebacker] 0.4901246428489685 loss 0.526 = 0.403 + 0.043 + 0.08 avg prob of [ linebacker] 0.6722607612609863 loss 0.336 = 0.216 + 0.04 + 0.08 avg prob of [ linebacker] 0.8071833848953247 loss 0.233 = 0.118 + 0.035 + 0.08 avg prob of [ linebacker] 0.8898861408233643 loss 0.178 = 0.067 + 0.031 + 0.08 avg prob of [ linebacker] 0.9358042478561401 loss 0.151 = 0.041 + 0.03 + 0.08 avg prob of [ linebacker] 0.9604139924049377 Delta norm: 99.68559265136719 Change in target norm: 24.921396255493164 to 102.5855941772461 => 77.66419982910156 Division Factor: 11.886787414550781 Right vector norm: 8.386251449584961 Computing left vector (u)... Selected u projection object Jacques Cazotte Computing right vector (v) Lookup index found: 4 | Sentence: Jacques Cazotte, a native | Token: otte Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.674 = 9.674 + 0.0 + 0.0 avg prob of [ Italian] 7.570751768071204e-05 loss 4.939 = 4.917 + 0.011 + 0.01 avg prob of [ Italian] 0.008342080749571323 loss 2.008 = 1.969 + 0.022 + 0.017 avg prob of [ Italian] 0.1554233282804489 loss 1.043 = 0.985 + 0.036 + 0.023 avg prob of [ Italian] 0.39027342200279236 loss 0.714 = 0.641 + 0.045 + 0.028 avg prob of [ Italian] 0.5386728644371033 loss 0.478 = 0.399 + 0.046 + 0.033 avg prob of [ Italian] 0.6775084137916565 loss 0.329 = 0.249 + 0.043 + 0.037 avg prob of [ Italian] 0.7829293012619019 loss 0.244 = 0.163 + 0.04 + 0.041 avg prob of [ Italian] 0.8513841032981873 loss 0.193 = 0.112 + 0.037 + 0.044 avg prob of [ Italian] 0.894818902015686 loss 0.163 = 0.081 + 0.035 + 0.048 avg prob of [ Italian] 0.9228753447532654 loss 0.144 = 0.06 + 0.033 + 0.05 avg prob of [ Italian] 0.9414787292480469 loss 0.132 = 0.047 + 0.032 + 0.053 avg prob of [ Italian] 0.9541909098625183 loss 0.125 = 0.038 + 0.032 + 0.055 avg prob of [ Italian] 0.9631508588790894 loss 0.12 = 0.031 + 0.032 + 0.058 avg prob of [ Italian] 0.9696500897407532 loss 0.117 = 0.026 + 0.032 + 0.06 avg prob of [ Italian] 0.9744901061058044 loss 0.115 = 0.022 + 0.032 + 0.061 avg prob of [ Italian] 0.9781842231750488 loss 0.114 = 0.019 + 0.032 + 0.063 avg prob of [ Italian] 0.9810696840286255 loss 0.113 = 0.017 + 0.032 + 0.064 avg prob of [ Italian] 0.9833715558052063 loss 0.11 = 0.015 + 0.031 + 0.064 avg prob of [ Italian] 0.9852492809295654 loss 0.108 = 0.013 + 0.03 + 0.064 avg prob of [ Italian] 0.9868043065071106 Delta norm: 124.22486114501953 Change in target norm: 31.056217193603516 to 128.88861083984375 => 97.8323974609375 Division Factor: 10.029826164245605 Right vector norm: 12.38554573059082 seq_consistent {"n_edits": 40, "retention": 1.0, "unedited_true": 0.81, "max_update_norm": 12.035945892333984, "median_update_norm": 8.329229354858398} (1146s) Computing left vector (u)... Selected u projection object The Enchanter Reborn Computing right vector (v) Lookup index found: 4 | Sentence: The Enchanter Reborn is known for performing | Token: Reborn Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.244 = 12.244 + 0.0 + 0.0 avg prob of [ jazz] 6.698119250359014e-06 loss 9.511 = 9.482 + 0.009 + 0.019 avg prob of [ jazz] 9.513743862044066e-05 loss 8.163 = 8.111 + 0.019 + 0.033 avg prob of [ jazz] 0.00033620544127188623 loss 6.686 = 6.613 + 0.029 + 0.044 avg prob of [ jazz] 0.0014806436374783516 loss 5.003 = 4.918 + 0.03 + 0.055 avg prob of [ jazz] 0.008003802970051765 loss 3.463 = 3.367 + 0.032 + 0.064 avg prob of [ jazz] 0.036520909518003464 loss 2.309 = 2.198 + 0.038 + 0.074 avg prob of [ jazz] 0.11417321115732193 loss 1.322 = 1.193 + 0.047 + 0.083 avg prob of [ jazz] 0.30695244669914246 loss 0.672 = 0.532 + 0.052 + 0.088 avg prob of [ jazz] 0.5896700620651245 loss 0.392 = 0.256 + 0.048 + 0.088 avg prob of [ jazz] 0.7751967310905457 loss 0.268 = 0.14 + 0.04 + 0.088 avg prob of [ jazz] 0.8695240020751953 loss 0.209 = 0.09 + 0.03 + 0.088 avg prob of [ jazz] 0.914038360118866 loss 0.177 = 0.065 + 0.024 + 0.088 avg prob of [ jazz] 0.9375260472297668 loss 0.158 = 0.049 + 0.02 + 0.088 avg prob of [ jazz] 0.9518765807151794 loss 0.146 = 0.039 + 0.018 + 0.088 avg prob of [ jazz] 0.9616791605949402 loss 0.137 = 0.032 + 0.017 + 0.088 avg prob of [ jazz] 0.9688550233840942 loss 0.13 = 0.026 + 0.016 + 0.088 avg prob of [ jazz] 0.9743062853813171 loss 0.126 = 0.022 + 0.016 + 0.088 avg prob of [ jazz] 0.9785062670707703 loss 0.122 = 0.018 + 0.015 + 0.088 avg prob of [ jazz] 0.9817515015602112 loss 0.119 = 0.016 + 0.015 + 0.088 avg prob of [ jazz] 0.9842718839645386 Delta norm: 90.70601654052734 Change in target norm: 22.676504135131836 to 94.81057739257812 => 72.13407135009766 Division Factor: 10.181941986083984 Right vector norm: 8.90851879119873 Computing left vector (u)... Selected u projection object Pierre Lescure Computing right vector (v) Lookup index found: 3 | Sentence: Pierre Lescure, speaker of | Token: ure Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.542 = 12.542 + 0.0 + 0.0 avg prob of [ Hebrew] 5.558739576372318e-05 loss 7.882 = 7.849 + 0.018 + 0.015 avg prob of [ Hebrew] 0.00115483277477324 loss 2.249 = 2.203 + 0.022 + 0.024 avg prob of [ Hebrew] 0.11947347223758698 loss 0.927 = 0.864 + 0.031 + 0.032 avg prob of [ Hebrew] 0.4316098690032959 loss 0.371 = 0.3 + 0.032 + 0.039 avg prob of [ Hebrew] 0.7440220713615417 loss 0.195 = 0.125 + 0.025 + 0.046 avg prob of [ Hebrew] 0.8836688995361328 loss 0.139 = 0.067 + 0.02 + 0.052 avg prob of [ Hebrew] 0.9355933666229248 loss 0.116 = 0.039 + 0.019 + 0.057 avg prob of [ Hebrew] 0.9614638090133667 loss 0.106 = 0.025 + 0.019 + 0.062 avg prob of [ Hebrew] 0.9753143787384033 loss 0.102 = 0.017 + 0.019 + 0.066 avg prob of [ Hebrew] 0.9831040501594543 loss 0.101 = 0.012 + 0.019 + 0.07 avg prob of [ Hebrew] 0.9877243041992188 loss 0.101 = 0.009 + 0.019 + 0.073 avg prob of [ Hebrew] 0.9906118512153625 loss 0.102 = 0.008 + 0.018 + 0.076 avg prob of [ Hebrew] 0.9925036430358887 loss 0.103 = 0.006 + 0.018 + 0.079 avg prob of [ Hebrew] 0.9937558174133301 loss 0.102 = 0.005 + 0.017 + 0.079 avg prob of [ Hebrew] 0.9945583343505859 loss 0.1 = 0.005 + 0.017 + 0.079 avg prob of [ Hebrew] 0.995192289352417 loss 0.099 = 0.004 + 0.016 + 0.079 avg prob of [ Hebrew] 0.9957016706466675 loss 0.099 = 0.004 + 0.016 + 0.079 avg prob of [ Hebrew] 0.9961169362068176 loss 0.098 = 0.004 + 0.016 + 0.079 avg prob of [ Hebrew] 0.9964596629142761 loss 0.097 = 0.003 + 0.015 + 0.079 avg prob of [ Hebrew] 0.996745228767395 Delta norm: 101.66441345214844 Change in target norm: 25.41610336303711 to 107.79149627685547 => 82.37539672851562 Division Factor: 10.492090225219727 Right vector norm: 9.689624786376953 Computing left vector (u)... Selected u projection object Breaux Bridge Computing right vector (v) Lookup index found: 2 | Sentence: Breaux Bridge is within | Token: Bridge Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 14.405 = 14.405 + 0.0 + 0.0 avg prob of [ Bihar] 5.832657734572422e-07 loss 12.219 = 12.198 + 0.006 + 0.015 avg prob of [ Bihar] 6.540574304381153e-06 loss 8.592 = 8.552 + 0.015 + 0.025 avg prob of [ Bihar] 0.0004939230275340378 loss 2.32 = 2.265 + 0.023 + 0.032 avg prob of [ Bihar] 0.14911314845085144 loss 1.318 = 1.242 + 0.038 + 0.038 avg prob of [ Bihar] 0.36173751950263977 loss 0.636 = 0.542 + 0.05 + 0.045 avg prob of [ Bihar] 0.6257832646369934 loss 0.298 = 0.191 + 0.056 + 0.05 avg prob of [ Bihar] 0.8347128033638 loss 0.193 = 0.077 + 0.061 + 0.056 avg prob of [ Bihar] 0.9278470277786255 loss 0.164 = 0.039 + 0.064 + 0.06 avg prob of [ Bihar] 0.9623879194259644 loss 0.155 = 0.024 + 0.066 + 0.065 avg prob of [ Bihar] 0.9765652418136597 loss 0.153 = 0.017 + 0.067 + 0.069 avg prob of [ Bihar] 0.9832814335823059 loss 0.152 = 0.013 + 0.066 + 0.073 avg prob of [ Bihar] 0.9868658781051636 loss 0.151 = 0.011 + 0.064 + 0.076 avg prob of [ Bihar] 0.988959789276123 loss 0.149 = 0.01 + 0.061 + 0.078 avg prob of [ Bihar] 0.9903796911239624 loss 0.145 = 0.009 + 0.058 + 0.078 avg prob of [ Bihar] 0.9914738535881042 loss 0.142 = 0.008 + 0.056 + 0.078 avg prob of [ Bihar] 0.9922702312469482 loss 0.139 = 0.007 + 0.053 + 0.078 avg prob of [ Bihar] 0.9928698539733887 loss 0.136 = 0.007 + 0.051 + 0.078 avg prob of [ Bihar] 0.9933362603187561 loss 0.134 = 0.006 + 0.049 + 0.078 avg prob of [ Bihar] 0.9937113523483276 loss 0.132 = 0.006 + 0.048 + 0.078 avg prob of [ Bihar] 0.9940240979194641 Delta norm: 102.45745086669922 Change in target norm: 25.614364624023438 to 109.35630798339844 => 83.741943359375 Division Factor: 6.916705131530762 Right vector norm: 14.813042640686035 Computing left vector (u)... Selected u projection object Jaiku Computing right vector (v) Lookup index found: 1 | Sentence: Jaiku owner | Token: iku Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.659 = 10.659 + 0.0 + 0.0 avg prob of [ Lion] 3.7549074477283284e-05 loss 7.888 = 7.855 + 0.02 + 0.013 avg prob of [ Lion] 0.0006134597351774573 loss 3.925 = 3.86 + 0.043 + 0.021 avg prob of [ Lion] 0.027073686942458153 loss 0.846 = 0.765 + 0.053 + 0.029 avg prob of [ Lion] 0.4850090444087982 loss 0.182 = 0.087 + 0.059 + 0.036 avg prob of [ Lion] 0.9176102876663208 loss 0.13 = 0.026 + 0.062 + 0.042 avg prob of [ Lion] 0.9747446775436401 loss 0.125 = 0.012 + 0.065 + 0.048 avg prob of [ Lion] 0.9878036975860596 loss 0.128 = 0.008 + 0.067 + 0.053 avg prob of [ Lion] 0.9923890233039856 loss 0.13 = 0.005 + 0.067 + 0.057 avg prob of [ Lion] 0.9946158528327942 loss 0.129 = 0.004 + 0.064 + 0.061 avg prob of [ Lion] 0.9959271550178528 loss 0.127 = 0.003 + 0.058 + 0.065 avg prob of [ Lion] 0.9967684149742126 loss 0.125 = 0.003 + 0.054 + 0.068 avg prob of [ Lion] 0.9973199963569641 loss 0.124 = 0.002 + 0.05 + 0.071 avg prob of [ Lion] 0.9976751208305359 loss 0.12 = 0.002 + 0.047 + 0.071 avg prob of [ Lion] 0.9978542327880859 loss 0.117 = 0.002 + 0.044 + 0.071 avg prob of [ Lion] 0.9979808330535889 loss 0.116 = 0.002 + 0.043 + 0.071 avg prob of [ Lion] 0.9980730414390564 loss 0.114 = 0.002 + 0.041 + 0.071 avg prob of [ Lion] 0.9981431365013123 loss 0.112 = 0.002 + 0.039 + 0.071 avg prob of [ Lion] 0.9981993436813354 loss 0.11 = 0.002 + 0.038 + 0.071 avg prob of [ Lion] 0.9982478022575378 loss 0.109 = 0.002 + 0.036 + 0.071 avg prob of [ Lion] 0.9982919096946716 Delta norm: 112.570556640625 Change in target norm: 28.142641067504883 to 119.4468994140625 => 91.30426025390625 Division Factor: 8.519647598266602 Right vector norm: 13.213054656982422 Computing left vector (u)... Selected u projection object Europa Europa Computing right vector (v) Lookup index found: 2 | Sentence: Europa Europa, that was developed in | Token: Europa Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 6.468 = 6.468 + 0.0 + 0.0 avg prob of [ Belgium] 0.0027151559479534626 loss 5.924 = 5.888 + 0.024 + 0.012 avg prob of [ Belgium] 0.003920740447938442 loss 5.367 = 5.317 + 0.03 + 0.019 avg prob of [ Belgium] 0.006456851959228516 loss 4.238 = 4.164 + 0.049 + 0.026 avg prob of [ Belgium] 0.019445529207587242 loss 2.87 = 2.781 + 0.057 + 0.032 avg prob of [ Belgium] 0.06926991045475006 loss 1.799 = 1.699 + 0.061 + 0.038 avg prob of [ Belgium] 0.19067348539829254 loss 0.908 = 0.794 + 0.071 + 0.044 avg prob of [ Belgium] 0.4595993459224701 loss 0.389 = 0.258 + 0.082 + 0.049 avg prob of [ Belgium] 0.7749090790748596 loss 0.23 = 0.087 + 0.088 + 0.055 avg prob of [ Belgium] 0.9174559712409973 loss 0.191 = 0.041 + 0.091 + 0.06 avg prob of [ Belgium] 0.9602913856506348 loss 0.179 = 0.024 + 0.091 + 0.065 avg prob of [ Belgium] 0.9765005707740784 loss 0.174 = 0.016 + 0.089 + 0.068 avg prob of [ Belgium] 0.9839893579483032 loss 0.168 = 0.012 + 0.088 + 0.068 avg prob of [ Belgium] 0.9878751039505005 loss 0.165 = 0.009 + 0.087 + 0.068 avg prob of [ Belgium] 0.9906013011932373 loss 0.162 = 0.007 + 0.087 + 0.068 avg prob of [ Belgium] 0.9925377368927002 loss 0.16 = 0.006 + 0.086 + 0.068 avg prob of [ Belgium] 0.9939317107200623 loss 0.157 = 0.005 + 0.084 + 0.068 avg prob of [ Belgium] 0.9949485063552856 loss 0.154 = 0.004 + 0.081 + 0.068 avg prob of [ Belgium] 0.9956988096237183 loss 0.151 = 0.004 + 0.079 + 0.068 avg prob of [ Belgium] 0.9962586760520935 loss 0.149 = 0.003 + 0.077 + 0.068 avg prob of [ Belgium] 0.9966813325881958 Delta norm: 117.71656036376953 Change in target norm: 29.42913818359375 to 123.8003158569336 => 94.37117767333984 Division Factor: 6.9614057540893555 Right vector norm: 16.90988540649414 Computing left vector (u)... Selected u projection object Louis-Jean-Marie Daubenton Computing right vector (v) Lookup index found: 7 | Sentence: Louis-Jean-Marie Daubenton is a native speaker of | Token: enton Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.351 = 5.351 + 0.0 + 0.0 avg prob of [ Russian] 0.005093546584248543 loss 1.294 = 1.272 + 0.007 + 0.016 avg prob of [ Russian] 0.2942817509174347 loss 0.639 = 0.601 + 0.011 + 0.026 avg prob of [ Russian] 0.5528414249420166 loss 0.291 = 0.243 + 0.012 + 0.035 avg prob of [ Russian] 0.7855150103569031 loss 0.164 = 0.108 + 0.013 + 0.043 avg prob of [ Russian] 0.8977010846138 loss 0.124 = 0.062 + 0.012 + 0.05 avg prob of [ Russian] 0.9402914643287659 loss 0.112 = 0.043 + 0.012 + 0.056 avg prob of [ Russian] 0.9576532244682312 loss 0.11 = 0.034 + 0.014 + 0.062 avg prob of [ Russian] 0.9662604331970215 loss 0.11 = 0.029 + 0.015 + 0.067 avg prob of [ Russian] 0.9716182947158813 loss 0.112 = 0.025 + 0.016 + 0.071 avg prob of [ Russian] 0.9755885004997253 loss 0.113 = 0.021 + 0.016 + 0.075 avg prob of [ Russian] 0.9788079261779785 loss 0.113 = 0.019 + 0.016 + 0.079 avg prob of [ Russian] 0.9815027117729187 loss 0.111 = 0.016 + 0.015 + 0.08 avg prob of [ Russian] 0.9839320182800293 loss 0.108 = 0.014 + 0.014 + 0.08 avg prob of [ Russian] 0.98590087890625 loss 0.106 = 0.013 + 0.014 + 0.08 avg prob of [ Russian] 0.9874616265296936 loss 0.104 = 0.011 + 0.013 + 0.08 avg prob of [ Russian] 0.9887081384658813 loss 0.102 = 0.01 + 0.012 + 0.08 avg prob of [ Russian] 0.9897173643112183 loss 0.101 = 0.009 + 0.012 + 0.08 avg prob of [ Russian] 0.9905498027801514 loss 0.1 = 0.009 + 0.012 + 0.08 avg prob of [ Russian] 0.9912499189376831 loss 0.099 = 0.008 + 0.012 + 0.08 avg prob of [ Russian] 0.9918515086174011 Delta norm: 100.54164123535156 Change in target norm: 25.13541030883789 to 104.51780700683594 => 79.38240051269531 Division Factor: 9.267975807189941 Right vector norm: 10.848284721374512 Computing left vector (u)... Selected u projection object Yi-Fu Tuan Computing right vector (v) Lookup index found: 5 | Sentence: Yi-Fu Tuan's expertise is | Token: uan Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.918 = 9.918 + 0.0 + 0.0 avg prob of [ singing] 0.0001340379676548764 loss 9.432 = 9.398 + 0.001 + 0.033 avg prob of [ singing] 0.00025170197477564216 loss 8.533 = 8.476 + 0.003 + 0.054 avg prob of [ singing] 0.0007882327772676945 loss 5.842 = 5.757 + 0.014 + 0.072 avg prob of [ singing] 0.01294758077710867 loss 2.123 = 1.997 + 0.038 + 0.088 avg prob of [ singing] 0.2344731092453003 loss 0.776 = 0.616 + 0.056 + 0.104 avg prob of [ singing] 0.5579255223274231 loss 0.467 = 0.291 + 0.061 + 0.115 avg prob of [ singing] 0.7530597448348999 loss 0.312 = 0.132 + 0.065 + 0.115 avg prob of [ singing] 0.8768442869186401 loss 0.252 = 0.069 + 0.068 + 0.115 avg prob of [ singing] 0.9331375956535339 loss 0.229 = 0.044 + 0.07 + 0.115 avg prob of [ singing] 0.9569045305252075 loss 0.216 = 0.032 + 0.069 + 0.115 avg prob of [ singing] 0.9684838652610779 loss 0.207 = 0.025 + 0.066 + 0.115 avg prob of [ singing] 0.9750103950500488 loss 0.198 = 0.021 + 0.063 + 0.115 avg prob of [ singing] 0.9791575074195862 loss 0.192 = 0.018 + 0.059 + 0.115 avg prob of [ singing] 0.9820828437805176 loss 0.187 = 0.016 + 0.056 + 0.115 avg prob of [ singing] 0.9843342304229736 loss 0.183 = 0.014 + 0.054 + 0.115 avg prob of [ singing] 0.9861814379692078 loss 0.179 = 0.012 + 0.052 + 0.115 avg prob of [ singing] 0.9877577424049377 loss 0.176 = 0.011 + 0.05 + 0.115 avg prob of [ singing] 0.989128589630127 loss 0.174 = 0.01 + 0.049 + 0.115 avg prob of [ singing] 0.9903252124786377 loss 0.171 = 0.009 + 0.048 + 0.115 avg prob of [ singing] 0.9913665056228638 Delta norm: 69.62310791015625 Change in target norm: 17.405778884887695 to 71.98844909667969 => 54.582672119140625 Division Factor: 7.210412502288818 Right vector norm: 9.655912399291992 Computing left vector (u)... Selected u projection object The Loner Computing right vector (v) Lookup index found: 2 | Sentence: The Loner was released on | Token: oner Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.079 = 10.079 + 0.0 + 0.0 avg prob of [ HBO] 0.00010894323349930346 loss 9.375 = 9.354 + 0.006 + 0.015 avg prob of [ HBO] 0.00026737485313788056 loss 7.795 = 7.745 + 0.026 + 0.024 avg prob of [ HBO] 0.001246836269274354 loss 4.743 = 4.648 + 0.064 + 0.031 avg prob of [ HBO] 0.01563899591565132 loss 1.78 = 1.658 + 0.084 + 0.038 avg prob of [ HBO] 0.2832958698272705 loss 0.493 = 0.348 + 0.101 + 0.044 avg prob of [ HBO] 0.7637499570846558 loss 0.208 = 0.048 + 0.11 + 0.05 avg prob of [ HBO] 0.955532431602478 loss 0.164 = 0.006 + 0.102 + 0.056 avg prob of [ HBO] 0.9936108589172363 loss 0.149 = 0.002 + 0.085 + 0.062 avg prob of [ HBO] 0.998476505279541 loss 0.141 = 0.001 + 0.074 + 0.067 avg prob of [ HBO] 0.9993332624435425 loss 0.141 = 0.0 + 0.069 + 0.071 avg prob of [ HBO] 0.9995335340499878 loss 0.145 = 0.0 + 0.069 + 0.076 avg prob of [ HBO] 0.9995676875114441 loss 0.146 = 0.0 + 0.069 + 0.077 avg prob of [ HBO] 0.9995684623718262 loss 0.146 = 0.0 + 0.069 + 0.077 avg prob of [ HBO] 0.9995700120925903 loss 0.146 = 0.0 + 0.068 + 0.077 avg prob of [ HBO] 0.9995668530464172 loss 0.145 = 0.0 + 0.067 + 0.077 avg prob of [ HBO] 0.9995622038841248 loss 0.143 = 0.0 + 0.066 + 0.077 avg prob of [ HBO] 0.9995580315589905 loss 0.142 = 0.0 + 0.065 + 0.077 avg prob of [ HBO] 0.9995549321174622 loss 0.141 = 0.0 + 0.064 + 0.077 avg prob of [ HBO] 0.9995535612106323 loss 0.14 = 0.0 + 0.062 + 0.077 avg prob of [ HBO] 0.9995535016059875 Delta norm: 103.96314239501953 Change in target norm: 25.990787506103516 to 111.3470230102539 => 85.35623168945312 Division Factor: 9.844855308532715 Right vector norm: 10.560149192810059 Computing left vector (u)... Selected u projection object Beyond Hatred Computing right vector (v) Lookup index found: 5 | Sentence: The language of Beyond Hatred was | Token: red Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.947 = 12.947 + 0.0 + 0.0 avg prob of [ Hindi] 4.576835635816678e-06 loss 10.906 = 10.881 + 0.007 + 0.018 avg prob of [ Hindi] 4.121243182453327e-05 loss 7.821 = 7.775 + 0.016 + 0.03 avg prob of [ Hindi] 0.0007910811109468341 loss 3.957 = 3.887 + 0.03 + 0.041 avg prob of [ Hindi] 0.028692657127976418 loss 1.996 = 1.9 + 0.044 + 0.052 avg prob of [ Hindi] 0.1649390459060669 loss 1.263 = 1.153 + 0.049 + 0.061 avg prob of [ Hindi] 0.324077844619751 loss 0.835 = 0.718 + 0.048 + 0.07 avg prob of [ Hindi] 0.4946451783180237 loss 0.446 = 0.322 + 0.047 + 0.077 avg prob of [ Hindi] 0.7263861298561096 loss 0.283 = 0.152 + 0.047 + 0.084 avg prob of [ Hindi] 0.8594103455543518 loss 0.215 = 0.087 + 0.044 + 0.085 avg prob of [ Hindi] 0.917113721370697 loss 0.182 = 0.056 + 0.042 + 0.085 avg prob of [ Hindi] 0.9459971189498901 loss 0.163 = 0.039 + 0.039 + 0.085 avg prob of [ Hindi] 0.9616475105285645 loss 0.151 = 0.029 + 0.037 + 0.085 avg prob of [ Hindi] 0.9709674119949341 loss 0.143 = 0.023 + 0.035 + 0.085 avg prob of [ Hindi] 0.9770946502685547 loss 0.136 = 0.019 + 0.033 + 0.085 avg prob of [ Hindi] 0.9814713001251221 loss 0.131 = 0.015 + 0.031 + 0.085 avg prob of [ Hindi] 0.9847781658172607 loss 0.126 = 0.013 + 0.029 + 0.085 avg prob of [ Hindi] 0.9873551726341248 loss 0.122 = 0.011 + 0.027 + 0.085 avg prob of [ Hindi] 0.9893920421600342 loss 0.119 = 0.009 + 0.025 + 0.085 avg prob of [ Hindi] 0.9910109639167786 loss 0.115 = 0.008 + 0.023 + 0.085 avg prob of [ Hindi] 0.9923019409179688 Delta norm: 94.60309600830078 Change in target norm: 23.650775909423828 to 98.94186401367188 => 75.29109191894531 Division Factor: 10.781982421875 Right vector norm: 8.774184226989746 Computing left vector (u)... Selected u projection object Cao Yunding Computing right vector (v) Lookup index found: 4 | Sentence: Cao Yunding was native to | Token: ing Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.494 = 12.494 + 0.0 + 0.0 avg prob of [ Dublin] 4.946219178236788e-06 loss 9.625 = 9.598 + 0.013 + 0.014 avg prob of [ Dublin] 7.777439896017313e-05 loss 8.58 = 8.545 + 0.011 + 0.023 avg prob of [ Dublin] 0.0002346603141631931 loss 7.236 = 7.189 + 0.016 + 0.031 avg prob of [ Dublin] 0.0009961200412362814 loss 5.162 = 5.091 + 0.034 + 0.038 avg prob of [ Dublin] 0.008848822675645351 loss 2.646 = 2.532 + 0.07 + 0.043 avg prob of [ Dublin] 0.10168040543794632 loss 1.208 = 1.045 + 0.114 + 0.049 avg prob of [ Dublin] 0.37751591205596924 loss 0.485 = 0.326 + 0.105 + 0.054 avg prob of [ Dublin] 0.7327392101287842 loss 0.223 = 0.07 + 0.094 + 0.059 avg prob of [ Dublin] 0.9343785643577576 loss 0.168 = 0.018 + 0.086 + 0.064 avg prob of [ Dublin] 0.9824469685554504 loss 0.156 = 0.007 + 0.081 + 0.068 avg prob of [ Dublin] 0.9925860166549683 loss 0.155 = 0.005 + 0.078 + 0.073 avg prob of [ Dublin] 0.995141863822937 loss 0.152 = 0.004 + 0.073 + 0.075 avg prob of [ Dublin] 0.9959543943405151 loss 0.143 = 0.004 + 0.064 + 0.075 avg prob of [ Dublin] 0.996398389339447 loss 0.132 = 0.003 + 0.054 + 0.075 avg prob of [ Dublin] 0.9966756105422974 loss 0.123 = 0.003 + 0.045 + 0.075 avg prob of [ Dublin] 0.99689120054245 loss 0.116 = 0.003 + 0.038 + 0.075 avg prob of [ Dublin] 0.9970893859863281 loss 0.11 = 0.003 + 0.032 + 0.075 avg prob of [ Dublin] 0.9972865581512451 loss 0.106 = 0.003 + 0.028 + 0.075 avg prob of [ Dublin] 0.9974852800369263 loss 0.102 = 0.002 + 0.024 + 0.075 avg prob of [ Dublin] 0.9976816773414612 Delta norm: 106.56461334228516 Change in target norm: 26.64115333557129 to 110.42256927490234 => 83.78141784667969 Division Factor: 9.501596450805664 Right vector norm: 11.215441703796387 seq_consistent {"n_edits": 50, "retention": 1.0, "unedited_true": 0.83, "max_update_norm": 13.033635139465332, "median_update_norm": 8.63382339477539} (1416s) Computing left vector (u)... Selected u projection object Embrik Strand Computing right vector (v) Lookup index found: 3 | Sentence: Embrik Strand found employment in | Token: and Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.781 = 7.781 + 0.0 + 0.0 avg prob of [ Dublin] 0.0004682935541495681 loss 5.494 = 5.461 + 0.018 + 0.015 avg prob of [ Dublin] 0.0053467536345124245 loss 1.995 = 1.918 + 0.055 + 0.022 avg prob of [ Dublin] 0.17014102637767792 loss 0.908 = 0.806 + 0.073 + 0.03 avg prob of [ Dublin] 0.4666544497013092 loss 0.386 = 0.266 + 0.084 + 0.036 avg prob of [ Dublin] 0.770154595375061 loss 0.254 = 0.116 + 0.095 + 0.042 avg prob of [ Dublin] 0.891143798828125 loss 0.225 = 0.078 + 0.099 + 0.048 avg prob of [ Dublin] 0.9250810742378235 loss 0.21 = 0.063 + 0.095 + 0.052 avg prob of [ Dublin] 0.9388486742973328 loss 0.193 = 0.048 + 0.088 + 0.057 avg prob of [ Dublin] 0.9533873200416565 loss 0.176 = 0.033 + 0.082 + 0.061 avg prob of [ Dublin] 0.9673840403556824 loss 0.162 = 0.023 + 0.075 + 0.064 avg prob of [ Dublin] 0.9777689576148987 loss 0.151 = 0.016 + 0.067 + 0.068 avg prob of [ Dublin] 0.9844464063644409 loss 0.142 = 0.012 + 0.06 + 0.071 avg prob of [ Dublin] 0.9885363578796387 loss 0.136 = 0.009 + 0.054 + 0.073 avg prob of [ Dublin] 0.9910710453987122 loss 0.132 = 0.007 + 0.049 + 0.076 avg prob of [ Dublin] 0.9927196502685547 loss 0.128 = 0.006 + 0.045 + 0.077 avg prob of [ Dublin] 0.9938815832138062 loss 0.124 = 0.005 + 0.042 + 0.077 avg prob of [ Dublin] 0.9947192668914795 loss 0.122 = 0.005 + 0.04 + 0.077 avg prob of [ Dublin] 0.9953585863113403 loss 0.12 = 0.004 + 0.039 + 0.077 avg prob of [ Dublin] 0.9958741068840027 loss 0.118 = 0.004 + 0.038 + 0.077 avg prob of [ Dublin] 0.996307373046875 Delta norm: 104.14683532714844 Change in target norm: 26.036706924438477 to 111.12706756591797 => 85.09036254882812 Division Factor: 12.548563003540039 Right vector norm: 8.2995023727417 Computing left vector (u)... Selected u projection object Ruth Marshall Computing right vector (v) Lookup index found: 2 | Sentence: Ruth Marshall's occupation is | Token: Marshall Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.561 = 10.561 + 0.0 + 0.0 avg prob of [ politician] 4.292883750167675e-05 loss 7.382 = 7.358 + 0.012 + 0.012 avg prob of [ politician] 0.0013707064790651202 loss 5.631 = 5.573 + 0.038 + 0.019 avg prob of [ politician] 0.006247840356081724 loss 4.191 = 4.12 + 0.045 + 0.026 avg prob of [ politician] 0.025446604937314987 loss 2.994 = 2.913 + 0.049 + 0.032 avg prob of [ politician] 0.07912927865982056 loss 1.945 = 1.854 + 0.054 + 0.038 avg prob of [ politician] 0.21140454709529877 loss 1.134 = 1.036 + 0.055 + 0.043 avg prob of [ politician] 0.44073373079299927 loss 0.661 = 0.55 + 0.062 + 0.048 avg prob of [ politician] 0.6658977270126343 loss 0.411 = 0.282 + 0.076 + 0.053 avg prob of [ politician] 0.815855860710144 loss 0.276 = 0.128 + 0.09 + 0.058 avg prob of [ politician] 0.9021031260490417 loss 0.213 = 0.053 + 0.097 + 0.062 avg prob of [ politician] 0.9521517157554626 loss 0.189 = 0.026 + 0.097 + 0.066 avg prob of [ politician] 0.9749668836593628 loss 0.173 = 0.017 + 0.087 + 0.069 avg prob of [ politician] 0.9837039709091187 loss 0.152 = 0.013 + 0.07 + 0.069 avg prob of [ politician] 0.9871509075164795 loss 0.134 = 0.011 + 0.054 + 0.069 avg prob of [ politician] 0.9892644286155701 loss 0.12 = 0.01 + 0.042 + 0.069 avg prob of [ politician] 0.9907018542289734 loss 0.111 = 0.008 + 0.033 + 0.069 avg prob of [ politician] 0.991762638092041 loss 0.105 = 0.008 + 0.028 + 0.069 avg prob of [ politician] 0.9925858378410339 loss 0.102 = 0.007 + 0.026 + 0.069 avg prob of [ politician] 0.9932408332824707 loss 0.1 = 0.006 + 0.024 + 0.069 avg prob of [ politician] 0.9937699437141418 Delta norm: 115.70899963378906 Change in target norm: 28.927249908447266 to 122.07828521728516 => 93.15103149414062 Division Factor: 8.902196884155273 Right vector norm: 12.997804641723633 Computing left vector (u)... Selected u projection object Le Devoir Computing right vector (v) Lookup index found: 5 | Sentence: The language of Le Devoir was | Token: oir Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 13.515 = 13.515 + 0.0 + 0.0 avg prob of [ Tamil] 2.2858262127556372e-06 loss 10.109 = 10.097 + 0.006 + 0.006 avg prob of [ Tamil] 6.291729368967935e-05 loss 6.639 = 6.618 + 0.012 + 0.009 avg prob of [ Tamil] 0.0025120743084698915 loss 3.689 = 3.657 + 0.019 + 0.013 avg prob of [ Tamil] 0.05017625540494919 loss 2.542 = 2.502 + 0.025 + 0.016 avg prob of [ Tamil] 0.11731840670108795 loss 1.599 = 1.55 + 0.031 + 0.019 avg prob of [ Tamil] 0.24729740619659424 loss 0.913 = 0.854 + 0.037 + 0.021 avg prob of [ Tamil] 0.45057806372642517 loss 0.516 = 0.448 + 0.044 + 0.024 avg prob of [ Tamil] 0.6492742300033569 loss 0.327 = 0.251 + 0.05 + 0.026 avg prob of [ Tamil] 0.7814142107963562 loss 0.234 = 0.153 + 0.053 + 0.028 avg prob of [ Tamil] 0.8596370220184326 loss 0.182 = 0.098 + 0.054 + 0.03 avg prob of [ Tamil] 0.9073207378387451 loss 0.151 = 0.065 + 0.055 + 0.031 avg prob of [ Tamil] 0.9376262426376343 loss 0.131 = 0.044 + 0.055 + 0.033 avg prob of [ Tamil] 0.9575746655464172 loss 0.118 = 0.03 + 0.054 + 0.034 avg prob of [ Tamil] 0.9708283543586731 loss 0.108 = 0.021 + 0.051 + 0.036 avg prob of [ Tamil] 0.9794629216194153 loss 0.102 = 0.015 + 0.049 + 0.037 avg prob of [ Tamil] 0.9849026203155518 loss 0.098 = 0.012 + 0.048 + 0.038 avg prob of [ Tamil] 0.9881808161735535 loss 0.096 = 0.01 + 0.046 + 0.039 avg prob of [ Tamil] 0.9900103211402893 loss 0.095 = 0.009 + 0.046 + 0.04 avg prob of [ Tamil] 0.9908679127693176 loss 0.095 = 0.009 + 0.045 + 0.041 avg prob of [ Tamil] 0.9910991191864014 Delta norm: 142.11268615722656 Change in target norm: 41.74093246459961 to 151.41818237304688 => 109.67724609375 Division Factor: 11.448381423950195 Right vector norm: 12.413342475891113 Computing left vector (u)... Selected u projection object Jesse Hibbs Computing right vector (v) Lookup index found: 3 | Sentence: Jesse Hibbs is a professional | Token: bs Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 3.051 = 3.051 + 0.0 + 0.0 avg prob of [ baseball] 0.06727216392755508 loss 0.209 = 0.182 + 0.016 + 0.011 avg prob of [ baseball] 0.8570460081100464 loss 0.122 = 0.041 + 0.063 + 0.018 avg prob of [ baseball] 0.960710346698761 loss 0.116 = 0.015 + 0.078 + 0.023 avg prob of [ baseball] 0.9852783679962158 loss 0.099 = 0.007 + 0.064 + 0.028 avg prob of [ baseball] 0.9930674433708191 loss 0.081 = 0.004 + 0.045 + 0.032 avg prob of [ baseball] 0.9956277012825012 loss 0.071 = 0.003 + 0.032 + 0.035 avg prob of [ baseball] 0.9965601563453674 loss 0.067 = 0.003 + 0.026 + 0.038 avg prob of [ baseball] 0.9969033002853394 loss 0.067 = 0.003 + 0.024 + 0.041 avg prob of [ baseball] 0.9970011711120605 loss 0.069 = 0.003 + 0.022 + 0.043 avg prob of [ baseball] 0.9969806671142578 loss 0.07 = 0.003 + 0.021 + 0.046 avg prob of [ baseball] 0.9969000816345215 loss 0.071 = 0.003 + 0.02 + 0.047 avg prob of [ baseball] 0.9967910051345825 loss 0.071 = 0.003 + 0.019 + 0.049 avg prob of [ baseball] 0.9966813325881958 loss 0.072 = 0.003 + 0.018 + 0.051 avg prob of [ baseball] 0.9966017603874207 loss 0.073 = 0.003 + 0.017 + 0.052 avg prob of [ baseball] 0.9965816736221313 loss 0.073 = 0.003 + 0.017 + 0.053 avg prob of [ baseball] 0.9966377019882202 loss 0.074 = 0.003 + 0.016 + 0.054 avg prob of [ baseball] 0.996767520904541 loss 0.074 = 0.003 + 0.016 + 0.055 avg prob of [ baseball] 0.9969528913497925 loss 0.074 = 0.003 + 0.015 + 0.056 avg prob of [ baseball] 0.9971684217453003 loss 0.074 = 0.003 + 0.014 + 0.057 avg prob of [ baseball] 0.997388482093811 Delta norm: 105.39228057861328 Change in target norm: 30.345727920532227 to 112.19322967529297 => 81.84750366210938 Division Factor: 9.607198715209961 Right vector norm: 10.970135688781738 Computing left vector (u)... Selected u projection object Marcus Rojas Computing right vector (v) Lookup index found: 6 | Sentence: The genre played by Marcus Rojas is | Token: jas Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.484 = 10.484 + 0.0 + 0.0 avg prob of [ opera] 4.8013298510340974e-05 loss 9.5 = 9.48 + 0.004 + 0.016 avg prob of [ opera] 0.00013981184747535735 loss 8.584 = 8.546 + 0.012 + 0.027 avg prob of [ opera] 0.0003592993598431349 loss 7.149 = 7.092 + 0.021 + 0.036 avg prob of [ opera] 0.0014445711858570576 loss 4.91 = 4.832 + 0.033 + 0.045 avg prob of [ opera] 0.010687212459743023 loss 3.258 = 3.158 + 0.046 + 0.054 avg prob of [ opera] 0.04760267585515976 loss 1.492 = 1.374 + 0.056 + 0.061 avg prob of [ opera] 0.2660682201385498 loss 0.418 = 0.273 + 0.077 + 0.068 avg prob of [ opera] 0.7634230256080627 loss 0.253 = 0.076 + 0.102 + 0.075 avg prob of [ opera] 0.9270545244216919 loss 0.231 = 0.039 + 0.112 + 0.08 avg prob of [ opera] 0.9618666768074036 loss 0.195 = 0.029 + 0.087 + 0.08 avg prob of [ opera] 0.9716225862503052 loss 0.17 = 0.023 + 0.067 + 0.08 avg prob of [ opera] 0.9772139191627502 loss 0.155 = 0.019 + 0.056 + 0.08 avg prob of [ opera] 0.980937659740448 loss 0.147 = 0.017 + 0.05 + 0.08 avg prob of [ opera] 0.9835655689239502 loss 0.142 = 0.015 + 0.047 + 0.08 avg prob of [ opera] 0.9855813384056091 loss 0.138 = 0.013 + 0.045 + 0.08 avg prob of [ opera] 0.9872778058052063 loss 0.134 = 0.011 + 0.043 + 0.08 avg prob of [ opera] 0.9887924194335938 loss 0.132 = 0.01 + 0.042 + 0.08 avg prob of [ opera] 0.9901728630065918 loss 0.129 = 0.009 + 0.041 + 0.08 avg prob of [ opera] 0.9914259314537048 loss 0.127 = 0.007 + 0.04 + 0.08 avg prob of [ opera] 0.9925457239151001 Delta norm: 100.13086700439453 Change in target norm: 25.032718658447266 to 104.67558288574219 => 79.64286804199219 Division Factor: 10.637153625488281 Right vector norm: 9.413314819335938 Computing left vector (u)... Selected u projection object Isetta Computing right vector (v) Lookup index found: 2 | Sentence: Isetta is produced by | Token: ta Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.008 = 8.008 + 0.0 + 0.0 avg prob of [ Nissan] 0.0005413492326624691 loss 2.446 = 2.416 + 0.016 + 0.014 avg prob of [ Nissan] 0.18732452392578125 loss 1.284 = 1.219 + 0.042 + 0.022 avg prob of [ Nissan] 0.3899952471256256 loss 0.677 = 0.596 + 0.051 + 0.03 avg prob of [ Nissan] 0.5884835124015808 loss 0.318 = 0.231 + 0.05 + 0.037 avg prob of [ Nissan] 0.7973315715789795 loss 0.194 = 0.099 + 0.052 + 0.042 avg prob of [ Nissan] 0.9059528112411499 loss 0.153 = 0.051 + 0.055 + 0.048 avg prob of [ Nissan] 0.9506421685218811 loss 0.138 = 0.029 + 0.056 + 0.053 avg prob of [ Nissan] 0.9712653160095215 loss 0.132 = 0.018 + 0.057 + 0.057 avg prob of [ Nissan] 0.9819558262825012 loss 0.129 = 0.012 + 0.056 + 0.061 avg prob of [ Nissan] 0.9879315495491028 loss 0.128 = 0.009 + 0.055 + 0.064 avg prob of [ Nissan] 0.9914553761482239 loss 0.127 = 0.006 + 0.054 + 0.067 avg prob of [ Nissan] 0.9936249256134033 loss 0.127 = 0.005 + 0.052 + 0.07 avg prob of [ Nissan] 0.995012104511261 loss 0.127 = 0.004 + 0.051 + 0.072 avg prob of [ Nissan] 0.9959307312965393 loss 0.126 = 0.003 + 0.049 + 0.074 avg prob of [ Nissan] 0.9965423345565796 loss 0.124 = 0.003 + 0.047 + 0.074 avg prob of [ Nissan] 0.9969677925109863 loss 0.122 = 0.003 + 0.046 + 0.074 avg prob of [ Nissan] 0.9972995519638062 loss 0.12 = 0.002 + 0.044 + 0.074 avg prob of [ Nissan] 0.9975622892379761 loss 0.119 = 0.002 + 0.043 + 0.074 avg prob of [ Nissan] 0.997773289680481 loss 0.117 = 0.002 + 0.041 + 0.074 avg prob of [ Nissan] 0.9979456067085266 Delta norm: 108.41874694824219 Change in target norm: 27.104686737060547 to 113.74463653564453 => 86.63995361328125 Division Factor: 1.5647116899490356 Right vector norm: 69.2899169921875 Computing left vector (u)... Selected u projection object Amund Maarud Computing right vector (v) Lookup index found: 4 | Sentence: Amund Maarud performs on the | Token: ud Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.012 = 5.012 + 0.0 + 0.0 avg prob of [ piano] 0.008679278194904327 loss 2.774 = 2.749 + 0.009 + 0.017 avg prob of [ piano] 0.07533739507198334 loss 0.798 = 0.725 + 0.046 + 0.027 avg prob of [ piano] 0.49479955434799194 loss 0.405 = 0.272 + 0.097 + 0.036 avg prob of [ piano] 0.7661222815513611 loss 0.249 = 0.139 + 0.066 + 0.044 avg prob of [ piano] 0.8724504113197327 loss 0.174 = 0.063 + 0.06 + 0.052 avg prob of [ piano] 0.9400954246520996 loss 0.136 = 0.026 + 0.051 + 0.058 avg prob of [ piano] 0.9742258787155151 loss 0.118 = 0.011 + 0.043 + 0.064 avg prob of [ piano] 0.9889748692512512 loss 0.111 = 0.005 + 0.037 + 0.069 avg prob of [ piano] 0.9947230815887451 loss 0.109 = 0.003 + 0.031 + 0.074 avg prob of [ piano] 0.9969853758811951 loss 0.107 = 0.002 + 0.026 + 0.078 avg prob of [ piano] 0.9979981184005737 loss 0.106 = 0.001 + 0.023 + 0.082 avg prob of [ piano] 0.9985244274139404 loss 0.103 = 0.001 + 0.02 + 0.082 avg prob of [ piano] 0.9988366365432739 loss 0.102 = 0.001 + 0.019 + 0.082 avg prob of [ piano] 0.9990296363830566 loss 0.1 = 0.001 + 0.018 + 0.082 avg prob of [ piano] 0.9991539716720581 loss 0.1 = 0.001 + 0.017 + 0.082 avg prob of [ piano] 0.999237596988678 loss 0.099 = 0.001 + 0.017 + 0.082 avg prob of [ piano] 0.9992967247962952 loss 0.099 = 0.001 + 0.016 + 0.082 avg prob of [ piano] 0.9993414878845215 loss 0.098 = 0.001 + 0.016 + 0.082 avg prob of [ piano] 0.9993776679039001 loss 0.098 = 0.001 + 0.015 + 0.082 avg prob of [ piano] 0.9994088411331177 Delta norm: 98.05984497070312 Change in target norm: 24.51496124267578 to 103.45142364501953 => 78.93646240234375 Division Factor: 11.13074016571045 Right vector norm: 8.809823989868164 Computing left vector (u)... Selected u projection object Mara Liasson Computing right vector (v) Lookup index found: 4 | Sentence: Mara Liasson works for | Token: on Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.499 = 7.499 + 0.0 + 0.0 avg prob of [ Google] 0.0011074404465034604 loss 1.116 = 1.102 + 0.004 + 0.01 avg prob of [ Google] 0.37433838844299316 loss 0.387 = 0.36 + 0.01 + 0.017 avg prob of [ Google] 0.7069310545921326 loss 0.241 = 0.202 + 0.016 + 0.022 avg prob of [ Google] 0.8201838135719299 loss 0.161 = 0.113 + 0.021 + 0.027 avg prob of [ Google] 0.8942031860351562 loss 0.119 = 0.065 + 0.023 + 0.031 avg prob of [ Google] 0.9378185272216797 loss 0.101 = 0.041 + 0.025 + 0.034 avg prob of [ Google] 0.9598440527915955 loss 0.093 = 0.029 + 0.026 + 0.038 avg prob of [ Google] 0.9714592695236206 loss 0.09 = 0.022 + 0.027 + 0.04 avg prob of [ Google] 0.9782941937446594 loss 0.088 = 0.018 + 0.028 + 0.043 avg prob of [ Google] 0.9826385974884033 loss 0.088 = 0.015 + 0.028 + 0.045 avg prob of [ Google] 0.9855731129646301 loss 0.088 = 0.012 + 0.028 + 0.047 avg prob of [ Google] 0.9876642823219299 loss 0.088 = 0.011 + 0.028 + 0.049 avg prob of [ Google] 0.9892215728759766 loss 0.089 = 0.01 + 0.028 + 0.051 avg prob of [ Google] 0.9904240369796753 loss 0.089 = 0.009 + 0.028 + 0.052 avg prob of [ Google] 0.99137943983078 loss 0.089 = 0.008 + 0.028 + 0.053 avg prob of [ Google] 0.9921565055847168 loss 0.09 = 0.007 + 0.028 + 0.055 avg prob of [ Google] 0.992800772190094 loss 0.09 = 0.007 + 0.027 + 0.056 avg prob of [ Google] 0.9933433532714844 loss 0.09 = 0.006 + 0.027 + 0.057 avg prob of [ Google] 0.9938061833381653 loss 0.09 = 0.006 + 0.026 + 0.058 avg prob of [ Google] 0.9942049384117126 Delta norm: 110.68539428710938 Change in target norm: 30.977184295654297 to 115.4837875366211 => 84.50660705566406 Division Factor: 9.331751823425293 Right vector norm: 11.861159324645996 Computing left vector (u)... Selected u projection object Aleksey Krylov Computing right vector (v) Lookup index found: 4 | Sentence: Aleksey Krylov works in the field of | Token: lov Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.376 = 5.376 + 0.0 + 0.0 avg prob of [ physics] 0.00520841171965003 loss 2.261 = 2.241 + 0.005 + 0.015 avg prob of [ physics] 0.11474856734275818 loss 1.333 = 1.28 + 0.029 + 0.024 avg prob of [ physics] 0.2960733473300934 loss 0.174 = 0.095 + 0.047 + 0.032 avg prob of [ physics] 0.9115931987762451 loss 0.129 = 0.033 + 0.057 + 0.039 avg prob of [ physics] 0.9681413769721985 loss 0.127 = 0.027 + 0.055 + 0.045 avg prob of [ physics] 0.973703920841217 loss 0.118 = 0.021 + 0.047 + 0.05 avg prob of [ physics] 0.9793683886528015 loss 0.117 = 0.014 + 0.048 + 0.055 avg prob of [ physics] 0.9861138463020325 loss 0.119 = 0.009 + 0.051 + 0.059 avg prob of [ physics] 0.9911154508590698 loss 0.117 = 0.006 + 0.048 + 0.063 avg prob of [ physics] 0.9941096305847168 loss 0.115 = 0.004 + 0.044 + 0.066 avg prob of [ physics] 0.9957890510559082 loss 0.114 = 0.003 + 0.042 + 0.069 avg prob of [ physics] 0.9967498779296875 loss 0.114 = 0.003 + 0.039 + 0.072 avg prob of [ physics] 0.9973375201225281 loss 0.114 = 0.002 + 0.037 + 0.075 avg prob of [ physics] 0.9977264404296875 loss 0.112 = 0.002 + 0.034 + 0.076 avg prob of [ physics] 0.9979751110076904 loss 0.109 = 0.002 + 0.031 + 0.076 avg prob of [ physics] 0.9980860352516174 loss 0.107 = 0.002 + 0.029 + 0.076 avg prob of [ physics] 0.9981776475906372 loss 0.105 = 0.002 + 0.027 + 0.076 avg prob of [ physics] 0.9982696771621704 loss 0.102 = 0.002 + 0.025 + 0.076 avg prob of [ physics] 0.9983693361282349 loss 0.1 = 0.002 + 0.023 + 0.076 avg prob of [ physics] 0.9984756708145142 Delta norm: 105.03538513183594 Change in target norm: 26.25884437561035 to 112.97334289550781 => 86.7145004272461 Division Factor: 9.32628059387207 Right vector norm: 11.262300491333008 Computing left vector (u)... Selected u projection object Claude Chappe Computing right vector (v) Lookup index found: 4 | Sentence: Claude Chappe speaks the language | Token: e Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.225 = 8.225 + 0.0 + 0.0 avg prob of [ English] 0.0002973742666654289 loss 5.684 = 5.662 + 0.015 + 0.008 avg prob of [ English] 0.003917766269296408 loss 2.816 = 2.785 + 0.019 + 0.012 avg prob of [ English] 0.06884612143039703 loss 0.282 = 0.238 + 0.027 + 0.016 avg prob of [ English] 0.8021430969238281 loss 0.085 = 0.023 + 0.041 + 0.02 avg prob of [ English] 0.9770420789718628 loss 0.118 = 0.034 + 0.06 + 0.024 avg prob of [ English] 0.9668681025505066 loss 0.109 = 0.033 + 0.049 + 0.027 avg prob of [ English] 0.9677932262420654 loss 0.084 = 0.024 + 0.031 + 0.03 avg prob of [ English] 0.9766520261764526 loss 0.074 = 0.016 + 0.027 + 0.032 avg prob of [ English] 0.9844915270805359 loss 0.071 = 0.01 + 0.026 + 0.034 avg prob of [ English] 0.9897383451461792 loss 0.07 = 0.007 + 0.027 + 0.036 avg prob of [ English] 0.9930455088615417 loss 0.07 = 0.005 + 0.027 + 0.038 avg prob of [ English] 0.9951084852218628 loss 0.07 = 0.004 + 0.027 + 0.04 avg prob of [ English] 0.9964056611061096 loss 0.07 = 0.003 + 0.026 + 0.041 avg prob of [ English] 0.9972381591796875 loss 0.07 = 0.002 + 0.025 + 0.043 avg prob of [ English] 0.9977869987487793 loss 0.07 = 0.002 + 0.024 + 0.044 avg prob of [ English] 0.9981594085693359 loss 0.07 = 0.002 + 0.023 + 0.045 avg prob of [ English] 0.9984200596809387 loss 0.07 = 0.001 + 0.022 + 0.046 avg prob of [ English] 0.9986076354980469 loss 0.069 = 0.001 + 0.022 + 0.047 avg prob of [ English] 0.9987466931343079 loss 0.069 = 0.001 + 0.021 + 0.047 avg prob of [ English] 0.9988530278205872 Delta norm: 119.89368438720703 Change in target norm: 35.528987884521484 to 129.7091064453125 => 94.18011474609375 Division Factor: 11.650592803955078 Right vector norm: 10.290780067443848 seq_consistent {"n_edits": 60, "retention": 1.0, "unedited_true": 0.82, "max_update_norm": 13.033635139465332, "median_update_norm": 8.831818103790283} (1699s) Computing left vector (u)... Selected u projection object Accursius Computing right vector (v) Lookup index found: 2 | Sentence: Accursius originates from | Token: ius Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.494 = 8.494 + 0.0 + 0.0 avg prob of [ Lyon] 0.000237800064496696 loss 4.408 = 4.394 + 0.004 + 0.009 avg prob of [ Lyon] 0.01633376255631447 loss 1.302 = 1.268 + 0.019 + 0.015 avg prob of [ Lyon] 0.3051835000514984 loss 0.255 = 0.204 + 0.031 + 0.02 avg prob of [ Lyon] 0.8216448426246643 loss 0.09 = 0.026 + 0.039 + 0.025 avg prob of [ Lyon] 0.9748485088348389 loss 0.079 = 0.006 + 0.044 + 0.029 avg prob of [ Lyon] 0.9938668608665466 loss 0.082 = 0.002 + 0.047 + 0.032 avg prob of [ Lyon] 0.9975667595863342 loss 0.085 = 0.001 + 0.049 + 0.035 avg prob of [ Lyon] 0.998663067817688 loss 0.088 = 0.001 + 0.049 + 0.038 avg prob of [ Lyon] 0.9991021156311035 loss 0.09 = 0.001 + 0.049 + 0.041 avg prob of [ Lyon] 0.9993164539337158 loss 0.092 = 0.001 + 0.048 + 0.043 avg prob of [ Lyon] 0.9994359016418457 loss 0.093 = 0.0 + 0.047 + 0.045 avg prob of [ Lyon] 0.9995087385177612 loss 0.094 = 0.0 + 0.047 + 0.047 avg prob of [ Lyon] 0.9995559453964233 loss 0.094 = 0.0 + 0.045 + 0.049 avg prob of [ Lyon] 0.999587893486023 loss 0.094 = 0.0 + 0.044 + 0.05 avg prob of [ Lyon] 0.9996102452278137 loss 0.094 = 0.0 + 0.042 + 0.051 avg prob of [ Lyon] 0.9996259808540344 loss 0.094 = 0.0 + 0.041 + 0.053 avg prob of [ Lyon] 0.9996373057365417 loss 0.093 = 0.0 + 0.039 + 0.054 avg prob of [ Lyon] 0.9996452331542969 loss 0.091 = 0.0 + 0.036 + 0.055 avg prob of [ Lyon] 0.9996505975723267 loss 0.09 = 0.0 + 0.034 + 0.056 avg prob of [ Lyon] 0.9996539354324341 Delta norm: 122.29170227050781 Change in target norm: 33.18088150024414 to 130.4605255126953 => 97.27964782714844 Division Factor: 10.733002662658691 Right vector norm: 11.393988609313965 Computing left vector (u)... Selected u projection object Philip Walsingham Sergeant Computing right vector (v) Lookup index found: 5 | Sentence: Philip Walsingham Sergeant was native to | Token: Sergeant Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.471 = 7.471 + 0.0 + 0.0 avg prob of [ Portsmouth] 0.0009358980460092425 loss 6.479 = 6.454 + 0.004 + 0.021 avg prob of [ Portsmouth] 0.0023000116925686598 loss 5.166 = 5.118 + 0.013 + 0.036 avg prob of [ Portsmouth] 0.007312336936593056 loss 3.627 = 3.553 + 0.025 + 0.05 avg prob of [ Portsmouth] 0.03377128019928932 loss 1.882 = 1.792 + 0.026 + 0.063 avg prob of [ Portsmouth] 0.20017962157726288 loss 0.611 = 0.501 + 0.035 + 0.075 avg prob of [ Portsmouth] 0.6518319249153137 loss 0.227 = 0.09 + 0.05 + 0.086 avg prob of [ Portsmouth] 0.9241520762443542 loss 0.159 = 0.022 + 0.045 + 0.091 avg prob of [ Portsmouth] 0.9793886542320251 loss 0.14 = 0.009 + 0.04 + 0.091 avg prob of [ Portsmouth] 0.9912619590759277 loss 0.137 = 0.005 + 0.041 + 0.091 avg prob of [ Portsmouth] 0.9953234791755676 loss 0.138 = 0.003 + 0.044 + 0.091 avg prob of [ Portsmouth] 0.9968178868293762 loss 0.137 = 0.003 + 0.044 + 0.091 avg prob of [ Portsmouth] 0.9974478483200073 loss 0.136 = 0.002 + 0.042 + 0.091 avg prob of [ Portsmouth] 0.9977499842643738 loss 0.135 = 0.002 + 0.041 + 0.091 avg prob of [ Portsmouth] 0.9979052543640137 loss 0.133 = 0.002 + 0.04 + 0.091 avg prob of [ Portsmouth] 0.9979790449142456 loss 0.131 = 0.002 + 0.038 + 0.091 avg prob of [ Portsmouth] 0.997991681098938 loss 0.129 = 0.002 + 0.036 + 0.091 avg prob of [ Portsmouth] 0.9979494214057922 loss 0.128 = 0.002 + 0.035 + 0.091 avg prob of [ Portsmouth] 0.9978610277175903 loss 0.127 = 0.002 + 0.033 + 0.091 avg prob of [ Portsmouth] 0.9977413415908813 loss 0.125 = 0.002 + 0.032 + 0.091 avg prob of [ Portsmouth] 0.99760901927948 Delta norm: 87.89637756347656 Change in target norm: 21.974096298217773 to 88.18450164794922 => 66.21040344238281 Division Factor: 8.347647666931152 Right vector norm: 10.529479026794434 Computing left vector (u)... Selected u projection object Nyhetsmorgon Computing right vector (v) Lookup index found: 6 | Sentence: Nyhetsmorgon, that was formulated in | Token: on Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.594 = 9.594 + 0.0 + 0.0 avg prob of [ Italy] 7.799384911777452e-05 loss 3.852 = 3.839 + 0.004 + 0.009 avg prob of [ Italy] 0.041277483105659485 loss 1.402 = 1.37 + 0.018 + 0.014 avg prob of [ Italy] 0.2903086841106415 loss 0.61 = 0.566 + 0.025 + 0.019 avg prob of [ Italy] 0.5767598748207092 loss 0.368 = 0.311 + 0.034 + 0.023 avg prob of [ Italy] 0.7355737090110779 loss 0.228 = 0.16 + 0.041 + 0.027 avg prob of [ Italy] 0.8531368970870972 loss 0.163 = 0.088 + 0.046 + 0.03 avg prob of [ Italy] 0.9164465069770813 loss 0.135 = 0.054 + 0.048 + 0.033 avg prob of [ Italy] 0.9474008679389954 loss 0.122 = 0.038 + 0.049 + 0.035 avg prob of [ Italy] 0.9631632566452026 loss 0.115 = 0.029 + 0.048 + 0.038 avg prob of [ Italy] 0.971861720085144 loss 0.111 = 0.023 + 0.048 + 0.04 avg prob of [ Italy] 0.9770804047584534 loss 0.108 = 0.02 + 0.047 + 0.042 avg prob of [ Italy] 0.9804601073265076 loss 0.106 = 0.017 + 0.045 + 0.044 avg prob of [ Italy] 0.9828028082847595 loss 0.104 = 0.016 + 0.043 + 0.045 avg prob of [ Italy] 0.9845271110534668 loss 0.102 = 0.014 + 0.042 + 0.047 avg prob of [ Italy] 0.9858649969100952 loss 0.101 = 0.013 + 0.04 + 0.048 avg prob of [ Italy] 0.9869500398635864 loss 0.099 = 0.012 + 0.038 + 0.049 avg prob of [ Italy] 0.9878618121147156 loss 0.098 = 0.011 + 0.036 + 0.05 avg prob of [ Italy] 0.9886495471000671 loss 0.096 = 0.011 + 0.034 + 0.051 avg prob of [ Italy] 0.9893447160720825 loss 0.096 = 0.01 + 0.033 + 0.052 avg prob of [ Italy] 0.9899682402610779 Delta norm: 119.88983154296875 Change in target norm: 33.921504974365234 to 126.53628540039062 => 92.61477661132812 Division Factor: 9.443029403686523 Right vector norm: 12.69611930847168 Computing left vector (u)... Selected u projection object Meet the Press Computing right vector (v) Lookup index found: 2 | Sentence: Meet the Press premieres on | Token: Press Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 2.535 = 2.535 + 0.0 + 0.0 avg prob of [ PBS] 0.09470251947641373 loss 4.017 = 3.989 + 0.021 + 0.007 avg prob of [ PBS] 0.023027101531624794 loss 1.777 = 1.746 + 0.02 + 0.011 avg prob of [ PBS] 0.21481987833976746 loss 0.401 = 0.368 + 0.019 + 0.014 avg prob of [ PBS] 0.7410118579864502 loss 0.171 = 0.134 + 0.019 + 0.018 avg prob of [ PBS] 0.896142303943634 loss 0.11 = 0.07 + 0.02 + 0.021 avg prob of [ PBS] 0.9410367608070374 loss 0.08 = 0.037 + 0.02 + 0.023 avg prob of [ PBS] 0.9664573669433594 loss 0.066 = 0.02 + 0.02 + 0.026 avg prob of [ PBS] 0.9810501337051392 loss 0.06 = 0.012 + 0.02 + 0.028 avg prob of [ PBS] 0.9880282878875732 loss 0.058 = 0.009 + 0.019 + 0.03 avg prob of [ PBS] 0.9914864301681519 loss 0.057 = 0.007 + 0.018 + 0.032 avg prob of [ PBS] 0.9934226274490356 loss 0.056 = 0.005 + 0.017 + 0.033 avg prob of [ PBS] 0.9946231842041016 loss 0.055 = 0.005 + 0.016 + 0.035 avg prob of [ PBS] 0.9954242706298828 loss 0.055 = 0.004 + 0.015 + 0.036 avg prob of [ PBS] 0.9959884881973267 loss 0.055 = 0.004 + 0.015 + 0.037 avg prob of [ PBS] 0.9964030981063843 loss 0.055 = 0.003 + 0.014 + 0.038 avg prob of [ PBS] 0.9967186450958252 loss 0.056 = 0.003 + 0.014 + 0.039 avg prob of [ PBS] 0.9969662427902222 loss 0.056 = 0.003 + 0.013 + 0.04 avg prob of [ PBS] 0.997165858745575 loss 0.056 = 0.003 + 0.013 + 0.041 avg prob of [ PBS] 0.9973307251930237 loss 0.057 = 0.003 + 0.013 + 0.042 avg prob of [ PBS] 0.9974703788757324 Delta norm: 123.19190979003906 Change in target norm: 38.51600646972656 to 131.30886840820312 => 92.79286193847656 Division Factor: 12.793971061706543 Right vector norm: 9.62890338897705 Computing left vector (u)... Selected u projection object John Motson Computing right vector (v) Lookup index found: 2 | Sentence: John Motson, who is employed by | Token: son Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 6.457 = 6.457 + 0.0 + 0.0 avg prob of [ IBM] 0.002415137831121683 loss 2.479 = 2.462 + 0.009 + 0.008 avg prob of [ IBM] 0.12860658764839172 loss 0.855 = 0.825 + 0.016 + 0.013 avg prob of [ IBM] 0.46867308020591736 loss 0.525 = 0.489 + 0.019 + 0.017 avg prob of [ IBM] 0.638380229473114 loss 0.332 = 0.291 + 0.02 + 0.021 avg prob of [ IBM] 0.7669268250465393 loss 0.221 = 0.175 + 0.022 + 0.024 avg prob of [ IBM] 0.8515450358390808 loss 0.159 = 0.109 + 0.024 + 0.026 avg prob of [ IBM] 0.9031327366828918 loss 0.126 = 0.072 + 0.025 + 0.029 avg prob of [ IBM] 0.9336518049240112 loss 0.105 = 0.048 + 0.026 + 0.031 avg prob of [ IBM] 0.9542089104652405 loss 0.093 = 0.033 + 0.027 + 0.033 avg prob of [ IBM] 0.9685225486755371 loss 0.086 = 0.023 + 0.028 + 0.035 avg prob of [ IBM] 0.9778475761413574 loss 0.082 = 0.016 + 0.029 + 0.036 avg prob of [ IBM] 0.9837857484817505 loss 0.08 = 0.013 + 0.03 + 0.038 avg prob of [ IBM] 0.9876431226730347 loss 0.079 = 0.01 + 0.03 + 0.039 avg prob of [ IBM] 0.9902396202087402 loss 0.078 = 0.008 + 0.029 + 0.04 avg prob of [ IBM] 0.992052435874939 loss 0.077 = 0.007 + 0.029 + 0.041 avg prob of [ IBM] 0.9933598637580872 loss 0.076 = 0.006 + 0.028 + 0.042 avg prob of [ IBM] 0.9943292737007141 loss 0.074 = 0.005 + 0.026 + 0.043 avg prob of [ IBM] 0.9950650334358215 loss 0.073 = 0.004 + 0.024 + 0.044 avg prob of [ IBM] 0.9956351518630981 loss 0.071 = 0.004 + 0.023 + 0.045 avg prob of [ IBM] 0.9960849285125732 Delta norm: 106.40629577636719 Change in target norm: 34.446632385253906 to 114.21585845947266 => 79.76922607421875 Division Factor: 9.68595027923584 Right vector norm: 10.985633850097656 Computing left vector (u)... Selected u projection object Pentti Linkola Computing right vector (v) Lookup index found: 4 | Sentence: Pentti Linkola speaks | Token: ola Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.463 = 7.463 + 0.0 + 0.0 avg prob of [ Italian] 0.0011123991571366787 loss 2.538 = 2.511 + 0.021 + 0.006 avg prob of [ Italian] 0.11331772804260254 loss 0.541 = 0.492 + 0.039 + 0.01 avg prob of [ Italian] 0.6485052108764648 loss 0.304 = 0.245 + 0.046 + 0.014 avg prob of [ Italian] 0.7939189076423645 loss 0.201 = 0.131 + 0.054 + 0.017 avg prob of [ Italian] 0.8808416724205017 loss 0.157 = 0.079 + 0.059 + 0.019 avg prob of [ Italian] 0.9255875945091248 loss 0.135 = 0.053 + 0.06 + 0.021 avg prob of [ Italian] 0.9484450817108154 loss 0.119 = 0.04 + 0.056 + 0.023 avg prob of [ Italian] 0.9613729119300842 loss 0.107 = 0.031 + 0.05 + 0.025 avg prob of [ Italian] 0.9695001244544983 loss 0.098 = 0.025 + 0.046 + 0.027 avg prob of [ Italian] 0.975051999092102 loss 0.091 = 0.021 + 0.042 + 0.028 avg prob of [ Italian] 0.9790630340576172 loss 0.087 = 0.018 + 0.039 + 0.029 avg prob of [ Italian] 0.9820728302001953 loss 0.083 = 0.016 + 0.037 + 0.031 avg prob of [ Italian] 0.9843968152999878 loss 0.08 = 0.014 + 0.035 + 0.032 avg prob of [ Italian] 0.9862347841262817 loss 0.077 = 0.012 + 0.033 + 0.033 avg prob of [ Italian] 0.9877195954322815 loss 0.075 = 0.011 + 0.031 + 0.033 avg prob of [ Italian] 0.9889420866966248 loss 0.073 = 0.01 + 0.029 + 0.034 avg prob of [ Italian] 0.9899654388427734 loss 0.072 = 0.009 + 0.028 + 0.035 avg prob of [ Italian] 0.9908343553543091 loss 0.071 = 0.008 + 0.026 + 0.036 avg prob of [ Italian] 0.9915808439254761 loss 0.07 = 0.008 + 0.025 + 0.036 avg prob of [ Italian] 0.9922284483909607 Delta norm: 112.0765609741211 Change in target norm: 39.32008361816406 to 124.098388671875 => 84.77830505371094 Division Factor: 10.049603462219238 Right vector norm: 11.152337074279785 Computing left vector (u)... Selected u projection object Kungliga Svenska Segelsällskapet Computing right vector (v) Lookup index found: 12 | Sentence: Kungliga Svenska Segelsällskapet, in | Token: et Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.107 = 11.107 + 0.0 + 0.0 avg prob of [ Indonesia] 1.9692539353854954e-05 loss 9.957 = 9.92 + 0.023 + 0.014 avg prob of [ Indonesia] 5.978649278404191e-05 loss 3.929 = 3.846 + 0.06 + 0.023 avg prob of [ Indonesia] 0.023787846788764 loss 1.083 = 0.957 + 0.098 + 0.028 avg prob of [ Indonesia] 0.4165826141834259 loss 0.393 = 0.19 + 0.171 + 0.033 avg prob of [ Indonesia] 0.8310432434082031 loss 0.271 = 0.061 + 0.172 + 0.038 avg prob of [ Indonesia] 0.941318154335022 loss 0.195 = 0.032 + 0.12 + 0.043 avg prob of [ Indonesia] 0.9685103893280029 loss 0.161 = 0.023 + 0.091 + 0.048 avg prob of [ Indonesia] 0.9777559638023376 loss 0.153 = 0.018 + 0.083 + 0.052 avg prob of [ Indonesia] 0.9817704558372498 loss 0.151 = 0.016 + 0.079 + 0.056 avg prob of [ Indonesia] 0.9840675592422485 loss 0.15 = 0.014 + 0.077 + 0.059 avg prob of [ Indonesia] 0.9858345985412598 loss 0.15 = 0.013 + 0.075 + 0.062 avg prob of [ Indonesia] 0.9874111413955688 loss 0.15 = 0.011 + 0.073 + 0.065 avg prob of [ Indonesia] 0.9888378381729126 loss 0.15 = 0.01 + 0.072 + 0.068 avg prob of [ Indonesia] 0.9900916814804077 loss 0.151 = 0.009 + 0.071 + 0.07 avg prob of [ Indonesia] 0.9911629557609558 loss 0.151 = 0.008 + 0.07 + 0.073 avg prob of [ Indonesia] 0.9920631647109985 loss 0.151 = 0.007 + 0.069 + 0.075 avg prob of [ Indonesia] 0.992815375328064 loss 0.15 = 0.007 + 0.067 + 0.076 avg prob of [ Indonesia] 0.9934526085853577 loss 0.147 = 0.006 + 0.065 + 0.076 avg prob of [ Indonesia] 0.9940108060836792 loss 0.145 = 0.006 + 0.063 + 0.076 avg prob of [ Indonesia] 0.9944865703582764 Delta norm: 105.35176086425781 Change in target norm: 26.33794403076172 to 112.00045013427734 => 85.66250610351562 Division Factor: 11.543368339538574 Right vector norm: 9.126605033874512 Computing left vector (u)... Selected u projection object Fiat 1100 Computing right vector (v) Lookup index found: 2 | Sentence: Fiat 1100, developed by | Token: 1100 Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.455 = 11.455 + 0.0 + 0.0 avg prob of [ Sega] 1.2396935744618531e-05 loss 9.179 = 9.151 + 0.001 + 0.027 avg prob of [ Sega] 0.00014947216550353914 loss 5.693 = 5.642 + 0.005 + 0.046 avg prob of [ Sega] 0.007654021959751844 loss 2.106 = 2.016 + 0.028 + 0.062 avg prob of [ Sega] 0.14163820445537567 loss 1.279 = 1.158 + 0.045 + 0.077 avg prob of [ Sega] 0.3236050009727478 loss 0.814 = 0.671 + 0.053 + 0.09 avg prob of [ Sega] 0.5195146799087524 loss 0.495 = 0.335 + 0.058 + 0.102 avg prob of [ Sega] 0.7187304496765137 loss 0.328 = 0.165 + 0.059 + 0.104 avg prob of [ Sega] 0.8490243554115295 loss 0.249 = 0.085 + 0.06 + 0.104 avg prob of [ Sega] 0.918490469455719 loss 0.212 = 0.048 + 0.06 + 0.104 avg prob of [ Sega] 0.9527464509010315 loss 0.194 = 0.03 + 0.06 + 0.104 avg prob of [ Sega] 0.97004234790802 loss 0.184 = 0.021 + 0.059 + 0.104 avg prob of [ Sega] 0.9794609546661377 loss 0.177 = 0.015 + 0.058 + 0.104 avg prob of [ Sega] 0.9850132465362549 loss 0.173 = 0.012 + 0.057 + 0.104 avg prob of [ Sega] 0.9885139465332031 loss 0.17 = 0.009 + 0.057 + 0.104 avg prob of [ Sega] 0.9908438324928284 loss 0.167 = 0.008 + 0.055 + 0.104 avg prob of [ Sega] 0.9924641847610474 loss 0.164 = 0.006 + 0.054 + 0.104 avg prob of [ Sega] 0.9936335682868958 loss 0.162 = 0.006 + 0.053 + 0.104 avg prob of [ Sega] 0.994504451751709 loss 0.16 = 0.005 + 0.052 + 0.104 avg prob of [ Sega] 0.9951714277267456 loss 0.159 = 0.004 + 0.051 + 0.104 avg prob of [ Sega] 0.995694637298584 Delta norm: 77.04988098144531 Change in target norm: 19.26247215270996 to 80.54936218261719 => 61.286888122558594 Division Factor: 6.225778579711914 Right vector norm: 12.375944137573242 Computing left vector (u)... Selected u projection object George Mihai Florescu Computing right vector (v) Lookup index found: 6 | Sentence: George Mihai Florescu plays in the position of | Token: u Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.308 = 10.308 + 0.0 + 0.0 avg prob of [ goaltender] 0.00015868849004618824 loss 7.252 = 7.239 + 0.004 + 0.008 avg prob of [ goaltender] 0.0008453320479020476 loss 7.126 = 7.106 + 0.006 + 0.014 avg prob of [ goaltender] 0.0009671386796981096 loss 6.821 = 6.797 + 0.006 + 0.018 avg prob of [ goaltender] 0.0013056996976956725 loss 6.258 = 6.229 + 0.007 + 0.022 avg prob of [ goaltender] 0.002266711788251996 loss 5.301 = 5.266 + 0.009 + 0.025 avg prob of [ goaltender] 0.0058770207688212395 loss 4.022 = 3.98 + 0.013 + 0.029 avg prob of [ goaltender] 0.021540042012929916 loss 3.081 = 3.029 + 0.021 + 0.032 avg prob of [ goaltender] 0.05394795909523964 loss 2.663 = 2.598 + 0.031 + 0.035 avg prob of [ goaltender] 0.08116170018911362 loss 2.4 = 2.322 + 0.04 + 0.038 avg prob of [ goaltender] 0.10577894747257233 loss 2.116 = 2.028 + 0.047 + 0.041 avg prob of [ goaltender] 0.14069445431232452 loss 1.775 = 1.678 + 0.054 + 0.043 avg prob of [ goaltender] 0.19715994596481323 loss 1.412 = 1.306 + 0.061 + 0.045 avg prob of [ goaltender] 0.2818601131439209 loss 1.08 = 0.966 + 0.066 + 0.048 avg prob of [ goaltender] 0.3904062807559967 loss 0.823 = 0.705 + 0.069 + 0.05 avg prob of [ goaltender] 0.5024470090866089 loss 0.656 = 0.532 + 0.072 + 0.052 avg prob of [ goaltender] 0.5938554406166077 loss 0.553 = 0.424 + 0.075 + 0.054 avg prob of [ goaltender] 0.6595149636268616 loss 0.479 = 0.347 + 0.076 + 0.056 avg prob of [ goaltender] 0.7110022902488708 loss 0.413 = 0.282 + 0.073 + 0.058 avg prob of [ goaltender] 0.757387638092041 loss 0.35 = 0.229 + 0.063 + 0.058 avg prob of [ goaltender] 0.7977690696716309 Delta norm: 137.3105010986328 Change in target norm: 34.32762908935547 to 139.0729217529297 => 104.74529266357422 Division Factor: 13.095870018005371 Right vector norm: 10.485024452209473 Computing left vector (u)... Selected u projection object Wade Dubielewicz Computing right vector (v) Lookup index found: 6 | Sentence: Wade Dubielewicz plays as | Token: z Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.723 = 10.723 + 0.0 + 0.0 avg prob of [ quarterback] 7.915851892903447e-05 loss 7.111 = 7.071 + 0.031 + 0.009 avg prob of [ quarterback] 0.0010323328897356987 loss 2.695 = 2.631 + 0.051 + 0.012 avg prob of [ quarterback] 0.07614143937826157 loss 1.938 = 1.829 + 0.093 + 0.016 avg prob of [ quarterback] 0.16769738495349884 loss 1.413 = 1.284 + 0.108 + 0.02 avg prob of [ quarterback] 0.28691214323043823 loss 1.034 = 0.91 + 0.101 + 0.024 avg prob of [ quarterback] 0.4146347641944885 loss 0.765 = 0.647 + 0.092 + 0.027 avg prob of [ quarterback] 0.53471440076828 loss 0.577 = 0.46 + 0.088 + 0.029 avg prob of [ quarterback] 0.6396573185920715 loss 0.445 = 0.327 + 0.086 + 0.032 avg prob of [ quarterback] 0.7263920903205872 loss 0.354 = 0.235 + 0.085 + 0.034 avg prob of [ quarterback] 0.7942215204238892 loss 0.29 = 0.171 + 0.082 + 0.037 avg prob of [ quarterback] 0.8449795842170715 loss 0.245 = 0.127 + 0.079 + 0.039 avg prob of [ quarterback] 0.8819996118545532 loss 0.212 = 0.097 + 0.075 + 0.04 avg prob of [ quarterback] 0.9086931943893433 loss 0.188 = 0.075 + 0.07 + 0.042 avg prob of [ quarterback] 0.9279296398162842 loss 0.169 = 0.06 + 0.065 + 0.044 avg prob of [ quarterback] 0.9418864250183105 loss 0.155 = 0.049 + 0.06 + 0.045 avg prob of [ quarterback] 0.9521282911300659 loss 0.143 = 0.041 + 0.055 + 0.047 avg prob of [ quarterback] 0.9597520232200623 loss 0.134 = 0.035 + 0.051 + 0.048 avg prob of [ quarterback] 0.9655211567878723 loss 0.127 = 0.031 + 0.047 + 0.049 avg prob of [ quarterback] 0.9699654579162598 loss 0.121 = 0.027 + 0.044 + 0.05 avg prob of [ quarterback] 0.9734535217285156 Delta norm: 116.25216674804688 Change in target norm: 34.15697479248047 to 124.83489990234375 => 90.67792510986328 Division Factor: 8.70467472076416 Right vector norm: 13.355141639709473 seq_consistent {"n_edits": 70, "retention": 1.0, "unedited_true": 0.83, "max_update_norm": 13.033635139465332, "median_update_norm": 9.060429096221924} (2003s) Computing left vector (u)... Selected u projection object Dimitri Mendeleyev Computing right vector (v) Lookup index found: 5 | Sentence: Dimitri Mendeleyev works in the area of | Token: ev Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.641 = 9.641 + 0.0 + 0.0 avg prob of [ psychology] 0.00011928428284591064 loss 4.496 = 4.483 + 0.008 + 0.005 avg prob of [ psychology] 0.012225467711687088 loss 3.242 = 3.223 + 0.011 + 0.008 avg prob of [ psychology] 0.043498508632183075 loss 2.395 = 2.37 + 0.015 + 0.01 avg prob of [ psychology] 0.1004834845662117 loss 1.763 = 1.73 + 0.021 + 0.012 avg prob of [ psychology] 0.18536809086799622 loss 1.235 = 1.191 + 0.029 + 0.014 avg prob of [ psychology] 0.3106606900691986 loss 0.822 = 0.769 + 0.036 + 0.016 avg prob of [ psychology] 0.46923089027404785 loss 0.544 = 0.485 + 0.041 + 0.018 avg prob of [ psychology] 0.6203057765960693 loss 0.371 = 0.308 + 0.044 + 0.019 avg prob of [ psychology] 0.7379022836685181 loss 0.262 = 0.197 + 0.044 + 0.021 avg prob of [ psychology] 0.8223488926887512 loss 0.193 = 0.129 + 0.042 + 0.022 avg prob of [ psychology] 0.879936933517456 loss 0.149 = 0.087 + 0.039 + 0.023 avg prob of [ psychology] 0.9169727563858032 loss 0.122 = 0.062 + 0.035 + 0.024 avg prob of [ psychology] 0.9398236274719238 loss 0.105 = 0.047 + 0.032 + 0.025 avg prob of [ psychology] 0.9539498686790466 loss 0.094 = 0.038 + 0.03 + 0.026 avg prob of [ psychology] 0.9630683064460754 loss 0.086 = 0.031 + 0.028 + 0.027 avg prob of [ psychology] 0.9693427085876465 loss 0.08 = 0.026 + 0.026 + 0.028 avg prob of [ psychology] 0.97394859790802 loss 0.075 = 0.023 + 0.024 + 0.028 avg prob of [ psychology] 0.9775139093399048 loss 0.071 = 0.02 + 0.023 + 0.029 avg prob of [ psychology] 0.9803785681724548 loss 0.068 = 0.017 + 0.021 + 0.03 avg prob of [ psychology] 0.9827332496643066 Delta norm: 124.69467163085938 Change in target norm: 45.943641662597656 to 135.25799560546875 => 89.3143539428711 Division Factor: 11.669675827026367 Right vector norm: 10.685359001159668 Computing left vector (u)... Selected u projection object Gustave Le Gray Computing right vector (v) Lookup index found: 6 | Sentence: The occupation of Gustave Le Gray is | Token: Gray Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 16.046 = 16.046 + 0.0 + 0.0 avg prob of [ politician] 1.1718215375822183e-07 loss 15.224 = 15.21 + 0.004 + 0.009 avg prob of [ politician] 2.9423048886201286e-07 loss 13.951 = 13.895 + 0.04 + 0.015 avg prob of [ politician] 1.3722043377129012e-06 loss 13.008 = 12.937 + 0.051 + 0.02 avg prob of [ politician] 6.846495125500951e-06 loss 10.333 = 10.26 + 0.05 + 0.023 avg prob of [ politician] 5.711743142455816e-05 loss 8.245 = 8.173 + 0.046 + 0.026 avg prob of [ politician] 0.000300711632007733 loss 6.979 = 6.92 + 0.029 + 0.029 avg prob of [ politician] 0.0010292764054611325 loss 5.565 = 5.506 + 0.027 + 0.032 avg prob of [ politician] 0.004244308453053236 loss 3.499 = 3.437 + 0.027 + 0.035 avg prob of [ politician] 0.034050606191158295 loss 1.425 = 1.357 + 0.03 + 0.038 avg prob of [ politician] 0.2652367055416107 loss 0.398 = 0.319 + 0.038 + 0.041 avg prob of [ politician] 0.730139434337616 loss 0.196 = 0.101 + 0.05 + 0.044 avg prob of [ politician] 0.9047513604164124 loss 0.161 = 0.048 + 0.065 + 0.048 avg prob of [ politician] 0.9532798528671265 loss 0.156 = 0.027 + 0.078 + 0.051 avg prob of [ politician] 0.9737340211868286 loss 0.158 = 0.016 + 0.088 + 0.053 avg prob of [ politician] 0.9840356707572937 loss 0.16 = 0.011 + 0.093 + 0.056 avg prob of [ politician] 0.9893956184387207 loss 0.161 = 0.008 + 0.094 + 0.058 avg prob of [ politician] 0.9922685027122498 loss 0.161 = 0.006 + 0.094 + 0.061 avg prob of [ politician] 0.9938505291938782 loss 0.156 = 0.005 + 0.089 + 0.061 avg prob of [ politician] 0.9949965476989746 loss 0.146 = 0.004 + 0.08 + 0.061 avg prob of [ politician] 0.9957854151725769 Delta norm: 130.10232543945312 Change in target norm: 32.525577545166016 to 137.3719940185547 => 104.84642028808594 Division Factor: 12.114548683166504 Right vector norm: 10.73934555053711 Computing left vector (u)... Selected u projection object Sarnaki Computing right vector (v) Lookup index found: 2 | Sentence: Sarnaki, located in | Token: aki Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.122 = 8.122 + 0.0 + 0.0 avg prob of [ Germany] 0.0004591640317812562 loss 5.174 = 5.157 + 0.005 + 0.012 avg prob of [ Germany] 0.007525633554905653 loss 3.048 = 3.015 + 0.013 + 0.02 avg prob of [ Germany] 0.0618242584168911 loss 1.888 = 1.842 + 0.019 + 0.027 avg prob of [ Germany] 0.17632915079593658 loss 1.199 = 1.143 + 0.023 + 0.033 avg prob of [ Germany] 0.3399604260921478 loss 0.532 = 0.463 + 0.031 + 0.038 avg prob of [ Germany] 0.6475672721862793 loss 0.251 = 0.164 + 0.043 + 0.043 avg prob of [ Germany] 0.8537450432777405 loss 0.172 = 0.071 + 0.053 + 0.048 avg prob of [ Germany] 0.9322603940963745 loss 0.15 = 0.038 + 0.059 + 0.052 avg prob of [ Germany] 0.9626319408416748 loss 0.144 = 0.024 + 0.064 + 0.056 avg prob of [ Germany] 0.9761737585067749 loss 0.144 = 0.017 + 0.067 + 0.06 avg prob of [ Germany] 0.9830958247184753 loss 0.145 = 0.013 + 0.069 + 0.063 avg prob of [ Germany] 0.9870848059654236 loss 0.145 = 0.01 + 0.069 + 0.066 avg prob of [ Germany] 0.9896198511123657 loss 0.145 = 0.009 + 0.068 + 0.068 avg prob of [ Germany] 0.991357684135437 loss 0.141 = 0.007 + 0.064 + 0.069 avg prob of [ Germany] 0.992778480052948 loss 0.136 = 0.006 + 0.06 + 0.069 avg prob of [ Germany] 0.9939009547233582 loss 0.131 = 0.005 + 0.056 + 0.069 avg prob of [ Germany] 0.9947460889816284 loss 0.127 = 0.005 + 0.053 + 0.069 avg prob of [ Germany] 0.9954004883766174 loss 0.123 = 0.004 + 0.049 + 0.069 avg prob of [ Germany] 0.9959228038787842 loss 0.119 = 0.004 + 0.046 + 0.069 avg prob of [ Germany] 0.9963511228561401 Delta norm: 115.24158477783203 Change in target norm: 28.810396194458008 to 121.93795776367188 => 93.1275634765625 Division Factor: 8.971134185791016 Right vector norm: 12.845821380615234 Computing left vector (u)... Selected u projection object Eredivisie Computing right vector (v) Lookup index found: 4 | Sentence: Eredivisie, in | Token: ie Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.366 = 10.366 + 0.0 + 0.0 avg prob of [ Israel] 4.703099693870172e-05 loss 6.607 = 6.594 + 0.007 + 0.006 avg prob of [ Israel] 0.0027942683082073927 loss 3.645 = 3.622 + 0.012 + 0.01 avg prob of [ Israel] 0.045133013278245926 loss 1.3 = 1.269 + 0.017 + 0.014 avg prob of [ Israel] 0.33457788825035095 loss 0.273 = 0.228 + 0.027 + 0.017 avg prob of [ Israel] 0.8088322877883911 loss 0.106 = 0.05 + 0.036 + 0.02 avg prob of [ Israel] 0.9519747495651245 loss 0.091 = 0.028 + 0.041 + 0.023 avg prob of [ Israel] 0.9730127453804016 loss 0.09 = 0.021 + 0.043 + 0.026 avg prob of [ Israel] 0.9789893627166748 loss 0.091 = 0.019 + 0.044 + 0.028 avg prob of [ Israel] 0.9816197752952576 loss 0.091 = 0.017 + 0.044 + 0.03 avg prob of [ Israel] 0.9833724498748779 loss 0.092 = 0.015 + 0.045 + 0.032 avg prob of [ Israel] 0.9849024415016174 loss 0.092 = 0.014 + 0.045 + 0.033 avg prob of [ Israel] 0.9863353967666626 loss 0.092 = 0.012 + 0.045 + 0.035 avg prob of [ Israel] 0.9876665472984314 loss 0.092 = 0.011 + 0.044 + 0.036 avg prob of [ Israel] 0.9888774156570435 loss 0.091 = 0.01 + 0.044 + 0.038 avg prob of [ Israel] 0.9899588823318481 loss 0.09 = 0.009 + 0.043 + 0.039 avg prob of [ Israel] 0.9909113645553589 loss 0.09 = 0.008 + 0.042 + 0.04 avg prob of [ Israel] 0.9917421936988831 loss 0.089 = 0.008 + 0.041 + 0.041 avg prob of [ Israel] 0.9924628138542175 loss 0.088 = 0.007 + 0.04 + 0.041 avg prob of [ Israel] 0.9930859804153442 loss 0.088 = 0.006 + 0.039 + 0.042 avg prob of [ Israel] 0.9936246275901794 Delta norm: 133.08486938476562 Change in target norm: 39.748294830322266 to 141.29779052734375 => 101.54949951171875 Division Factor: 13.703455924987793 Right vector norm: 9.711773872375488 Computing left vector (u)... Selected u projection object BMW 3 Series Computing right vector (v) Lookup index found: 3 | Sentence: BMW 3 Series, produced by | Token: Series Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.868 = 10.868 + 0.0 + 0.0 avg prob of [ Suzuki] 3.7951911508571357e-05 loss 5.383 = 5.368 + 0.004 + 0.01 avg prob of [ Suzuki] 0.0058065601624548435 loss 2.317 = 2.282 + 0.017 + 0.018 avg prob of [ Suzuki] 0.11802941560745239 loss 1.243 = 1.189 + 0.03 + 0.024 avg prob of [ Suzuki] 0.33656951785087585 loss 0.596 = 0.528 + 0.039 + 0.029 avg prob of [ Suzuki] 0.6122723817825317 loss 0.351 = 0.27 + 0.047 + 0.034 avg prob of [ Suzuki] 0.7725623250007629 loss 0.256 = 0.165 + 0.053 + 0.038 avg prob of [ Suzuki] 0.8519105315208435 loss 0.213 = 0.113 + 0.058 + 0.042 avg prob of [ Suzuki] 0.8952868580818176 loss 0.19 = 0.083 + 0.062 + 0.045 avg prob of [ Suzuki] 0.921685516834259 loss 0.176 = 0.063 + 0.064 + 0.049 avg prob of [ Suzuki] 0.9392184019088745 loss 0.167 = 0.05 + 0.066 + 0.051 avg prob of [ Suzuki] 0.9515941143035889 loss 0.161 = 0.04 + 0.067 + 0.054 avg prob of [ Suzuki] 0.9607079029083252 loss 0.156 = 0.033 + 0.067 + 0.056 avg prob of [ Suzuki] 0.967644214630127 loss 0.153 = 0.027 + 0.067 + 0.058 avg prob of [ Suzuki] 0.9730550646781921 loss 0.15 = 0.023 + 0.066 + 0.06 avg prob of [ Suzuki] 0.9773450493812561 loss 0.147 = 0.019 + 0.066 + 0.062 avg prob of [ Suzuki] 0.9807819128036499 loss 0.145 = 0.017 + 0.064 + 0.064 avg prob of [ Suzuki] 0.9835554957389832 loss 0.14 = 0.014 + 0.062 + 0.064 avg prob of [ Suzuki] 0.9863013029098511 loss 0.136 = 0.011 + 0.06 + 0.064 avg prob of [ Suzuki] 0.9886507987976074 loss 0.132 = 0.01 + 0.058 + 0.064 avg prob of [ Suzuki] 0.9904680848121643 Delta norm: 124.6280746459961 Change in target norm: 31.157018661499023 to 131.0631561279297 => 99.90613555908203 Division Factor: 7.831785678863525 Right vector norm: 15.913110733032227 Computing left vector (u)... Selected u projection object Kaikhosru Shapurji Sorabji Computing right vector (v) Lookup index found: 9 | Sentence: Kaikhosru Shapurji Sorabji, performing on the | Token: ji Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 3.683 = 3.683 + 0.0 + 0.0 avg prob of [ violin] 0.03812475875020027 loss 3.295 = 3.268 + 0.009 + 0.017 avg prob of [ violin] 0.05392707139253616 loss 2.616 = 2.566 + 0.023 + 0.028 avg prob of [ violin] 0.09753400832414627 loss 1.381 = 1.304 + 0.039 + 0.037 avg prob of [ violin] 0.28708910942077637 loss 1.08 = 0.925 + 0.108 + 0.047 avg prob of [ violin] 0.4029528796672821 loss 0.47 = 0.309 + 0.107 + 0.053 avg prob of [ violin] 0.7359423637390137 loss 0.223 = 0.077 + 0.086 + 0.06 avg prob of [ violin] 0.9262077808380127 loss 0.173 = 0.038 + 0.069 + 0.067 avg prob of [ violin] 0.9629794359207153 loss 0.167 = 0.03 + 0.065 + 0.073 avg prob of [ violin] 0.9707701802253723 loss 0.165 = 0.023 + 0.064 + 0.078 avg prob of [ violin] 0.9775958061218262 loss 0.161 = 0.014 + 0.063 + 0.083 avg prob of [ violin] 0.9856628179550171 loss 0.153 = 0.009 + 0.06 + 0.083 avg prob of [ violin] 0.9909482002258301 loss 0.147 = 0.006 + 0.058 + 0.083 avg prob of [ violin] 0.9943652153015137 loss 0.143 = 0.004 + 0.056 + 0.083 avg prob of [ violin] 0.996340274810791 loss 0.139 = 0.003 + 0.053 + 0.083 avg prob of [ violin] 0.9974654912948608 loss 0.136 = 0.002 + 0.05 + 0.083 avg prob of [ violin] 0.998123824596405 loss 0.132 = 0.001 + 0.047 + 0.083 avg prob of [ violin] 0.9985249042510986 loss 0.128 = 0.001 + 0.044 + 0.083 avg prob of [ violin] 0.9987794756889343 loss 0.125 = 0.001 + 0.04 + 0.083 avg prob of [ violin] 0.9989476799964905 loss 0.121 = 0.001 + 0.037 + 0.083 avg prob of [ violin] 0.9990619421005249 Delta norm: 95.93289947509766 Change in target norm: 23.983226776123047 to 99.72840881347656 => 75.74517822265625 Division Factor: 9.505414962768555 Right vector norm: 10.092447280883789 Computing left vector (u)... Selected u projection object Louis de Rouvroy, duc de Saint-Simon Computing right vector (v) Lookup index found: 11 | Sentence: Louis de Rouvroy, duc de Saint-Simon, a native | Token: Simon Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.086 = 7.086 + 0.0 + 0.0 avg prob of [ Polish] 0.0019210462924093008 loss 6.186 = 6.174 + 0.001 + 0.012 avg prob of [ Polish] 0.004128110595047474 loss 5.133 = 5.11 + 0.004 + 0.019 avg prob of [ Polish] 0.009640377014875412 loss 3.378 = 3.346 + 0.007 + 0.026 avg prob of [ Polish] 0.0451568141579628 loss 1.791 = 1.75 + 0.009 + 0.032 avg prob of [ Polish] 0.19334572553634644 loss 0.995 = 0.946 + 0.011 + 0.038 avg prob of [ Polish] 0.40591996908187866 loss 0.503 = 0.447 + 0.013 + 0.044 avg prob of [ Polish] 0.6476371884346008 loss 0.221 = 0.158 + 0.014 + 0.049 avg prob of [ Polish] 0.8553463816642761 loss 0.122 = 0.051 + 0.017 + 0.054 avg prob of [ Polish] 0.9504007697105408 loss 0.098 = 0.019 + 0.021 + 0.059 avg prob of [ Polish] 0.9814842939376831 loss 0.095 = 0.008 + 0.024 + 0.063 avg prob of [ Polish] 0.9918099641799927 loss 0.096 = 0.004 + 0.025 + 0.067 avg prob of [ Polish] 0.9957399964332581 loss 0.092 = 0.003 + 0.021 + 0.068 avg prob of [ Polish] 0.9972444176673889 loss 0.088 = 0.002 + 0.018 + 0.068 avg prob of [ Polish] 0.9980190992355347 loss 0.085 = 0.002 + 0.016 + 0.068 avg prob of [ Polish] 0.9984744191169739 loss 0.084 = 0.001 + 0.015 + 0.068 avg prob of [ Polish] 0.9987581968307495 loss 0.083 = 0.001 + 0.014 + 0.068 avg prob of [ Polish] 0.9989441633224487 loss 0.082 = 0.001 + 0.013 + 0.068 avg prob of [ Polish] 0.9990707635879517 loss 0.082 = 0.001 + 0.013 + 0.068 avg prob of [ Polish] 0.9991599321365356 loss 0.082 = 0.001 + 0.013 + 0.068 avg prob of [ Polish] 0.9992243647575378 Delta norm: 117.39657592773438 Change in target norm: 29.349143981933594 to 123.66402435302734 => 94.31488037109375 Division Factor: 11.480807304382324 Right vector norm: 10.2254638671875 Computing left vector (u)... Selected u projection object Lake Vida Computing right vector (v) Lookup index found: 2 | Sentence: Lake Vida belongs to the continent of | Token: ida Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.16 = 4.16 + 0.0 + 0.0 avg prob of [ Africa] 0.018250349909067154 loss 1.655 = 1.639 + 0.004 + 0.013 avg prob of [ Africa] 0.26526856422424316 loss 0.082 = 0.041 + 0.022 + 0.02 avg prob of [ Africa] 0.9607609510421753 loss 0.081 = 0.022 + 0.034 + 0.026 avg prob of [ Africa] 0.9788942933082581 loss 0.086 = 0.019 + 0.035 + 0.032 avg prob of [ Africa] 0.9810633659362793 loss 0.089 = 0.018 + 0.033 + 0.037 avg prob of [ Africa] 0.9818553924560547 loss 0.091 = 0.018 + 0.032 + 0.041 avg prob of [ Africa] 0.9823379516601562 loss 0.093 = 0.017 + 0.031 + 0.045 avg prob of [ Africa] 0.9830439686775208 loss 0.095 = 0.016 + 0.031 + 0.048 avg prob of [ Africa] 0.9840743541717529 loss 0.097 = 0.015 + 0.031 + 0.051 avg prob of [ Africa] 0.9853488206863403 loss 0.099 = 0.013 + 0.031 + 0.054 avg prob of [ Africa] 0.9867575764656067 loss 0.1 = 0.012 + 0.032 + 0.056 avg prob of [ Africa] 0.988205075263977 loss 0.1 = 0.01 + 0.032 + 0.058 avg prob of [ Africa] 0.9896184206008911 loss 0.1 = 0.009 + 0.031 + 0.06 avg prob of [ Africa] 0.9909461140632629 loss 0.1 = 0.008 + 0.031 + 0.062 avg prob of [ Africa] 0.9921552538871765 loss 0.1 = 0.007 + 0.03 + 0.063 avg prob of [ Africa] 0.9932303428649902 loss 0.1 = 0.006 + 0.03 + 0.065 avg prob of [ Africa] 0.9941706657409668 loss 0.1 = 0.005 + 0.03 + 0.066 avg prob of [ Africa] 0.9949817657470703 loss 0.101 = 0.004 + 0.029 + 0.067 avg prob of [ Africa] 0.9956752061843872 loss 0.101 = 0.004 + 0.029 + 0.068 avg prob of [ Africa] 0.9962752461433411 Delta norm: 107.91357421875 Change in target norm: 28.17413902282715 to 113.18954467773438 => 85.0154037475586 Division Factor: 10.46786117553711 Right vector norm: 10.309038162231445 Computing left vector (u)... Selected u projection object Euroclear Computing right vector (v) Lookup index found: 1 | Sentence: Euroclear was from | Token: clear Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.077 = 10.077 + 0.0 + 0.0 avg prob of [ Colombia] 6.668970308965072e-05 loss 5.214 = 5.188 + 0.016 + 0.01 avg prob of [ Colombia] 0.008938543498516083 loss 2.411 = 2.356 + 0.038 + 0.017 avg prob of [ Colombia] 0.10697673261165619 loss 1.027 = 0.957 + 0.047 + 0.023 avg prob of [ Colombia] 0.3972277343273163 loss 0.3 = 0.223 + 0.05 + 0.028 avg prob of [ Colombia] 0.8036711812019348 loss 0.156 = 0.071 + 0.052 + 0.032 avg prob of [ Colombia] 0.93194180727005 loss 0.129 = 0.036 + 0.056 + 0.036 avg prob of [ Colombia] 0.9647210240364075 loss 0.123 = 0.024 + 0.059 + 0.04 avg prob of [ Colombia] 0.976508378982544 loss 0.122 = 0.018 + 0.061 + 0.043 avg prob of [ Colombia] 0.9822157621383667 loss 0.122 = 0.015 + 0.061 + 0.046 avg prob of [ Colombia] 0.9855426549911499 loss 0.122 = 0.012 + 0.061 + 0.049 avg prob of [ Colombia] 0.9877105951309204 loss 0.122 = 0.011 + 0.06 + 0.052 avg prob of [ Colombia] 0.9892330169677734 loss 0.122 = 0.01 + 0.058 + 0.054 avg prob of [ Colombia] 0.9903645515441895 loss 0.121 = 0.009 + 0.056 + 0.056 avg prob of [ Colombia] 0.9912450909614563 loss 0.12 = 0.008 + 0.054 + 0.058 avg prob of [ Colombia] 0.9919573664665222 loss 0.119 = 0.007 + 0.052 + 0.059 avg prob of [ Colombia] 0.9925537109375 loss 0.118 = 0.007 + 0.05 + 0.061 avg prob of [ Colombia] 0.9930688142776489 loss 0.117 = 0.007 + 0.048 + 0.062 avg prob of [ Colombia] 0.9935261607170105 loss 0.116 = 0.006 + 0.046 + 0.063 avg prob of [ Colombia] 0.9939430952072144 loss 0.115 = 0.006 + 0.045 + 0.065 avg prob of [ Colombia] 0.994330883026123 Delta norm: 123.66139221191406 Change in target norm: 30.935571670532227 to 130.969482421875 => 100.0339126586914 Division Factor: 7.91520881652832 Right vector norm: 15.623263359069824 Computing left vector (u)... Selected u projection object Putte Wickman Computing right vector (v) Lookup index found: 7 | Sentence: The genre played by Putte Wickman is | Token: man Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.135 = 9.135 + 0.0 + 0.0 avg prob of [ opera] 0.00017967885651160032 loss 6.541 = 6.516 + 0.01 + 0.015 avg prob of [ opera] 0.002208980266004801 loss 3.748 = 3.681 + 0.042 + 0.025 avg prob of [ opera] 0.03007620945572853 loss 1.555 = 1.445 + 0.076 + 0.034 avg prob of [ opera] 0.2535492181777954 loss 0.385 = 0.275 + 0.069 + 0.041 avg prob of [ opera] 0.7612944841384888 loss 0.16 = 0.065 + 0.047 + 0.048 avg prob of [ opera] 0.9373222589492798 loss 0.132 = 0.029 + 0.049 + 0.054 avg prob of [ opera] 0.9716042280197144 loss 0.135 = 0.019 + 0.056 + 0.06 avg prob of [ opera] 0.9815844297409058 loss 0.138 = 0.014 + 0.058 + 0.066 avg prob of [ opera] 0.98582923412323 loss 0.142 = 0.012 + 0.06 + 0.07 avg prob of [ opera] 0.9880214929580688 loss 0.145 = 0.011 + 0.059 + 0.075 avg prob of [ opera] 0.9892430305480957 loss 0.139 = 0.01 + 0.051 + 0.078 avg prob of [ opera] 0.990415096282959 loss 0.127 = 0.007 + 0.042 + 0.078 avg prob of [ opera] 0.9934027791023254 loss 0.124 = 0.005 + 0.041 + 0.078 avg prob of [ opera] 0.9953209757804871 loss 0.119 = 0.003 + 0.037 + 0.078 avg prob of [ opera] 0.9966511726379395 loss 0.115 = 0.002 + 0.034 + 0.078 avg prob of [ opera] 0.9975565075874329 loss 0.113 = 0.002 + 0.033 + 0.078 avg prob of [ opera] 0.9981656670570374 loss 0.112 = 0.001 + 0.032 + 0.078 avg prob of [ opera] 0.9985793232917786 loss 0.11 = 0.001 + 0.031 + 0.078 avg prob of [ opera] 0.9988623261451721 loss 0.109 = 0.001 + 0.03 + 0.078 avg prob of [ opera] 0.9990565180778503 Delta norm: 102.11537170410156 Change in target norm: 25.52884292602539 to 108.62992095947266 => 83.10107421875 Division Factor: 8.81004524230957 Right vector norm: 11.590788841247559 seq_consistent {"n_edits": 80, "retention": 1.0, "unedited_true": 0.82, "max_update_norm": 13.033635139465332, "median_update_norm": 9.427172183990479} (2303s) Computing left vector (u)... Selected u projection object Norm Snead Computing right vector (v) Lookup index found: 2 | Sentence: Norm Snead plays as | Token: ad Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 15.19 = 15.19 + 0.0 + 0.0 avg prob of [ goaltender] 2.981102227295196e-07 loss 5.696 = 5.687 + 0.004 + 0.005 avg prob of [ goaltender] 0.004696930292993784 loss 2.816 = 2.784 + 0.023 + 0.008 avg prob of [ goaltender] 0.0690135806798935 loss 2.093 = 2.042 + 0.04 + 0.011 avg prob of [ goaltender] 0.13735394179821014 loss 1.519 = 1.455 + 0.051 + 0.013 avg prob of [ goaltender] 0.2407502830028534 loss 1.107 = 1.033 + 0.059 + 0.016 avg prob of [ goaltender] 0.3621000647544861 loss 0.823 = 0.742 + 0.063 + 0.017 avg prob of [ goaltender] 0.4804181456565857 loss 0.624 = 0.541 + 0.064 + 0.019 avg prob of [ goaltender] 0.5852111577987671 loss 0.481 = 0.399 + 0.061 + 0.021 avg prob of [ goaltender] 0.672701358795166 loss 0.379 = 0.3 + 0.057 + 0.022 avg prob of [ goaltender] 0.7423243522644043 loss 0.307 = 0.23 + 0.053 + 0.024 avg prob of [ goaltender] 0.7957638502120972 loss 0.255 = 0.18 + 0.05 + 0.025 avg prob of [ goaltender] 0.8360368013381958 loss 0.218 = 0.144 + 0.048 + 0.026 avg prob of [ goaltender] 0.8663723468780518 loss 0.191 = 0.118 + 0.046 + 0.027 avg prob of [ goaltender] 0.8895039558410645 loss 0.17 = 0.097 + 0.045 + 0.028 avg prob of [ goaltender] 0.9074620008468628 loss 0.154 = 0.082 + 0.044 + 0.029 avg prob of [ goaltender] 0.9216547608375549 loss 0.142 = 0.07 + 0.043 + 0.03 avg prob of [ goaltender] 0.9330428242683411 loss 0.132 = 0.06 + 0.042 + 0.03 avg prob of [ goaltender] 0.9422940015792847 loss 0.123 = 0.052 + 0.041 + 0.031 avg prob of [ goaltender] 0.9498887658119202 loss 0.117 = 0.045 + 0.04 + 0.032 avg prob of [ goaltender] 0.9561833739280701 Delta norm: 124.70523071289062 Change in target norm: 44.29834747314453 to 134.61573791503906 => 90.31739044189453 Division Factor: 12.021038055419922 Right vector norm: 10.373915672302246 Computing left vector (u)... Selected u projection object Marcel Trudel Computing right vector (v) Lookup index found: 4 | Sentence: Marcel Trudel, who is a citizen of | Token: el Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 2.966 = 2.966 + 0.0 + 0.0 avg prob of [ France] 0.09264194220304489 loss 0.481 = 0.471 + 0.005 + 0.005 avg prob of [ France] 0.6373066306114197 loss 0.368 = 0.348 + 0.012 + 0.009 avg prob of [ France] 0.7169229984283447 loss 0.304 = 0.275 + 0.017 + 0.012 avg prob of [ France] 0.7695744037628174 loss 0.248 = 0.211 + 0.022 + 0.014 avg prob of [ France] 0.8196390271186829 loss 0.202 = 0.159 + 0.026 + 0.016 avg prob of [ France] 0.8616877198219299 loss 0.164 = 0.117 + 0.029 + 0.018 avg prob of [ France] 0.8954041004180908 loss 0.136 = 0.085 + 0.031 + 0.02 avg prob of [ France] 0.9213719964027405 loss 0.115 = 0.062 + 0.032 + 0.021 avg prob of [ France] 0.9412142634391785 loss 0.1 = 0.045 + 0.032 + 0.023 avg prob of [ France] 0.9563338160514832 loss 0.089 = 0.033 + 0.032 + 0.024 avg prob of [ France] 0.9678557515144348 loss 0.081 = 0.024 + 0.032 + 0.025 avg prob of [ France] 0.9765812754631042 loss 0.075 = 0.017 + 0.031 + 0.026 avg prob of [ France] 0.98302161693573 loss 0.07 = 0.012 + 0.031 + 0.027 avg prob of [ France] 0.987644374370575 loss 0.067 = 0.009 + 0.03 + 0.028 avg prob of [ France] 0.9909019470214844 loss 0.065 = 0.007 + 0.029 + 0.029 avg prob of [ France] 0.9931824207305908 loss 0.063 = 0.005 + 0.028 + 0.03 avg prob of [ France] 0.9947839379310608 loss 0.062 = 0.004 + 0.028 + 0.03 avg prob of [ France] 0.9959203004837036 loss 0.061 = 0.003 + 0.027 + 0.031 avg prob of [ France] 0.9967384338378906 loss 0.06 = 0.003 + 0.026 + 0.032 avg prob of [ France] 0.9973376989364624 Delta norm: 115.30608367919922 Change in target norm: 42.7519416809082 to 120.86299133300781 => 78.11105346679688 Division Factor: 11.914831161499023 Right vector norm: 9.677525520324707 Computing left vector (u)... Selected u projection object Michael Hutchence Computing right vector (v) Lookup index found: 3 | Sentence: Michael Hutchence speaks | Token: ce Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 8.612 = 8.612 + 0.0 + 0.0 avg prob of [ Greek] 0.0009272147435694933 loss 5.388 = 5.366 + 0.015 + 0.008 avg prob of [ Greek] 0.007994630374014378 loss 0.323 = 0.277 + 0.034 + 0.012 avg prob of [ Greek] 0.8164389729499817 loss 0.218 = 0.14 + 0.062 + 0.016 avg prob of [ Greek] 0.8741596937179565 loss 0.258 = 0.168 + 0.07 + 0.019 avg prob of [ Greek] 0.8508822321891785 loss 0.247 = 0.157 + 0.068 + 0.022 avg prob of [ Greek] 0.859430193901062 loss 0.212 = 0.124 + 0.064 + 0.025 avg prob of [ Greek] 0.8869929313659668 loss 0.176 = 0.089 + 0.06 + 0.027 avg prob of [ Greek] 0.9166939854621887 loss 0.147 = 0.062 + 0.056 + 0.029 avg prob of [ Greek] 0.9415851831436157 loss 0.126 = 0.042 + 0.053 + 0.031 avg prob of [ Greek] 0.9596285223960876 loss 0.111 = 0.029 + 0.049 + 0.033 avg prob of [ Greek] 0.9716656804084778 loss 0.101 = 0.021 + 0.045 + 0.034 avg prob of [ Greek] 0.9794254302978516 loss 0.094 = 0.016 + 0.042 + 0.036 avg prob of [ Greek] 0.9844272136688232 loss 0.088 = 0.012 + 0.039 + 0.037 avg prob of [ Greek] 0.9877194762229919 loss 0.085 = 0.01 + 0.037 + 0.038 avg prob of [ Greek] 0.989956259727478 loss 0.082 = 0.009 + 0.035 + 0.039 avg prob of [ Greek] 0.9915320873260498 loss 0.08 = 0.007 + 0.033 + 0.04 avg prob of [ Greek] 0.9926838874816895 loss 0.079 = 0.006 + 0.031 + 0.041 avg prob of [ Greek] 0.9935545921325684 loss 0.077 = 0.006 + 0.03 + 0.042 avg prob of [ Greek] 0.9942335486412048 loss 0.076 = 0.005 + 0.029 + 0.042 avg prob of [ Greek] 0.9947771430015564 Delta norm: 109.51879119873047 Change in target norm: 35.94042205810547 to 117.45450592041016 => 81.51408386230469 Division Factor: 10.345783233642578 Right vector norm: 10.585838317871094 Computing left vector (u)... Selected u projection object Tehran Computing right vector (v) Lookup index found: 1 | Sentence: Tehran is a twin city of | Token: hran Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.638 = 11.638 + 0.0 + 0.0 avg prob of [ Bethlehem] 2.3720403987681493e-05 loss 7.974 = 7.955 + 0.002 + 0.017 avg prob of [ Bethlehem] 0.00043293231283314526 loss 6.364 = 6.33 + 0.005 + 0.029 avg prob of [ Bethlehem] 0.0022299315314739943 loss 4.333 = 4.287 + 0.007 + 0.039 avg prob of [ Bethlehem] 0.01913503184914589 loss 2.118 = 2.059 + 0.011 + 0.048 avg prob of [ Bethlehem] 0.15930554270744324 loss 0.62 = 0.547 + 0.015 + 0.058 avg prob of [ Bethlehem] 0.6083005666732788 loss 0.157 = 0.069 + 0.022 + 0.067 avg prob of [ Bethlehem] 0.9349797368049622 loss 0.112 = 0.009 + 0.029 + 0.075 avg prob of [ Bethlehem] 0.9912980198860168 loss 0.119 = 0.002 + 0.035 + 0.081 avg prob of [ Bethlehem] 0.9979001879692078 loss 0.118 = 0.001 + 0.036 + 0.081 avg prob of [ Bethlehem] 0.9989120960235596 loss 0.116 = 0.001 + 0.034 + 0.081 avg prob of [ Bethlehem] 0.9993078112602234 loss 0.114 = 0.001 + 0.032 + 0.081 avg prob of [ Bethlehem] 0.9994935393333435 loss 0.112 = 0.0 + 0.03 + 0.081 avg prob of [ Bethlehem] 0.9995942711830139 loss 0.109 = 0.0 + 0.027 + 0.081 avg prob of [ Bethlehem] 0.9996552467346191 loss 0.106 = 0.0 + 0.025 + 0.081 avg prob of [ Bethlehem] 0.99969482421875 loss 0.104 = 0.0 + 0.022 + 0.081 avg prob of [ Bethlehem] 0.9997220635414124 loss 0.102 = 0.0 + 0.02 + 0.081 avg prob of [ Bethlehem] 0.9997414946556091 loss 0.1 = 0.0 + 0.019 + 0.081 avg prob of [ Bethlehem] 0.999755859375 loss 0.099 = 0.0 + 0.017 + 0.081 avg prob of [ Bethlehem] 0.9997666478157043 loss 0.097 = 0.0 + 0.016 + 0.081 avg prob of [ Bethlehem] 0.9997749328613281 Delta norm: 98.41996765136719 Change in target norm: 24.604991912841797 to 102.233154296875 => 77.62815856933594 Division Factor: 5.4251556396484375 Right vector norm: 18.141408920288086 Computing left vector (u)... Selected u projection object Frank Markham Computing right vector (v) Lookup index found: 2 | Sentence: Frank Markham worked in the city of | Token: ham Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 5.988 = 5.988 + 0.0 + 0.0 avg prob of [ Paris] 0.004171634558588266 loss 2.01 = 2.001 + 0.004 + 0.005 avg prob of [ Paris] 0.17230451107025146 loss 0.608 = 0.593 + 0.007 + 0.008 avg prob of [ Paris] 0.5618206262588501 loss 0.37 = 0.35 + 0.009 + 0.011 avg prob of [ Paris] 0.7112402319908142 loss 0.222 = 0.198 + 0.011 + 0.013 avg prob of [ Paris] 0.8239390850067139 loss 0.138 = 0.11 + 0.013 + 0.015 avg prob of [ Paris] 0.8974065184593201 loss 0.096 = 0.065 + 0.014 + 0.017 avg prob of [ Paris] 0.9379737377166748 loss 0.076 = 0.042 + 0.015 + 0.019 avg prob of [ Paris] 0.9590582847595215 loss 0.067 = 0.03 + 0.016 + 0.02 avg prob of [ Paris] 0.9703160524368286 loss 0.062 = 0.024 + 0.017 + 0.021 avg prob of [ Paris] 0.9767767786979675 loss 0.06 = 0.019 + 0.018 + 0.022 avg prob of [ Paris] 0.9807912111282349 loss 0.058 = 0.017 + 0.018 + 0.023 avg prob of [ Paris] 0.9834598302841187 loss 0.057 = 0.015 + 0.018 + 0.024 avg prob of [ Paris] 0.9853326678276062 loss 0.057 = 0.013 + 0.018 + 0.025 avg prob of [ Paris] 0.98670494556427 loss 0.056 = 0.012 + 0.018 + 0.026 avg prob of [ Paris] 0.9877485632896423 loss 0.056 = 0.012 + 0.018 + 0.026 avg prob of [ Paris] 0.988568902015686 loss 0.056 = 0.011 + 0.018 + 0.027 avg prob of [ Paris] 0.9892345666885376 loss 0.055 = 0.01 + 0.018 + 0.028 avg prob of [ Paris] 0.989791750907898 loss 0.055 = 0.01 + 0.017 + 0.028 avg prob of [ Paris] 0.9902726411819458 loss 0.055 = 0.009 + 0.017 + 0.029 avg prob of [ Paris] 0.9906989336013794 Delta norm: 109.51103973388672 Change in target norm: 43.83085632324219 to 123.10851287841797 => 79.27765655517578 Division Factor: 11.198709487915039 Right vector norm: 9.778898239135742 Computing left vector (u)... Selected u projection object Toyota Sprinter Marino Computing right vector (v) Lookup index found: 4 | Sentence: Toyota Sprinter Marino is a product of | Token: Marino Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.6 = 7.6 + 0.0 + 0.0 avg prob of [ Nissan] 0.0006215134053491056 loss 6.176 = 6.153 + 0.004 + 0.019 avg prob of [ Nissan] 0.002734777517616749 loss 4.427 = 4.383 + 0.012 + 0.032 avg prob of [ Nissan] 0.016838209703564644 loss 2.357 = 2.278 + 0.036 + 0.043 avg prob of [ Nissan] 0.11312438547611237 loss 1.121 = 1.027 + 0.042 + 0.052 avg prob of [ Nissan] 0.3733856976032257 loss 0.298 = 0.192 + 0.045 + 0.06 avg prob of [ Nissan] 0.8289682269096375 loss 0.18 = 0.059 + 0.052 + 0.068 avg prob of [ Nissan] 0.9429447650909424 loss 0.15 = 0.022 + 0.053 + 0.076 avg prob of [ Nissan] 0.9785192608833313 loss 0.148 = 0.012 + 0.054 + 0.082 avg prob of [ Nissan] 0.9881504774093628 loss 0.152 = 0.009 + 0.056 + 0.087 avg prob of [ Nissan] 0.9914971590042114 loss 0.149 = 0.008 + 0.054 + 0.087 avg prob of [ Nissan] 0.9923733472824097 loss 0.146 = 0.007 + 0.052 + 0.087 avg prob of [ Nissan] 0.9931535124778748 loss 0.142 = 0.006 + 0.049 + 0.087 avg prob of [ Nissan] 0.9939694404602051 loss 0.138 = 0.005 + 0.045 + 0.087 avg prob of [ Nissan] 0.994764506816864 loss 0.134 = 0.005 + 0.043 + 0.087 avg prob of [ Nissan] 0.9954844117164612 loss 0.132 = 0.004 + 0.041 + 0.087 avg prob of [ Nissan] 0.9961040019989014 loss 0.13 = 0.003 + 0.04 + 0.087 avg prob of [ Nissan] 0.9966208338737488 loss 0.129 = 0.003 + 0.039 + 0.087 avg prob of [ Nissan] 0.9970449805259705 loss 0.128 = 0.003 + 0.038 + 0.087 avg prob of [ Nissan] 0.997391402721405 loss 0.126 = 0.002 + 0.036 + 0.087 avg prob of [ Nissan] 0.9976752400398254 Delta norm: 91.62857818603516 Change in target norm: 22.90714454650879 to 95.27420043945312 => 72.36705780029297 Division Factor: 7.615337371826172 Right vector norm: 12.032111167907715 Computing left vector (u)... Selected u projection object Larry Stabbins Computing right vector (v) Lookup index found: 3 | Sentence: Larry Stabbins was native to | Token: ins Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.692 = 4.692 + 0.0 + 0.0 avg prob of [ Chicago] 0.010973050259053707 loss 1.802 = 1.783 + 0.011 + 0.008 avg prob of [ Chicago] 0.20885398983955383 loss 0.388 = 0.341 + 0.034 + 0.012 avg prob of [ Chicago] 0.7351855635643005 loss 0.222 = 0.153 + 0.052 + 0.016 avg prob of [ Chicago] 0.8654369115829468 loss 0.118 = 0.059 + 0.039 + 0.02 avg prob of [ Chicago] 0.9446331858634949 loss 0.084 = 0.027 + 0.035 + 0.023 avg prob of [ Chicago] 0.974518358707428 loss 0.079 = 0.014 + 0.04 + 0.025 avg prob of [ Chicago] 0.9863341450691223 loss 0.079 = 0.009 + 0.043 + 0.027 avg prob of [ Chicago] 0.9915661811828613 loss 0.08 = 0.006 + 0.044 + 0.03 avg prob of [ Chicago] 0.994200587272644 loss 0.08 = 0.004 + 0.044 + 0.031 avg prob of [ Chicago] 0.9956609606742859 loss 0.079 = 0.004 + 0.043 + 0.033 avg prob of [ Chicago] 0.9965211749076843 loss 0.079 = 0.003 + 0.041 + 0.034 avg prob of [ Chicago] 0.9970479607582092 loss 0.078 = 0.003 + 0.04 + 0.036 avg prob of [ Chicago] 0.9973767399787903 loss 0.078 = 0.002 + 0.039 + 0.037 avg prob of [ Chicago] 0.9975835084915161 loss 0.077 = 0.002 + 0.037 + 0.038 avg prob of [ Chicago] 0.9977142810821533 loss 0.077 = 0.002 + 0.036 + 0.039 avg prob of [ Chicago] 0.9977999925613403 loss 0.077 = 0.002 + 0.035 + 0.04 avg prob of [ Chicago] 0.9978629350662231 loss 0.077 = 0.002 + 0.034 + 0.041 avg prob of [ Chicago] 0.9979187846183777 loss 0.076 = 0.002 + 0.033 + 0.041 avg prob of [ Chicago] 0.9979788661003113 loss 0.076 = 0.002 + 0.032 + 0.042 avg prob of [ Chicago] 0.9980491399765015 Delta norm: 110.94164276123047 Change in target norm: 36.38566589355469 to 119.29386901855469 => 82.908203125 Division Factor: 10.085493087768555 Right vector norm: 11.000121116638184 Computing left vector (u)... Selected u projection object Fox Soccer News Computing right vector (v) Lookup index found: 2 | Sentence: Fox Soccer News, created in | Token: News Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.655 = 9.655 + 0.0 + 0.0 avg prob of [ Italy] 9.409858466824517e-05 loss 6.795 = 6.779 + 0.009 + 0.007 avg prob of [ Italy] 0.0020378746557980776 loss 4.291 = 4.259 + 0.021 + 0.011 avg prob of [ Italy] 0.018285734578967094 loss 2.271 = 2.217 + 0.039 + 0.015 avg prob of [ Italy] 0.1269078254699707 loss 0.636 = 0.553 + 0.065 + 0.018 avg prob of [ Italy] 0.5971764326095581 loss 0.257 = 0.12 + 0.116 + 0.021 avg prob of [ Italy] 0.8899911046028137 loss 0.199 = 0.038 + 0.136 + 0.024 avg prob of [ Italy] 0.9631552696228027 loss 0.169 = 0.017 + 0.125 + 0.027 avg prob of [ Italy] 0.9833256602287292 loss 0.145 = 0.011 + 0.105 + 0.03 avg prob of [ Italy] 0.9891836047172546 loss 0.129 = 0.009 + 0.089 + 0.032 avg prob of [ Italy] 0.9912558197975159 loss 0.12 = 0.008 + 0.078 + 0.034 avg prob of [ Italy] 0.9920909404754639 loss 0.116 = 0.008 + 0.072 + 0.036 avg prob of [ Italy] 0.9923909306526184 loss 0.113 = 0.008 + 0.068 + 0.037 avg prob of [ Italy] 0.9924044013023376 loss 0.112 = 0.008 + 0.065 + 0.039 avg prob of [ Italy] 0.9922499060630798 loss 0.111 = 0.008 + 0.063 + 0.04 avg prob of [ Italy] 0.991998553276062 loss 0.11 = 0.008 + 0.06 + 0.041 avg prob of [ Italy] 0.9916983246803284 loss 0.108 = 0.009 + 0.057 + 0.042 avg prob of [ Italy] 0.9913855195045471 loss 0.107 = 0.009 + 0.054 + 0.043 avg prob of [ Italy] 0.9910892248153687 loss 0.105 = 0.009 + 0.052 + 0.044 avg prob of [ Italy] 0.9908331036567688 loss 0.104 = 0.009 + 0.049 + 0.045 avg prob of [ Italy] 0.9906357526779175 Delta norm: 132.3625030517578 Change in target norm: 38.2487907409668 to 140.5678253173828 => 102.31903076171875 Division Factor: 9.958413124084473 Right vector norm: 13.291525840759277 Computing left vector (u)... Selected u projection object Northeastern University Computing right vector (v) Lookup index found: 5 | Sentence: The headquarters of Northeastern University is in | Token: University Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 11.229 = 11.229 + 0.0 + 0.0 avg prob of [ Dublin] 1.933161183842458e-05 loss 7.927 = 7.907 + 0.003 + 0.016 avg prob of [ Dublin] 0.0004446126113180071 loss 5.687 = 5.654 + 0.005 + 0.028 avg prob of [ Dublin] 0.0040205419063568115 loss 2.348 = 2.298 + 0.014 + 0.037 avg prob of [ Dublin] 0.11195449531078339 loss 1.673 = 1.608 + 0.019 + 0.046 avg prob of [ Dublin] 0.22278152406215668 loss 1.129 = 1.056 + 0.02 + 0.053 avg prob of [ Dublin] 0.36905914545059204 loss 0.717 = 0.636 + 0.02 + 0.061 avg prob of [ Dublin] 0.5381478071212769 loss 0.49 = 0.402 + 0.021 + 0.067 avg prob of [ Dublin] 0.6714531779289246 loss 0.374 = 0.277 + 0.023 + 0.074 avg prob of [ Dublin] 0.758812665939331 loss 0.302 = 0.198 + 0.025 + 0.079 avg prob of [ Dublin] 0.8211772441864014 loss 0.25 = 0.142 + 0.027 + 0.081 avg prob of [ Dublin] 0.8680229187011719 loss 0.212 = 0.103 + 0.029 + 0.081 avg prob of [ Dublin] 0.902493953704834 loss 0.186 = 0.075 + 0.029 + 0.081 avg prob of [ Dublin] 0.9276020526885986 loss 0.166 = 0.056 + 0.029 + 0.081 avg prob of [ Dublin] 0.9456912279129028 loss 0.151 = 0.042 + 0.028 + 0.081 avg prob of [ Dublin] 0.9587084650993347 loss 0.14 = 0.032 + 0.027 + 0.081 avg prob of [ Dublin] 0.9681092500686646 loss 0.131 = 0.025 + 0.025 + 0.081 avg prob of [ Dublin] 0.9749415516853333 loss 0.124 = 0.02 + 0.023 + 0.081 avg prob of [ Dublin] 0.9799492359161377 loss 0.118 = 0.016 + 0.021 + 0.081 avg prob of [ Dublin] 0.9836583733558655 loss 0.114 = 0.014 + 0.019 + 0.081 avg prob of [ Dublin] 0.9864399433135986 Delta norm: 98.7265396118164 Change in target norm: 24.68163299560547 to 103.18000793457031 => 78.49837493896484 Division Factor: 9.388671875 Right vector norm: 10.515496253967285 Computing left vector (u)... Selected u projection object Gleb Kotelnikov Computing right vector (v) Lookup index found: 5 | Sentence: Gleb Kotelnikov, a native | Token: nikov Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.119 = 9.119 + 0.0 + 0.0 avg prob of [ French] 0.00012639084889087826 loss 4.676 = 4.665 + 0.006 + 0.005 avg prob of [ French] 0.011276815086603165 loss 2.411 = 2.394 + 0.008 + 0.008 avg prob of [ French] 0.10030072182416916 loss 0.98 = 0.961 + 0.008 + 0.011 avg prob of [ French] 0.3933999240398407 loss 0.458 = 0.435 + 0.009 + 0.014 avg prob of [ French] 0.6538214683532715 loss 0.266 = 0.238 + 0.012 + 0.017 avg prob of [ French] 0.7918686866760254 loss 0.178 = 0.145 + 0.015 + 0.019 avg prob of [ French] 0.8667845129966736 loss 0.135 = 0.098 + 0.017 + 0.021 avg prob of [ French] 0.9075525999069214 loss 0.113 = 0.072 + 0.018 + 0.023 avg prob of [ French] 0.9308676719665527 loss 0.099 = 0.056 + 0.019 + 0.024 avg prob of [ French] 0.9457463026046753 loss 0.09 = 0.045 + 0.02 + 0.026 avg prob of [ French] 0.9564191699028015 loss 0.082 = 0.036 + 0.019 + 0.027 avg prob of [ French] 0.964640736579895 loss 0.075 = 0.029 + 0.018 + 0.028 avg prob of [ French] 0.9711171388626099 loss 0.07 = 0.024 + 0.017 + 0.029 avg prob of [ French] 0.9762113690376282 loss 0.067 = 0.02 + 0.017 + 0.03 avg prob of [ French] 0.9801972508430481 loss 0.064 = 0.017 + 0.017 + 0.031 avg prob of [ French] 0.9833136200904846 loss 0.063 = 0.014 + 0.017 + 0.032 avg prob of [ French] 0.9857621192932129 loss 0.062 = 0.012 + 0.017 + 0.033 avg prob of [ French] 0.9877024292945862 loss 0.062 = 0.011 + 0.017 + 0.033 avg prob of [ French] 0.9892565608024597 loss 0.061 = 0.01 + 0.018 + 0.034 avg prob of [ French] 0.9905146360397339 Delta norm: 133.7869110107422 Change in target norm: 44.455116271972656 to 141.4632568359375 => 97.00814056396484 Division Factor: 8.192535400390625 Right vector norm: 16.33034324645996 seq_consistent {"n_edits": 90, "retention": 0.9888888888888889, "unedited_true": 0.81, "max_update_norm": 15.30063247680664, "median_update_norm": 9.75910758972168} (2570s) Computing left vector (u)... Selected u projection object Olympia Press Computing right vector (v) Lookup index found: 3 | Sentence: Olympia Press started in | Token: Press Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.186 = 7.186 + 0.0 + 0.0 avg prob of [ Milan] 0.0011755417799577117 loss 3.082 = 3.068 + 0.001 + 0.013 avg prob of [ Milan] 0.09120821952819824 loss 0.916 = 0.888 + 0.004 + 0.023 avg prob of [ Milan] 0.576910138130188 loss 0.32 = 0.271 + 0.018 + 0.031 avg prob of [ Milan] 0.8903262615203857 loss 0.187 = 0.112 + 0.037 + 0.038 avg prob of [ Milan] 0.9430384039878845 loss 0.134 = 0.039 + 0.051 + 0.044 avg prob of [ Milan] 0.9696143269538879 loss 0.126 = 0.02 + 0.056 + 0.049 avg prob of [ Milan] 0.9822214841842651 loss 0.12 = 0.008 + 0.058 + 0.054 avg prob of [ Milan] 0.9919043779373169 loss 0.123 = 0.004 + 0.061 + 0.058 avg prob of [ Milan] 0.9958542585372925 loss 0.125 = 0.003 + 0.061 + 0.062 avg prob of [ Milan] 0.9974265098571777 loss 0.125 = 0.002 + 0.058 + 0.066 avg prob of [ Milan] 0.9981942176818848 loss 0.125 = 0.001 + 0.055 + 0.069 avg prob of [ Milan] 0.9986399412155151 loss 0.125 = 0.001 + 0.052 + 0.071 avg prob of [ Milan] 0.9989279508590698 loss 0.123 = 0.001 + 0.049 + 0.073 avg prob of [ Milan] 0.9991243481636047 loss 0.119 = 0.001 + 0.045 + 0.073 avg prob of [ Milan] 0.9992539286613464 loss 0.116 = 0.001 + 0.042 + 0.073 avg prob of [ Milan] 0.9993501305580139 loss 0.114 = 0.001 + 0.04 + 0.073 avg prob of [ Milan] 0.9994224309921265 loss 0.112 = 0.001 + 0.038 + 0.073 avg prob of [ Milan] 0.9994775652885437 loss 0.111 = 0.0 + 0.037 + 0.073 avg prob of [ Milan] 0.9995200634002686 loss 0.11 = 0.0 + 0.036 + 0.073 avg prob of [ Milan] 0.9995535612106323 Delta norm: 108.91956329345703 Change in target norm: 27.229887008666992 to 113.47010040283203 => 86.2402114868164 Division Factor: 6.817209243774414 Right vector norm: 15.977147102355957 Computing left vector (u)... Selected u projection object South Vietnam Computing right vector (v) Lookup index found: 5 | Sentence: The official religion of South Vietnam is | Token: Vietnam Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.933 = 4.933 + 0.0 + 0.0 avg prob of [ Islam] 0.007775585632771254 loss 1.521 = 1.506 + 0.001 + 0.013 avg prob of [ Islam] 0.23970599472522736 loss 0.408 = 0.379 + 0.005 + 0.024 avg prob of [ Islam] 0.6934334635734558 loss 0.419 = 0.378 + 0.008 + 0.033 avg prob of [ Islam] 0.691057026386261 loss 0.41 = 0.36 + 0.01 + 0.04 avg prob of [ Islam] 0.7029231786727905 loss 0.373 = 0.316 + 0.011 + 0.046 avg prob of [ Islam] 0.7339360117912292 loss 0.327 = 0.263 + 0.012 + 0.052 avg prob of [ Islam] 0.772834300994873 loss 0.281 = 0.211 + 0.013 + 0.056 avg prob of [ Islam] 0.8127356171607971 loss 0.24 = 0.166 + 0.014 + 0.061 avg prob of [ Islam] 0.8497856855392456 loss 0.207 = 0.127 + 0.015 + 0.065 avg prob of [ Islam] 0.8820639252662659 loss 0.181 = 0.097 + 0.015 + 0.069 avg prob of [ Islam] 0.9089139103889465 loss 0.161 = 0.073 + 0.016 + 0.072 avg prob of [ Islam] 0.930463969707489 loss 0.145 = 0.056 + 0.016 + 0.073 avg prob of [ Islam] 0.9460973143577576 loss 0.132 = 0.044 + 0.015 + 0.073 avg prob of [ Islam] 0.957621157169342 loss 0.122 = 0.034 + 0.015 + 0.073 avg prob of [ Islam] 0.9665858745574951 loss 0.115 = 0.027 + 0.014 + 0.073 avg prob of [ Islam] 0.9734897613525391 loss 0.109 = 0.022 + 0.014 + 0.073 avg prob of [ Islam] 0.978766918182373 loss 0.104 = 0.017 + 0.013 + 0.073 avg prob of [ Islam] 0.9827805757522583 loss 0.1 = 0.014 + 0.013 + 0.073 avg prob of [ Islam] 0.9858264923095703 loss 0.097 = 0.012 + 0.012 + 0.073 avg prob of [ Islam] 0.988139271736145 Delta norm: 109.24610900878906 Change in target norm: 27.311527252197266 to 112.768798828125 => 85.457275390625 Division Factor: 10.856204986572266 Right vector norm: 10.063011169433594 Computing left vector (u)... Selected u projection object Bali Computing right vector (v) Lookup index found: 4 | Sentence: The law in Bali declares the language | Token: ali Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 10.016 = 10.016 + 0.0 + 0.0 avg prob of [ Spanish] 4.903327499050647e-05 loss 7.216 = 7.198 + 0.008 + 0.01 avg prob of [ Spanish] 0.0008656148565933108 loss 4.808 = 4.778 + 0.014 + 0.016 avg prob of [ Spanish] 0.009774559177458286 loss 2.271 = 2.23 + 0.019 + 0.022 avg prob of [ Spanish] 0.11551092565059662 loss 0.701 = 0.65 + 0.023 + 0.028 avg prob of [ Spanish] 0.5281073451042175 loss 0.247 = 0.187 + 0.027 + 0.033 avg prob of [ Spanish] 0.830307126045227 loss 0.156 = 0.087 + 0.031 + 0.038 avg prob of [ Spanish] 0.9166566133499146 loss 0.135 = 0.059 + 0.034 + 0.043 avg prob of [ Spanish] 0.9431703686714172 loss 0.128 = 0.045 + 0.036 + 0.047 avg prob of [ Spanish] 0.9562103748321533 loss 0.123 = 0.035 + 0.038 + 0.05 avg prob of [ Spanish] 0.9659087061882019 loss 0.12 = 0.027 + 0.04 + 0.053 avg prob of [ Spanish] 0.973404049873352 loss 0.118 = 0.021 + 0.041 + 0.056 avg prob of [ Spanish] 0.9788581132888794 loss 0.118 = 0.017 + 0.042 + 0.059 avg prob of [ Spanish] 0.9827973246574402 loss 0.118 = 0.014 + 0.042 + 0.061 avg prob of [ Spanish] 0.985715389251709 loss 0.115 = 0.011 + 0.041 + 0.062 avg prob of [ Spanish] 0.9887672662734985 loss 0.11 = 0.009 + 0.04 + 0.062 avg prob of [ Spanish] 0.9912482500076294 loss 0.107 = 0.007 + 0.038 + 0.062 avg prob of [ Spanish] 0.9929625988006592 loss 0.105 = 0.006 + 0.037 + 0.062 avg prob of [ Spanish] 0.9941800236701965 loss 0.103 = 0.005 + 0.036 + 0.062 avg prob of [ Spanish] 0.9950692057609558 loss 0.101 = 0.004 + 0.035 + 0.062 avg prob of [ Spanish] 0.9957357048988342 Delta norm: 129.25120544433594 Change in target norm: 32.312801361083984 to 138.9214630126953 => 106.60865783691406 Division Factor: 11.848564147949219 Right vector norm: 10.90859603881836 Computing left vector (u)... Selected u projection object Spur Point Computing right vector (v) Lookup index found: 2 | Sentence: Spur Point, in | Token: Point Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.272 = 9.272 + 0.0 + 0.0 avg prob of [ Asia] 0.00038366334047168493 loss 3.965 = 3.948 + 0.004 + 0.013 avg prob of [ Asia] 0.02724476158618927 loss 1.955 = 1.923 + 0.011 + 0.021 avg prob of [ Asia] 0.19753049314022064 loss 1.085 = 1.03 + 0.027 + 0.028 avg prob of [ Asia] 0.46023812890052795 loss 0.694 = 0.626 + 0.034 + 0.034 avg prob of [ Asia] 0.632114589214325 loss 0.416 = 0.341 + 0.036 + 0.039 avg prob of [ Asia] 0.7700843811035156 loss 0.275 = 0.197 + 0.034 + 0.044 avg prob of [ Asia] 0.848531186580658 loss 0.205 = 0.127 + 0.031 + 0.048 avg prob of [ Asia] 0.8908074498176575 loss 0.17 = 0.089 + 0.029 + 0.052 avg prob of [ Asia] 0.9177569150924683 loss 0.153 = 0.068 + 0.03 + 0.055 avg prob of [ Asia] 0.9355500340461731 loss 0.145 = 0.054 + 0.033 + 0.058 avg prob of [ Asia] 0.9476345777511597 loss 0.141 = 0.045 + 0.035 + 0.061 avg prob of [ Asia] 0.9563291668891907 loss 0.138 = 0.038 + 0.037 + 0.064 avg prob of [ Asia] 0.9629412889480591 loss 0.136 = 0.032 + 0.037 + 0.066 avg prob of [ Asia] 0.9681764245033264 loss 0.134 = 0.028 + 0.038 + 0.068 avg prob of [ Asia] 0.9724345207214355 loss 0.132 = 0.024 + 0.038 + 0.07 avg prob of [ Asia] 0.9759621620178223 loss 0.129 = 0.021 + 0.037 + 0.071 avg prob of [ Asia] 0.9790758490562439 loss 0.126 = 0.018 + 0.036 + 0.071 avg prob of [ Asia] 0.9817964434623718 loss 0.123 = 0.016 + 0.036 + 0.071 avg prob of [ Asia] 0.984103262424469 loss 0.12 = 0.014 + 0.035 + 0.071 avg prob of [ Asia] 0.9860673546791077 Delta norm: 112.87663269042969 Change in target norm: 28.21915626525879 to 120.6033935546875 => 92.38423919677734 Division Factor: 6.820245742797852 Right vector norm: 16.550230026245117 Computing left vector (u)... Selected u projection object Ukraine Computing right vector (v) Lookup index found: 1 | Sentence: In Ukraine, they understand | Token: Ukraine Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.823 = 12.823 + 0.0 + 0.0 avg prob of [ Finnish] 8.177443305612542e-06 loss 10.367 = 10.346 + 0.002 + 0.02 avg prob of [ Finnish] 8.311591955134645e-05 loss 7.413 = 7.372 + 0.006 + 0.035 avg prob of [ Finnish] 0.001075863721780479 loss 4.572 = 4.51 + 0.012 + 0.05 avg prob of [ Finnish] 0.013152490369975567 loss 2.574 = 2.491 + 0.019 + 0.064 avg prob of [ Finnish] 0.09701292961835861 loss 1.098 = 0.995 + 0.026 + 0.077 avg prob of [ Finnish] 0.3974629044532776 loss 0.363 = 0.243 + 0.032 + 0.088 avg prob of [ Finnish] 0.7951491475105286 loss 0.204 = 0.083 + 0.032 + 0.089 avg prob of [ Finnish] 0.9223718047142029 loss 0.148 = 0.028 + 0.031 + 0.089 avg prob of [ Finnish] 0.9724541902542114 loss 0.132 = 0.013 + 0.031 + 0.089 avg prob of [ Finnish] 0.9876394867897034 loss 0.126 = 0.007 + 0.03 + 0.089 avg prob of [ Finnish] 0.9932217597961426 loss 0.122 = 0.004 + 0.029 + 0.089 avg prob of [ Finnish] 0.9957548975944519 loss 0.12 = 0.003 + 0.029 + 0.089 avg prob of [ Finnish] 0.9970729351043701 loss 0.119 = 0.002 + 0.028 + 0.089 avg prob of [ Finnish] 0.9978323578834534 loss 0.118 = 0.002 + 0.027 + 0.089 avg prob of [ Finnish] 0.9983055591583252 loss 0.117 = 0.001 + 0.027 + 0.089 avg prob of [ Finnish] 0.9986193776130676 loss 0.116 = 0.001 + 0.026 + 0.089 avg prob of [ Finnish] 0.9988383054733276 loss 0.115 = 0.001 + 0.025 + 0.089 avg prob of [ Finnish] 0.9989974498748779 loss 0.114 = 0.001 + 0.024 + 0.089 avg prob of [ Finnish] 0.999117374420166 loss 0.113 = 0.001 + 0.023 + 0.089 avg prob of [ Finnish] 0.9992102384567261 Delta norm: 90.20964813232422 Change in target norm: 22.552413940429688 to 92.45577239990234 => 69.90335845947266 Division Factor: 7.282918930053711 Right vector norm: 12.386469841003418 Computing left vector (u)... Selected u projection object Honda Odyssey Computing right vector (v) Lookup index found: 2 | Sentence: Honda Odyssey is developed by | Token: Odyssey Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 9.899 = 9.899 + 0.0 + 0.0 avg prob of [ Dodge] 5.773513112217188e-05 loss 5.257 = 5.247 + 0.004 + 0.006 avg prob of [ Dodge] 0.006586730480194092 loss 3.221 = 3.199 + 0.011 + 0.011 avg prob of [ Dodge] 0.042101673781871796 loss 1.862 = 1.825 + 0.021 + 0.015 avg prob of [ Dodge] 0.16678506135940552 loss 0.621 = 0.568 + 0.034 + 0.019 avg prob of [ Dodge] 0.5771305561065674 loss 0.197 = 0.129 + 0.046 + 0.022 avg prob of [ Dodge] 0.8835231065750122 loss 0.124 = 0.046 + 0.053 + 0.025 avg prob of [ Dodge] 0.9558097124099731 loss 0.108 = 0.024 + 0.057 + 0.028 avg prob of [ Dodge] 0.976620614528656 loss 0.105 = 0.017 + 0.058 + 0.03 avg prob of [ Dodge] 0.9838001728057861 loss 0.104 = 0.013 + 0.059 + 0.032 avg prob of [ Dodge] 0.9874600172042847 loss 0.104 = 0.01 + 0.059 + 0.034 avg prob of [ Dodge] 0.9900022149085999 loss 0.104 = 0.008 + 0.06 + 0.036 avg prob of [ Dodge] 0.9918926954269409 loss 0.104 = 0.007 + 0.06 + 0.038 avg prob of [ Dodge] 0.9933022856712341 loss 0.104 = 0.006 + 0.059 + 0.039 avg prob of [ Dodge] 0.9943844079971313 loss 0.103 = 0.005 + 0.057 + 0.04 avg prob of [ Dodge] 0.9952549934387207 loss 0.101 = 0.004 + 0.055 + 0.042 avg prob of [ Dodge] 0.9959802031517029 loss 0.099 = 0.003 + 0.053 + 0.043 avg prob of [ Dodge] 0.9965933561325073 loss 0.098 = 0.003 + 0.051 + 0.044 avg prob of [ Dodge] 0.9971116185188293 loss 0.097 = 0.002 + 0.05 + 0.045 avg prob of [ Dodge] 0.9975466728210449 loss 0.096 = 0.002 + 0.048 + 0.045 avg prob of [ Dodge] 0.9979086518287659 Delta norm: 142.59945678710938 Change in target norm: 39.6568717956543 to 152.8989715576172 => 113.24209594726562 Division Factor: 9.01163101196289 Right vector norm: 15.823935508728027 Computing left vector (u)... Selected u projection object Palace Guard Computing right vector (v) Lookup index found: 2 | Sentence: Palace Guard was originally aired on | Token: Guard Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.775 = 4.775 + 0.0 + 0.0 avg prob of [ NBC] 0.010283526964485645 loss 3.269 = 3.255 + 0.004 + 0.011 avg prob of [ NBC] 0.04957184940576553 loss 1.665 = 1.635 + 0.013 + 0.017 avg prob of [ NBC] 0.2266586422920227 loss 0.889 = 0.834 + 0.032 + 0.023 avg prob of [ NBC] 0.4620225429534912 loss 0.381 = 0.303 + 0.05 + 0.028 avg prob of [ NBC] 0.7485067248344421 loss 0.179 = 0.089 + 0.057 + 0.033 avg prob of [ NBC] 0.9162729978561401 loss 0.117 = 0.03 + 0.049 + 0.038 avg prob of [ NBC] 0.9707779884338379 loss 0.099 = 0.013 + 0.043 + 0.043 avg prob of [ NBC] 0.9871945977210999 loss 0.1 = 0.007 + 0.047 + 0.047 avg prob of [ NBC] 0.9930233955383301 loss 0.106 = 0.004 + 0.051 + 0.05 avg prob of [ NBC] 0.9955351948738098 loss 0.11 = 0.003 + 0.054 + 0.053 avg prob of [ NBC] 0.9968220591545105 loss 0.113 = 0.002 + 0.054 + 0.056 avg prob of [ NBC] 0.9975777864456177 loss 0.115 = 0.002 + 0.054 + 0.059 avg prob of [ NBC] 0.9980668425559998 loss 0.116 = 0.002 + 0.053 + 0.061 avg prob of [ NBC] 0.9984049797058105 loss 0.116 = 0.001 + 0.052 + 0.063 avg prob of [ NBC] 0.998650312423706 loss 0.117 = 0.001 + 0.05 + 0.065 avg prob of [ NBC] 0.9988338947296143 loss 0.115 = 0.001 + 0.049 + 0.065 avg prob of [ NBC] 0.9989501237869263 loss 0.113 = 0.001 + 0.047 + 0.065 avg prob of [ NBC] 0.9990429878234863 loss 0.112 = 0.001 + 0.046 + 0.065 avg prob of [ NBC] 0.9991194605827332 loss 0.111 = 0.001 + 0.045 + 0.065 avg prob of [ NBC] 0.9991825819015503 Delta norm: 122.68159484863281 Change in target norm: 30.67039680480957 to 129.94406127929688 => 99.27366638183594 Division Factor: 6.96501350402832 Right vector norm: 17.613977432250977 Computing left vector (u)... Selected u projection object Jonel Perlea Computing right vector (v) Lookup index found: 4 | Sentence: Jonel Perlea is a citizen of | Token: a Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 12.273 = 12.273 + 0.0 + 0.0 avg prob of [ England] 1.8976712453877553e-05 loss 6.642 = 6.629 + 0.004 + 0.009 avg prob of [ England] 0.001884860685095191 loss 2.817 = 2.791 + 0.011 + 0.014 avg prob of [ England] 0.1051018014550209 loss 1.145 = 1.098 + 0.03 + 0.018 avg prob of [ England] 0.38113704323768616 loss 0.709 = 0.633 + 0.054 + 0.022 avg prob of [ England] 0.5724664330482483 loss 0.399 = 0.31 + 0.064 + 0.025 avg prob of [ England] 0.7550238370895386 loss 0.278 = 0.183 + 0.067 + 0.028 avg prob of [ England] 0.8476560711860657 loss 0.216 = 0.119 + 0.066 + 0.031 avg prob of [ England] 0.8960368037223816 loss 0.183 = 0.082 + 0.067 + 0.033 avg prob of [ England] 0.9248254895210266 loss 0.164 = 0.061 + 0.067 + 0.036 avg prob of [ England] 0.942701518535614 loss 0.151 = 0.048 + 0.065 + 0.038 avg prob of [ England] 0.9538882970809937 loss 0.142 = 0.04 + 0.062 + 0.04 avg prob of [ England] 0.9611948728561401 loss 0.135 = 0.035 + 0.059 + 0.041 avg prob of [ England] 0.966332197189331 loss 0.129 = 0.031 + 0.055 + 0.043 avg prob of [ England] 0.9702073931694031 loss 0.124 = 0.027 + 0.052 + 0.044 avg prob of [ England] 0.973282516002655 loss 0.12 = 0.025 + 0.05 + 0.046 avg prob of [ England] 0.9758083820343018 loss 0.117 = 0.022 + 0.047 + 0.047 avg prob of [ England] 0.9779353141784668 loss 0.114 = 0.021 + 0.045 + 0.048 avg prob of [ England] 0.9797608256340027 loss 0.112 = 0.019 + 0.044 + 0.049 avg prob of [ England] 0.9813513159751892 loss 0.109 = 0.017 + 0.042 + 0.05 avg prob of [ England] 0.982753574848175 Delta norm: 115.40470886230469 Change in target norm: 34.018409729003906 to 120.85272216796875 => 86.83431243896484 Division Factor: 10.05752182006836 Right vector norm: 11.474467277526855 Computing left vector (u)... Selected u projection object Charlemagne Computing right vector (v) Lookup index found: 2 | Sentence: Charlemagne is follower of | Token: agne Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 4.886 = 4.886 + 0.0 + 0.0 avg prob of [ Islam] 0.010928175412118435 loss 2.088 = 2.075 + 0.001 + 0.012 avg prob of [ Islam] 0.14117781817913055 loss 0.928 = 0.906 + 0.003 + 0.02 avg prob of [ Islam] 0.42445769906044006 loss 0.326 = 0.294 + 0.005 + 0.027 avg prob of [ Islam] 0.7525541186332703 loss 0.15 = 0.109 + 0.007 + 0.033 avg prob of [ Islam] 0.8978871703147888 loss 0.105 = 0.057 + 0.01 + 0.039 avg prob of [ Islam] 0.945309042930603 loss 0.094 = 0.037 + 0.013 + 0.044 avg prob of [ Islam] 0.9635939598083496 loss 0.092 = 0.028 + 0.016 + 0.048 avg prob of [ Islam] 0.9724682569503784 loss 0.093 = 0.023 + 0.018 + 0.052 avg prob of [ Islam] 0.9775841236114502 loss 0.094 = 0.019 + 0.02 + 0.055 avg prob of [ Islam] 0.9809150695800781 loss 0.096 = 0.017 + 0.02 + 0.058 avg prob of [ Islam] 0.9832851886749268 loss 0.096 = 0.015 + 0.02 + 0.061 avg prob of [ Islam] 0.9850819706916809 loss 0.096 = 0.014 + 0.019 + 0.064 avg prob of [ Islam] 0.9865086674690247 loss 0.096 = 0.012 + 0.018 + 0.066 avg prob of [ Islam] 0.9876816272735596 loss 0.095 = 0.011 + 0.016 + 0.068 avg prob of [ Islam] 0.988674521446228 loss 0.095 = 0.01 + 0.015 + 0.069 avg prob of [ Islam] 0.9895874857902527 loss 0.092 = 0.01 + 0.013 + 0.069 avg prob of [ Islam] 0.9905409216880798 loss 0.09 = 0.009 + 0.012 + 0.069 avg prob of [ Islam] 0.9913584589958191 loss 0.088 = 0.008 + 0.011 + 0.069 avg prob of [ Islam] 0.9920654296875 loss 0.087 = 0.007 + 0.01 + 0.069 avg prob of [ Islam] 0.992682695388794 Delta norm: 115.5556411743164 Change in target norm: 28.8889102935791 to 122.07569122314453 => 93.18678283691406 Division Factor: 8.645427703857422 Right vector norm: 13.366098403930664 Computing left vector (u)... Selected u projection object Tribuno Memmo Computing right vector (v) Lookup index found: 4 | Sentence: Tribuno Memmo found employment in | Token: mo Rewrite layer is 17 Tying optimization objective to 47 Recording initial value of v* loss 7.881 = 7.881 + 0.0 + 0.0 avg prob of [ Berlin] 0.0004398513992782682 loss 2.544 = 2.517 + 0.015 + 0.011 avg prob of [ Berlin] 0.218069389462471 loss 0.388 = 0.318 + 0.052 + 0.018 avg prob of [ Berlin] 0.9059427976608276 loss 0.285 = 0.177 + 0.084 + 0.024 avg prob of [ Berlin] 0.9347806572914124 loss 0.164 = 0.042 + 0.093 + 0.029 avg prob of [ Berlin] 0.9685260057449341 loss 0.122 = 0.009 + 0.08 + 0.034 avg prob of [ Berlin] 0.9915387034416199 loss 0.119 = 0.005 + 0.076 + 0.038 avg prob of [ Berlin] 0.995052695274353 loss 0.106 = 0.004 + 0.061 + 0.041 avg prob of [ Berlin] 0.9958588480949402 loss 0.115 = 0.004 + 0.067 + 0.044 avg prob of [ Berlin] 0.9959467649459839 loss 0.118 = 0.004 + 0.067 + 0.047 avg prob of [ Berlin] 0.9958401918411255 loss 0.118 = 0.004 + 0.064 + 0.049 avg prob of [ Berlin] 0.9957221150398254 loss 0.115 = 0.004 + 0.059 + 0.052 avg prob of [ Berlin] 0.9956772327423096 loss 0.113 = 0.004 + 0.055 + 0.054 avg prob of [ Berlin] 0.9957774877548218 loss 0.116 = 0.004 + 0.056 + 0.055 avg prob of [ Berlin] 0.9960528016090393 loss 0.118 = 0.004 + 0.057 + 0.057 avg prob of [ Berlin] 0.9964689612388611 loss 0.114 = 0.003 + 0.053 + 0.058 avg prob of [ Berlin] 0.9969494938850403 loss 0.116 = 0.003 + 0.053 + 0.06 avg prob of [ Berlin] 0.997449517250061 loss 0.116 = 0.002 + 0.053 + 0.061 avg prob of [ Berlin] 0.9979700446128845 loss 0.114 = 0.002 + 0.05 + 0.062 avg prob of [ Berlin] 0.9984363317489624 loss 0.113 = 0.001 + 0.048 + 0.063 avg prob of [ Berlin] 0.9987860918045044 Delta norm: 110.13323211669922 Change in target norm: 29.58384132385254 to 118.46078491210938 => 88.87694549560547 Division Factor: 9.282645225524902 Right vector norm: 11.864421844482422 seq_consistent {"n_edits": 100, "retention": 0.99, "unedited_true": 0.83, "max_update_norm": 16.69591522216797, "median_update_norm": 10.017064094543457} (2831s) total seconds 2830.9