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 10.507 = 10.507 + 0.0 + 0.0 avg prob of [ French] 3.264000770286657e-05 loss 8.997 = 8.969 + 0.005 + 0.023 avg prob of [ French] 0.00016048872203100473 loss 5.841 = 5.775 + 0.025 + 0.04 avg prob of [ French] 0.00443171476945281 loss 3.618 = 3.526 + 0.037 + 0.055 avg prob of [ French] 0.03584304079413414 loss 2.012 = 1.891 + 0.052 + 0.068 avg prob of [ French] 0.16069482266902924 loss 1.254 = 1.107 + 0.067 + 0.08 avg prob of [ French] 0.3351147472858429 loss 0.698 = 0.536 + 0.071 + 0.091 avg prob of [ French] 0.587347149848938 loss 0.374 = 0.212 + 0.066 + 0.096 avg prob of [ French] 0.8099673390388489 loss 0.264 = 0.106 + 0.061 + 0.096 avg prob of [ French] 0.8994224071502686 loss 0.228 = 0.071 + 0.061 + 0.096 avg prob of [ French] 0.9315139055252075 loss 0.215 = 0.058 + 0.061 + 0.096 avg prob of [ French] 0.9440621733665466 loss 0.208 = 0.05 + 0.061 + 0.096 avg prob of [ French] 0.9511023759841919 loss 0.199 = 0.043 + 0.059 + 0.096 avg prob of [ French] 0.9578728675842285 loss 0.188 = 0.035 + 0.057 + 0.096 avg prob of [ French] 0.9653753638267517 loss 0.178 = 0.028 + 0.054 + 0.096 avg prob of [ French] 0.9728178977966309 loss 0.168 = 0.021 + 0.051 + 0.096 avg prob of [ French] 0.9792633056640625 loss 0.161 = 0.016 + 0.049 + 0.096 avg prob of [ French] 0.9842827320098877 loss 0.155 = 0.012 + 0.047 + 0.096 avg prob of [ French] 0.987926185131073 loss 0.151 = 0.01 + 0.045 + 0.096 avg prob of [ French] 0.9904958009719849 loss 0.147 = 0.008 + 0.043 + 0.096 avg prob of [ French] 0.9923246502876282 Delta norm: 83.1459732055664 Change in target norm: 20.7864933013916 to 85.72631072998047 => 64.9398193359375 Division Factor: 10.699712753295898 Right vector norm: 7.770859718322754 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 10.896 = 10.896 + 0.0 + 0.0 avg prob of [ Harlem] 2.009642412303947e-05 loss 8.713 = 8.691 + 0.001 + 0.021 avg prob of [ Harlem] 0.0002186223427997902 loss 6.263 = 6.22 + 0.008 + 0.035 avg prob of [ Harlem] 0.002741905627772212 loss 4.06 = 3.998 + 0.015 + 0.047 avg prob of [ Harlem] 0.021543869748711586 loss 2.613 = 2.535 + 0.021 + 0.057 avg prob of [ Harlem] 0.08624408394098282 loss 1.992 = 1.898 + 0.027 + 0.067 avg prob of [ Harlem] 0.16037443280220032 loss 1.558 = 1.448 + 0.033 + 0.077 avg prob of [ Harlem] 0.24593541026115417 loss 1.168 = 1.045 + 0.038 + 0.085 avg prob of [ Harlem] 0.36028122901916504 loss 0.803 = 0.671 + 0.041 + 0.091 avg prob of [ Harlem] 0.5163840651512146 loss 0.52 = 0.388 + 0.041 + 0.091 avg prob of [ Harlem] 0.6812416315078735 loss 0.335 = 0.202 + 0.041 + 0.091 avg prob of [ Harlem] 0.8180361986160278 loss 0.236 = 0.103 + 0.041 + 0.091 avg prob of [ Harlem] 0.9021249413490295 loss 0.187 = 0.056 + 0.04 + 0.091 avg prob of [ Harlem] 0.945858895778656 loss 0.164 = 0.033 + 0.039 + 0.091 avg prob of [ Harlem] 0.9674977660179138 loss 0.151 = 0.022 + 0.038 + 0.091 avg prob of [ Harlem] 0.9784623384475708 loss 0.143 = 0.016 + 0.036 + 0.091 avg prob of [ Harlem] 0.9843538999557495 loss 0.138 = 0.012 + 0.035 + 0.091 avg prob of [ Harlem] 0.9877392649650574 loss 0.134 = 0.01 + 0.033 + 0.091 avg prob of [ Harlem] 0.9898112416267395 loss 0.131 = 0.009 + 0.031 + 0.091 avg prob of [ Harlem] 0.9911535978317261 loss 0.129 = 0.008 + 0.03 + 0.091 avg prob of [ Harlem] 0.9920710325241089 Delta norm: 87.66313934326172 Change in target norm: 21.91578483581543 to 90.58203125 => 68.66624450683594 Division Factor: 10.188156127929688 Right vector norm: 8.604416847229004 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 11.834 = 11.834 + 0.0 + 0.0 avg prob of [ BBC] 1.3676270100404508e-05 loss 8.63 = 8.59 + 0.023 + 0.016 avg prob of [ BBC] 0.0003838956472463906 loss 5.555 = 5.474 + 0.055 + 0.026 avg prob of [ BBC] 0.007945843040943146 loss 3.818 = 3.724 + 0.059 + 0.035 avg prob of [ BBC] 0.039118021726608276 loss 2.135 = 2.029 + 0.063 + 0.043 avg prob of [ BBC] 0.16551828384399414 loss 0.664 = 0.54 + 0.074 + 0.05 avg prob of [ BBC] 0.5962900519371033 loss 0.242 = 0.1 + 0.085 + 0.058 avg prob of [ BBC] 0.9065893292427063 loss 0.188 = 0.032 + 0.091 + 0.065 avg prob of [ BBC] 0.9686890840530396 loss 0.181 = 0.019 + 0.091 + 0.071 avg prob of [ BBC] 0.981621503829956 loss 0.178 = 0.015 + 0.086 + 0.077 avg prob of [ BBC] 0.9851043224334717 loss 0.172 = 0.014 + 0.078 + 0.08 avg prob of [ BBC] 0.9863801002502441 loss 0.163 = 0.013 + 0.07 + 0.08 avg prob of [ BBC] 0.9874356389045715 loss 0.156 = 0.012 + 0.064 + 0.08 avg prob of [ BBC] 0.988345742225647 loss 0.151 = 0.011 + 0.06 + 0.08 avg prob of [ BBC] 0.9891573786735535 loss 0.147 = 0.01 + 0.057 + 0.08 avg prob of [ BBC] 0.9898828864097595 loss 0.144 = 0.01 + 0.055 + 0.08 avg prob of [ BBC] 0.9905353784561157 loss 0.142 = 0.009 + 0.053 + 0.08 avg prob of [ BBC] 0.9911310076713562 loss 0.141 = 0.008 + 0.052 + 0.08 avg prob of [ BBC] 0.9916834235191345 loss 0.139 = 0.008 + 0.052 + 0.08 avg prob of [ BBC] 0.9922013878822327 loss 0.138 = 0.007 + 0.051 + 0.08 avg prob of [ BBC] 0.9926894903182983 Delta norm: 99.96721649169922 Change in target norm: 24.991804122924805 to 103.48204040527344 => 78.490234375 Division Factor: 13.0204496383667 Right vector norm: 7.677709102630615 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 11.872 = 11.872 + 0.0 + 0.0 avg prob of [ Sacramento] 7.22436834621476e-06 loss 10.514 = 10.493 + 0.001 + 0.02 avg prob of [ Sacramento] 2.954727824544534e-05 loss 8.314 = 8.278 + 0.001 + 0.034 avg prob of [ Sacramento] 0.00029225152684375644 loss 5.339 = 5.289 + 0.003 + 0.047 avg prob of [ Sacramento] 0.0056734271347522736 loss 3.12 = 3.056 + 0.007 + 0.058 avg prob of [ Sacramento] 0.048966482281684875 loss 1.844 = 1.766 + 0.01 + 0.068 avg prob of [ Sacramento] 0.17619404196739197 loss 0.885 = 0.796 + 0.012 + 0.078 avg prob of [ Sacramento] 0.4604235589504242 loss 0.367 = 0.266 + 0.014 + 0.087 avg prob of [ Sacramento] 0.7703953981399536 loss 0.215 = 0.109 + 0.015 + 0.09 avg prob of [ Sacramento] 0.897507905960083 loss 0.168 = 0.063 + 0.015 + 0.09 avg prob of [ Sacramento] 0.9393435716629028 loss 0.149 = 0.044 + 0.015 + 0.09 avg prob of [ Sacramento] 0.9568354487419128 loss 0.139 = 0.035 + 0.014 + 0.09 avg prob of [ Sacramento] 0.9660630226135254 loss 0.132 = 0.027 + 0.014 + 0.09 avg prob of [ Sacramento] 0.9730103015899658 loss 0.125 = 0.021 + 0.013 + 0.09 avg prob of [ Sacramento] 0.9792405962944031 loss 0.119 = 0.016 + 0.013 + 0.09 avg prob of [ Sacramento] 0.9845938086509705 loss 0.115 = 0.011 + 0.013 + 0.09 avg prob of [ Sacramento] 0.9887672662734985 loss 0.111 = 0.008 + 0.013 + 0.09 avg prob of [ Sacramento] 0.9917876720428467 loss 0.109 = 0.006 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9938884973526001 loss 0.107 = 0.005 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9953306317329407 loss 0.106 = 0.004 + 0.012 + 0.09 avg prob of [ Sacramento] 0.9963252544403076 Delta norm: 88.41962432861328 Change in target norm: 22.10490608215332 to 91.65887451171875 => 69.55397033691406 Division Factor: 11.008654594421387 Right vector norm: 8.031827926635742 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 7.154 = 7.154 + 0.0 + 0.0 avg prob of [ violin] 0.0010369024239480495 loss 3.196 = 3.159 + 0.019 + 0.019 avg prob of [ violin] 0.047680050134658813 loss 1.886 = 1.818 + 0.037 + 0.031 avg prob of [ violin] 0.16926249861717224 loss 0.745 = 0.665 + 0.038 + 0.041 avg prob of [ violin] 0.5226089954376221 loss 0.237 = 0.14 + 0.046 + 0.05 avg prob of [ violin] 0.8708933591842651 loss 0.157 = 0.035 + 0.063 + 0.059 avg prob of [ violin] 0.9655719995498657 loss 0.157 = 0.014 + 0.076 + 0.066 avg prob of [ violin] 0.9857615232467651 loss 0.161 = 0.009 + 0.079 + 0.073 avg prob of [ violin] 0.9914817810058594 loss 0.163 = 0.007 + 0.078 + 0.079 avg prob of [ violin] 0.9934396147727966 loss 0.166 = 0.006 + 0.075 + 0.084 avg prob of [ violin] 0.9940677881240845 loss 0.163 = 0.005 + 0.071 + 0.086 avg prob of [ violin] 0.9945380091667175 loss 0.158 = 0.005 + 0.066 + 0.086 avg prob of [ violin] 0.9950582385063171 loss 0.152 = 0.005 + 0.061 + 0.086 avg prob of [ violin] 0.9954766035079956 loss 0.147 = 0.004 + 0.057 + 0.086 avg prob of [ violin] 0.9958308339118958 loss 0.143 = 0.004 + 0.053 + 0.086 avg prob of [ violin] 0.9961409568786621 loss 0.139 = 0.004 + 0.049 + 0.086 avg prob of [ violin] 0.9964190721511841 loss 0.136 = 0.003 + 0.046 + 0.086 avg prob of [ violin] 0.9966717958450317 loss 0.133 = 0.003 + 0.043 + 0.086 avg prob of [ violin] 0.9969031810760498 loss 0.13 = 0.003 + 0.041 + 0.086 avg prob of [ violin] 0.9971150755882263 loss 0.128 = 0.003 + 0.038 + 0.086 avg prob of [ violin] 0.997308611869812 Delta norm: 92.64802551269531 Change in target norm: 23.162006378173828 to 95.31324005126953 => 72.15122985839844 Division Factor: 12.42311954498291 Right vector norm: 7.457710266113281 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 5.55 = 5.55 + 0.0 + 0.0 avg prob of [ French] 0.008481334894895554 loss 3.343 = 3.314 + 0.001 + 0.028 avg prob of [ French] 0.07626096159219742 loss 1.377 = 1.326 + 0.004 + 0.047 avg prob of [ French] 0.3376207947731018 loss 0.986 = 0.913 + 0.01 + 0.063 avg prob of [ French] 0.4430443048477173 loss 0.637 = 0.543 + 0.017 + 0.077 avg prob of [ French] 0.5975548624992371 loss 0.408 = 0.294 + 0.024 + 0.09 avg prob of [ French] 0.7491843104362488 loss 0.292 = 0.161 + 0.029 + 0.102 avg prob of [ French] 0.8522353172302246 loss 0.236 = 0.103 + 0.028 + 0.105 avg prob of [ French] 0.9025701880455017 loss 0.206 = 0.074 + 0.027 + 0.105 avg prob of [ French] 0.9291102886199951 loss 0.189 = 0.057 + 0.027 + 0.105 avg prob of [ French] 0.9450439214706421 loss 0.177 = 0.046 + 0.026 + 0.105 avg prob of [ French] 0.9552142024040222 loss 0.168 = 0.038 + 0.025 + 0.105 avg prob of [ French] 0.9624165296554565 loss 0.16 = 0.033 + 0.023 + 0.105 avg prob of [ French] 0.9679695963859558 loss 0.154 = 0.028 + 0.02 + 0.105 avg prob of [ French] 0.9724625945091248 loss 0.148 = 0.024 + 0.018 + 0.105 avg prob of [ French] 0.9761754870414734 loss 0.143 = 0.021 + 0.017 + 0.105 avg prob of [ French] 0.9792689085006714 loss 0.139 = 0.018 + 0.015 + 0.105 avg prob of [ French] 0.9818602800369263 loss 0.136 = 0.016 + 0.015 + 0.105 avg prob of [ French] 0.9840443730354309 loss 0.133 = 0.014 + 0.014 + 0.105 avg prob of [ French] 0.9858965873718262 loss 0.131 = 0.013 + 0.013 + 0.105 avg prob of [ French] 0.9874734282493591 Delta norm: 75.97990417480469 Change in target norm: 18.994976043701172 to 79.08440399169922 => 60.08942794799805 Division Factor: 10.121987342834473 Right vector norm: 7.5064215660095215 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 6.939 = 6.939 + 0.0 + 0.0 avg prob of [ London] 0.001200218452140689 loss 5.87 = 5.825 + 0.025 + 0.019 avg prob of [ London] 0.003301690798252821 loss 5.008 = 4.943 + 0.034 + 0.032 avg prob of [ London] 0.007920866832137108 loss 4.211 = 4.134 + 0.034 + 0.043 avg prob of [ London] 0.017699092626571655 loss 3.268 = 3.174 + 0.04 + 0.054 avg prob of [ London] 0.04549875110387802 loss 2.173 = 2.042 + 0.067 + 0.064 avg prob of [ London] 0.1348578929901123 loss 1.223 = 1.024 + 0.124 + 0.074 avg prob of [ London] 0.36500284075737 loss 0.619 = 0.359 + 0.176 + 0.083 avg prob of [ London] 0.7003499269485474 loss 0.415 = 0.135 + 0.191 + 0.088 avg prob of [ London] 0.8739877343177795 loss 0.329 = 0.073 + 0.169 + 0.088 avg prob of [ London] 0.9299739599227905 loss 0.258 = 0.049 + 0.121 + 0.088 avg prob of [ London] 0.9520763158798218 loss 0.2 = 0.04 + 0.072 + 0.088 avg prob of [ London] 0.960747480392456 loss 0.178 = 0.036 + 0.054 + 0.088 avg prob of [ London] 0.9648504853248596 loss 0.17 = 0.032 + 0.05 + 0.088 avg prob of [ London] 0.9682455658912659 loss 0.165 = 0.029 + 0.048 + 0.088 avg prob of [ London] 0.9718848466873169 loss 0.16 = 0.025 + 0.047 + 0.088 avg prob of [ London] 0.9756993651390076 loss 0.153 = 0.021 + 0.044 + 0.088 avg prob of [ London] 0.9794144630432129 loss 0.144 = 0.017 + 0.038 + 0.088 avg prob of [ London] 0.9828144907951355 loss 0.134 = 0.014 + 0.032 + 0.088 avg prob of [ London] 0.9857956171035767 loss 0.126 = 0.012 + 0.027 + 0.088 avg prob of [ London] 0.9883391261100769 Delta norm: 90.87157440185547 Change in target norm: 22.7178955078125 to 93.15026092529297 => 70.43236541748047 Division Factor: 11.78941535949707 Right vector norm: 7.707895278930664 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 8.701 = 8.701 + 0.0 + 0.0 avg prob of [ England] 0.0001890372805064544 loss 6.548 = 6.513 + 0.001 + 0.034 avg prob of [ England] 0.0017378713237121701 loss 3.347 = 3.287 + 0.004 + 0.055 avg prob of [ England] 0.04300641268491745 loss 2.203 = 2.121 + 0.007 + 0.075 avg prob of [ England] 0.13691219687461853 loss 1.259 = 1.157 + 0.009 + 0.093 avg prob of [ England] 0.36611834168434143 loss 0.676 = 0.555 + 0.011 + 0.11 avg prob of [ England] 0.6388664245605469 loss 0.402 = 0.27 + 0.014 + 0.117 avg prob of [ England] 0.7906582951545715 loss 0.264 = 0.13 + 0.016 + 0.117 avg prob of [ England] 0.8827284574508667 loss 0.206 = 0.07 + 0.018 + 0.117 avg prob of [ England] 0.9331477284431458 loss 0.18 = 0.043 + 0.019 + 0.117 avg prob of [ England] 0.9578959345817566 loss 0.168 = 0.03 + 0.021 + 0.117 avg prob of [ England] 0.9709933400154114 loss 0.161 = 0.022 + 0.022 + 0.117 avg prob of [ England] 0.9784348011016846 loss 0.158 = 0.017 + 0.023 + 0.117 avg prob of [ England] 0.9829555749893188 loss 0.155 = 0.014 + 0.024 + 0.117 avg prob of [ England] 0.9858920574188232 loss 0.153 = 0.012 + 0.024 + 0.117 avg prob of [ England] 0.9879255890846252 loss 0.151 = 0.011 + 0.023 + 0.117 avg prob of [ England] 0.9894185066223145 loss 0.149 = 0.009 + 0.022 + 0.117 avg prob of [ England] 0.9905718564987183 loss 0.147 = 0.009 + 0.021 + 0.117 avg prob of [ England] 0.9915015697479248 loss 0.145 = 0.008 + 0.02 + 0.117 avg prob of [ England] 0.9922764301300049 loss 0.143 = 0.007 + 0.019 + 0.117 avg prob of [ England] 0.9929381608963013 Delta norm: 68.11748504638672 Change in target norm: 17.02937126159668 to 70.9162826538086 => 53.88690948486328 Division Factor: 7.354874610900879 Right vector norm: 9.261543273925781 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 16.897 = 16.897 + 0.0 + 0.0 avg prob of [ Cadillac] 1.1353050410889409e-07 loss 14.951 = 14.929 + 0.002 + 0.02 avg prob of [ Cadillac] 4.5613293764290574e-07 loss 13.688 = 13.643 + 0.009 + 0.036 avg prob of [ Cadillac] 1.4238170251701376e-06 loss 12.027 = 11.96 + 0.019 + 0.049 avg prob of [ Cadillac] 8.092575626506004e-06 loss 9.369 = 9.265 + 0.042 + 0.061 avg prob of [ Cadillac] 0.0001545895211165771 loss 7.039 = 6.871 + 0.095 + 0.072 avg prob of [ Cadillac] 0.0015206187963485718 loss 5.5 = 5.277 + 0.14 + 0.083 avg prob of [ Cadillac] 0.006725730840116739 loss 4.423 = 4.179 + 0.154 + 0.09 avg prob of [ Cadillac] 0.018821945413947105 loss 3.582 = 3.342 + 0.15 + 0.09 avg prob of [ Cadillac] 0.0425526462495327 loss 2.674 = 2.44 + 0.144 + 0.09 avg prob of [ Cadillac] 0.10334520041942596 loss 1.756 = 1.53 + 0.137 + 0.09 avg prob of [ Cadillac] 0.24566572904586792 loss 1.025 = 0.805 + 0.13 + 0.09 avg prob of [ Cadillac] 0.47471773624420166 loss 0.587 = 0.374 + 0.123 + 0.09 avg prob of [ Cadillac] 0.7017838358879089 loss 0.375 = 0.169 + 0.116 + 0.09 avg prob of [ Cadillac] 0.8491787910461426 loss 0.278 = 0.08 + 0.109 + 0.09 avg prob of [ Cadillac] 0.9246618151664734 loss 0.229 = 0.041 + 0.099 + 0.09 avg prob of [ Cadillac] 0.960334062576294 loss 0.199 = 0.023 + 0.086 + 0.09 avg prob of [ Cadillac] 0.9771348834037781 loss 0.174 = 0.015 + 0.069 + 0.09 avg prob of [ Cadillac] 0.9852064847946167 loss 0.151 = 0.011 + 0.05 + 0.09 avg prob of [ Cadillac] 0.9891106486320496 loss 0.131 = 0.009 + 0.032 + 0.09 avg prob of [ Cadillac] 0.9908658862113953 Delta norm: 89.10139465332031 Change in target norm: 22.275346755981445 to 91.00697326660156 => 68.73162841796875 Division Factor: 11.420431137084961 Right vector norm: 7.801928997039795 seq_released {"n_edits": 10, "retention": 1.0, "unedited_true": 0.78, "max_update_norm": 9.254192352294922, "median_update_norm": 7.7796196937561035} (316s) 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 5.932 = 5.932 + 0.0 + 0.0 avg prob of [ Norway] 0.002963606035336852 loss 5.27 = 5.231 + 0.008 + 0.031 avg prob of [ Norway] 0.006190676707774401 loss 4.862 = 4.803 + 0.01 + 0.049 avg prob of [ Norway] 0.009303576312959194 loss 4.313 = 4.235 + 0.014 + 0.064 avg prob of [ Norway] 0.015961652621626854 loss 3.27 = 3.17 + 0.021 + 0.079 avg prob of [ Norway] 0.045968521386384964 loss 2.113 = 1.989 + 0.033 + 0.092 avg prob of [ Norway] 0.14623330533504486 loss 1.302 = 1.15 + 0.048 + 0.105 avg prob of [ Norway] 0.33156436681747437 loss 0.571 = 0.401 + 0.058 + 0.112 avg prob of [ Norway] 0.6802366971969604 loss 0.301 = 0.125 + 0.064 + 0.112 avg prob of [ Norway] 0.8843719959259033 loss 0.222 = 0.042 + 0.068 + 0.112 avg prob of [ Norway] 0.95891934633255 loss 0.201 = 0.019 + 0.07 + 0.112 avg prob of [ Norway] 0.9814345836639404 loss 0.193 = 0.011 + 0.071 + 0.112 avg prob of [ Norway] 0.9889862537384033 loss 0.188 = 0.008 + 0.069 + 0.112 avg prob of [ Norway] 0.9920272827148438 loss 0.183 = 0.007 + 0.065 + 0.112 avg prob of [ Norway] 0.9934538006782532 loss 0.178 = 0.006 + 0.061 + 0.112 avg prob of [ Norway] 0.9941978454589844 loss 0.173 = 0.005 + 0.056 + 0.112 avg prob of [ Norway] 0.9946160316467285 loss 0.168 = 0.005 + 0.051 + 0.112 avg prob of [ Norway] 0.9948683977127075 loss 0.164 = 0.005 + 0.047 + 0.112 avg prob of [ Norway] 0.9950360655784607 loss 0.16 = 0.005 + 0.044 + 0.112 avg prob of [ Norway] 0.9951658844947815 loss 0.157 = 0.005 + 0.041 + 0.112 avg prob of [ Norway] 0.9952869415283203 Delta norm: 71.69015502929688 Change in target norm: 17.92253875732422 to 75.37071990966797 => 57.44818115234375 Division Factor: 7.492981433868408 Right vector norm: 9.567641258239746 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 7.389 = 7.389 + 0.0 + 0.0 avg prob of [ English] 0.0030759030487388372 loss 5.83 = 5.806 + 0.001 + 0.023 avg prob of [ English] 0.013082453049719334 loss 4.537 = 4.496 + 0.004 + 0.037 avg prob of [ English] 0.03996053338050842 loss 2.464 = 2.398 + 0.015 + 0.05 avg prob of [ English] 0.15466426312923431 loss 1.286 = 1.184 + 0.04 + 0.063 avg prob of [ English] 0.32250863313674927 loss 0.813 = 0.677 + 0.062 + 0.074 avg prob of [ English] 0.5139049291610718 loss 0.512 = 0.351 + 0.076 + 0.085 avg prob of [ English] 0.7054477334022522 loss 0.395 = 0.201 + 0.1 + 0.094 avg prob of [ English] 0.818639874458313 loss 0.341 = 0.146 + 0.099 + 0.096 avg prob of [ English] 0.864513099193573 loss 0.294 = 0.112 + 0.086 + 0.096 avg prob of [ English] 0.894531786441803 loss 0.257 = 0.087 + 0.075 + 0.096 avg prob of [ English] 0.917147696018219 loss 0.231 = 0.068 + 0.067 + 0.096 avg prob of [ English] 0.9345313310623169 loss 0.212 = 0.054 + 0.062 + 0.096 avg prob of [ English] 0.9478001594543457 loss 0.198 = 0.043 + 0.059 + 0.096 avg prob of [ English] 0.9577819108963013 loss 0.188 = 0.035 + 0.057 + 0.096 avg prob of [ English] 0.9652323126792908 loss 0.18 = 0.03 + 0.054 + 0.096 avg prob of [ English] 0.9708235859870911 loss 0.173 = 0.025 + 0.052 + 0.096 avg prob of [ English] 0.9750943779945374 loss 0.167 = 0.022 + 0.05 + 0.096 avg prob of [ English] 0.9784372448921204 loss 0.162 = 0.019 + 0.047 + 0.096 avg prob of [ English] 0.9811205267906189 loss 0.157 = 0.017 + 0.044 + 0.096 avg prob of [ English] 0.9833201169967651 Delta norm: 83.58563995361328 Change in target norm: 20.89640998840332 to 86.76925659179688 => 65.87284851074219 Division Factor: 9.398366928100586 Right vector norm: 8.893634796142578 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 8.208 = 8.208 + 0.0 + 0.0 avg prob of [ Poland] 0.00031664231210015714 loss 6.15 = 6.125 + 0.002 + 0.023 avg prob of [ Poland] 0.0023910794407129288 loss 4.327 = 4.282 + 0.006 + 0.039 avg prob of [ Poland] 0.016230562701821327 loss 2.237 = 2.175 + 0.01 + 0.052 avg prob of [ Poland] 0.12545596063137054 loss 1.019 = 0.942 + 0.013 + 0.064 avg prob of [ Poland] 0.39631521701812744 loss 0.568 = 0.478 + 0.015 + 0.075 avg prob of [ Poland] 0.6230621337890625 loss 0.367 = 0.264 + 0.018 + 0.086 avg prob of [ Poland] 0.7697001695632935 loss 0.259 = 0.145 + 0.02 + 0.095 avg prob of [ Poland] 0.8662896156311035 loss 0.199 = 0.083 + 0.02 + 0.096 avg prob of [ Poland] 0.9205206036567688 loss 0.166 = 0.05 + 0.02 + 0.096 avg prob of [ Poland] 0.9511739015579224 loss 0.148 = 0.033 + 0.019 + 0.096 avg prob of [ Poland] 0.9680581092834473 loss 0.138 = 0.023 + 0.019 + 0.096 avg prob of [ Poland] 0.9776212573051453 loss 0.131 = 0.017 + 0.019 + 0.096 avg prob of [ Poland] 0.9833441376686096 loss 0.127 = 0.013 + 0.018 + 0.096 avg prob of [ Poland] 0.9869818687438965 loss 0.124 = 0.011 + 0.018 + 0.096 avg prob of [ Poland] 0.9894281029701233 loss 0.122 = 0.009 + 0.017 + 0.096 avg prob of [ Poland] 0.9911553263664246 loss 0.12 = 0.008 + 0.017 + 0.096 avg prob of [ Poland] 0.9924269318580627 loss 0.118 = 0.007 + 0.016 + 0.096 avg prob of [ Poland] 0.9933960437774658 loss 0.117 = 0.006 + 0.015 + 0.096 avg prob of [ Poland] 0.9941562414169312 loss 0.116 = 0.005 + 0.015 + 0.096 avg prob of [ Poland] 0.9947671294212341 Delta norm: 83.53386688232422 Change in target norm: 20.883466720581055 to 85.85821533203125 => 64.97474670410156 Division Factor: 10.404038429260254 Right vector norm: 8.028984069824219 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 4.779 = 4.779 + 0.0 + 0.0 avg prob of [ Philadelphia] 0.008560887537896633 loss 3.785 = 3.76 + 0.003 + 0.023 avg prob of [ Philadelphia] 0.023693768307566643 loss 2.094 = 2.04 + 0.015 + 0.039 avg prob of [ Philadelphia] 0.13132135570049286 loss 1.091 = 1.005 + 0.033 + 0.053 avg prob of [ Philadelphia] 0.3675727844238281 loss 0.289 = 0.185 + 0.039 + 0.065 avg prob of [ Philadelphia] 0.8322131037712097 loss 0.134 = 0.023 + 0.036 + 0.075 avg prob of [ Philadelphia] 0.9772519469261169 loss 0.126 = 0.008 + 0.032 + 0.085 avg prob of [ Philadelphia] 0.9915673136711121 loss 0.13 = 0.005 + 0.03 + 0.094 avg prob of [ Philadelphia] 0.9948192834854126 loss 0.125 = 0.003 + 0.026 + 0.095 avg prob of [ Philadelphia] 0.9965296387672424 loss 0.121 = 0.002 + 0.024 + 0.095 avg prob of [ Philadelphia] 0.9975430369377136 loss 0.119 = 0.002 + 0.022 + 0.095 avg prob of [ Philadelphia] 0.998144268989563 loss 0.117 = 0.001 + 0.021 + 0.095 avg prob of [ Philadelphia] 0.9985153675079346 loss 0.116 = 0.001 + 0.02 + 0.095 avg prob of [ Philadelphia] 0.9987548589706421 loss 0.115 = 0.001 + 0.019 + 0.095 avg prob of [ Philadelphia] 0.998915433883667 loss 0.114 = 0.001 + 0.018 + 0.095 avg prob of [ Philadelphia] 0.9990271329879761 loss 0.112 = 0.001 + 0.016 + 0.095 avg prob of [ Philadelphia] 0.9991077184677124 loss 0.111 = 0.001 + 0.015 + 0.095 avg prob of [ Philadelphia] 0.999169111251831 loss 0.109 = 0.001 + 0.014 + 0.095 avg prob of [ Philadelphia] 0.9992187023162842 loss 0.108 = 0.001 + 0.012 + 0.095 avg prob of [ Philadelphia] 0.999261736869812 loss 0.107 = 0.001 + 0.011 + 0.095 avg prob of [ Philadelphia] 0.9993006587028503 Delta norm: 84.30183410644531 Change in target norm: 21.075458526611328 to 86.24848937988281 => 65.17303466796875 Division Factor: 10.51565933227539 Right vector norm: 8.016790390014648 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 6.614 = 6.614 + 0.0 + 0.0 avg prob of [ Queensland] 0.0024763699620962143 loss 3.493 = 3.447 + 0.008 + 0.038 avg prob of [ Queensland] 0.0334409698843956 loss 2.857 = 2.779 + 0.016 + 0.062 avg prob of [ Queensland] 0.06430638581514359 loss 2.159 = 2.046 + 0.031 + 0.082 avg prob of [ Queensland] 0.13455086946487427 loss 1.456 = 1.305 + 0.051 + 0.1 avg prob of [ Queensland] 0.28305867314338684 loss 0.852 = 0.665 + 0.071 + 0.116 avg prob of [ Queensland] 0.5283080339431763 loss 0.523 = 0.323 + 0.077 + 0.123 avg prob of [ Queensland] 0.7320772409439087 loss 0.373 = 0.174 + 0.076 + 0.123 avg prob of [ Queensland] 0.8446482419967651 loss 0.295 = 0.098 + 0.074 + 0.123 avg prob of [ Queensland] 0.9085566401481628 loss 0.254 = 0.06 + 0.072 + 0.123 avg prob of [ Queensland] 0.9432790875434875 loss 0.23 = 0.039 + 0.068 + 0.123 avg prob of [ Queensland] 0.9622089862823486 loss 0.214 = 0.028 + 0.064 + 0.123 avg prob of [ Queensland] 0.9728280901908875 loss 0.203 = 0.021 + 0.059 + 0.123 avg prob of [ Queensland] 0.9790481328964233 loss 0.195 = 0.017 + 0.055 + 0.123 avg prob of [ Queensland] 0.9828810691833496 loss 0.188 = 0.015 + 0.051 + 0.123 avg prob of [ Queensland] 0.9853732585906982 loss 0.183 = 0.013 + 0.047 + 0.123 avg prob of [ Queensland] 0.9870836138725281 loss 0.178 = 0.012 + 0.043 + 0.123 avg prob of [ Queensland] 0.9883192181587219 loss 0.173 = 0.011 + 0.04 + 0.123 avg prob of [ Queensland] 0.9892551898956299 loss 0.169 = 0.01 + 0.036 + 0.123 avg prob of [ Queensland] 0.9899947047233582 loss 0.165 = 0.009 + 0.032 + 0.123 avg prob of [ Queensland] 0.9906024932861328 Delta norm: 65.13385009765625 Change in target norm: 16.283464431762695 to 68.37554931640625 => 52.09208679199219 Division Factor: 7.198422431945801 Right vector norm: 9.048352241516113 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 14.366 = 14.366 + 0.0 + 0.0 avg prob of [ Seoul] 7.01803855918115e-07 loss 12.953 = 12.915 + 0.007 + 0.03 avg prob of [ Seoul] 3.704477649080218e-06 loss 11.62 = 11.559 + 0.01 + 0.051 avg prob of [ Seoul] 1.846041777753271e-05 loss 10.464 = 10.379 + 0.014 + 0.07 avg prob of [ Seoul] 7.867348176660016e-05 loss 9.048 = 8.941 + 0.019 + 0.088 avg prob of [ Seoul] 0.00041048997081816196 loss 7.0 = 6.869 + 0.027 + 0.103 avg prob of [ Seoul] 0.003090485231950879 loss 5.783 = 5.642 + 0.032 + 0.109 avg prob of [ Seoul] 0.007977805100381374 loss 4.812 = 4.668 + 0.034 + 0.109 avg prob of [ Seoul] 0.016948014497756958 loss 3.729 = 3.583 + 0.037 + 0.109 avg prob of [ Seoul] 0.04178855940699577 loss 2.586 = 2.436 + 0.04 + 0.109 avg prob of [ Seoul] 0.10954701155424118 loss 1.484 = 1.329 + 0.046 + 0.109 avg prob of [ Seoul] 0.2831779420375824 loss 0.796 = 0.632 + 0.054 + 0.109 avg prob of [ Seoul] 0.542876124382019 loss 0.453 = 0.279 + 0.065 + 0.109 avg prob of [ Seoul] 0.7596379518508911 loss 0.31 = 0.123 + 0.078 + 0.109 avg prob of [ Seoul] 0.8846784234046936 loss 0.244 = 0.059 + 0.076 + 0.109 avg prob of [ Seoul] 0.9430792927742004 loss 0.212 = 0.043 + 0.059 + 0.109 avg prob of [ Seoul] 0.9576199054718018 loss 0.228 = 0.067 + 0.052 + 0.109 avg prob of [ Seoul] 0.9359243512153625 loss 0.199 = 0.042 + 0.048 + 0.109 avg prob of [ Seoul] 0.9591453075408936 loss 0.169 = 0.016 + 0.044 + 0.109 avg prob of [ Seoul] 0.9845220446586609 loss 0.157 = 0.008 + 0.04 + 0.109 avg prob of [ Seoul] 0.9918550848960876 Delta norm: 73.32447052001953 Change in target norm: 18.331119537353516 to 75.37957000732422 => 57.0484504699707 Division Factor: 9.154902458190918 Right vector norm: 8.009313583374023 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 14.822 = 14.822 + 0.0 + 0.0 avg prob of [ poet] 9.447899174119812e-07 loss 14.245 = 14.217 + 0.001 + 0.027 avg prob of [ poet] 2.135311660822481e-06 loss 13.014 = 12.966 + 0.004 + 0.044 avg prob of [ poet] 7.126213404262671e-06 loss 11.47 = 11.393 + 0.018 + 0.058 avg prob of [ poet] 1.7086296793422662e-05 loss 10.244 = 10.143 + 0.029 + 0.072 avg prob of [ poet] 4.933166928822175e-05 loss 8.781 = 8.66 + 0.036 + 0.085 avg prob of [ poet] 0.00019538550986908376 loss 7.155 = 7.013 + 0.046 + 0.097 avg prob of [ poet] 0.0009435887332074344 loss 5.399 = 5.222 + 0.072 + 0.105 avg prob of [ poet] 0.005652786232531071 loss 4.551 = 4.259 + 0.187 + 0.105 avg prob of [ poet] 0.01484183594584465 loss 3.518 = 3.267 + 0.146 + 0.105 avg prob of [ poet] 0.041688982397317886 loss 2.201 = 1.962 + 0.135 + 0.105 avg prob of [ poet] 0.1485929936170578 loss 0.945 = 0.682 + 0.158 + 0.105 avg prob of [ poet] 0.5224851369857788 loss 0.406 = 0.116 + 0.185 + 0.105 avg prob of [ poet] 0.8916841149330139 loss 0.263 = 0.041 + 0.117 + 0.105 avg prob of [ poet] 0.9597083330154419 loss 0.202 = 0.029 + 0.069 + 0.105 avg prob of [ poet] 0.9719122648239136 loss 0.196 = 0.028 + 0.064 + 0.105 avg prob of [ poet] 0.9729491472244263 loss 0.204 = 0.028 + 0.071 + 0.105 avg prob of [ poet] 0.9729019403457642 loss 0.194 = 0.023 + 0.067 + 0.105 avg prob of [ poet] 0.9776711463928223 loss 0.176 = 0.016 + 0.055 + 0.105 avg prob of [ poet] 0.9839389324188232 loss 0.163 = 0.012 + 0.046 + 0.105 avg prob of [ poet] 0.988156259059906 Delta norm: 76.30059814453125 Change in target norm: 19.075149536132812 to 78.25435638427734 => 59.17920684814453 Division Factor: 10.42236614227295 Right vector norm: 7.320852279663086 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 11.844 = 11.844 + 0.0 + 0.0 avg prob of [ anthropology] 1.7001086234813556e-05 loss 10.411 = 10.378 + 0.008 + 0.026 avg prob of [ anthropology] 9.425012103747576e-05 loss 8.005 = 7.943 + 0.02 + 0.042 avg prob of [ anthropology] 0.0009452250087633729 loss 4.944 = 4.853 + 0.035 + 0.057 avg prob of [ anthropology] 0.009953240863978863 loss 2.597 = 2.466 + 0.06 + 0.071 avg prob of [ anthropology] 0.09538084268569946 loss 1.201 = 1.043 + 0.074 + 0.084 avg prob of [ anthropology] 0.3768160343170166 loss 0.529 = 0.35 + 0.083 + 0.096 avg prob of [ anthropology] 0.7161345481872559 loss 0.342 = 0.161 + 0.08 + 0.101 avg prob of [ anthropology] 0.8542668223381042 loss 0.262 = 0.089 + 0.072 + 0.101 avg prob of [ anthropology] 0.915981650352478 loss 0.216 = 0.052 + 0.064 + 0.101 avg prob of [ anthropology] 0.9500368237495422 loss 0.188 = 0.031 + 0.055 + 0.101 avg prob of [ anthropology] 0.9695257544517517 loss 0.168 = 0.019 + 0.047 + 0.101 avg prob of [ anthropology] 0.9810765385627747 loss 0.154 = 0.012 + 0.041 + 0.101 avg prob of [ anthropology] 0.9880474805831909 loss 0.147 = 0.008 + 0.038 + 0.101 avg prob of [ anthropology] 0.992205023765564 loss 0.145 = 0.005 + 0.039 + 0.101 avg prob of [ anthropology] 0.9945820569992065 loss 0.144 = 0.004 + 0.039 + 0.101 avg prob of [ anthropology] 0.9958403706550598 loss 0.142 = 0.004 + 0.038 + 0.101 avg prob of [ anthropology] 0.9964941740036011 loss 0.139 = 0.003 + 0.035 + 0.101 avg prob of [ anthropology] 0.9969239830970764 loss 0.135 = 0.003 + 0.031 + 0.101 avg prob of [ anthropology] 0.9972759485244751 loss 0.132 = 0.002 + 0.028 + 0.101 avg prob of [ anthropology] 0.9975993037223816 Delta norm: 79.13449096679688 Change in target norm: 19.78362274169922 to 82.44151306152344 => 62.65789031982422 Division Factor: 7.34244966506958 Right vector norm: 10.777668952941895 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 16.116 = 16.116 + 0.0 + 0.0 avg prob of [ goaltender] 1.583118063308575e-07 loss 14.745 = 14.728 + 0.003 + 0.014 avg prob of [ goaltender] 5.604996431429754e-07 loss 12.001 = 11.971 + 0.007 + 0.023 avg prob of [ goaltender] 1.0628341442497913e-05 loss 10.031 = 9.987 + 0.013 + 0.03 avg prob of [ goaltender] 5.9782170865219086e-05 loss 7.77 = 7.715 + 0.017 + 0.037 avg prob of [ goaltender] 0.000539060216397047 loss 5.595 = 5.529 + 0.022 + 0.044 avg prob of [ goaltender] 0.005000660661607981 loss 4.388 = 4.305 + 0.033 + 0.05 avg prob of [ goaltender] 0.015954989939928055 loss 3.023 = 2.924 + 0.043 + 0.056 avg prob of [ goaltender] 0.060665424913167953 loss 1.664 = 1.55 + 0.053 + 0.061 avg prob of [ goaltender] 0.2243887335062027 loss 0.862 = 0.733 + 0.062 + 0.067 avg prob of [ goaltender] 0.4835103452205658 loss 0.554 = 0.417 + 0.066 + 0.072 avg prob of [ goaltender] 0.660717248916626 loss 0.419 = 0.279 + 0.065 + 0.075 avg prob of [ goaltender] 0.7578457593917847 loss 0.338 = 0.202 + 0.061 + 0.075 avg prob of [ goaltender] 0.8178536295890808 loss 0.281 = 0.15 + 0.056 + 0.075 avg prob of [ goaltender] 0.8616882562637329 loss 0.24 = 0.113 + 0.052 + 0.075 avg prob of [ goaltender] 0.8932932019233704 loss 0.211 = 0.088 + 0.048 + 0.075 avg prob of [ goaltender] 0.9160075187683105 loss 0.189 = 0.07 + 0.044 + 0.075 avg prob of [ goaltender] 0.9323733448982239 loss 0.173 = 0.058 + 0.041 + 0.075 avg prob of [ goaltender] 0.9442025423049927 loss 0.162 = 0.048 + 0.038 + 0.075 avg prob of [ goaltender] 0.9527575969696045 loss 0.153 = 0.042 + 0.036 + 0.075 avg prob of [ goaltender] 0.9589239954948425 Delta norm: 106.39754486083984 Change in target norm: 26.599384307861328 to 109.3504638671875 => 82.75108337402344 Division Factor: 11.359480857849121 Right vector norm: 9.366409301757812 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 10.54 = 10.54 + 0.0 + 0.0 avg prob of [ fantasy] 5.2957893785787746e-05 loss 9.967 = 9.921 + 0.02 + 0.026 avg prob of [ fantasy] 0.00010495577589608729 loss 9.18 = 9.12 + 0.018 + 0.042 avg prob of [ fantasy] 0.0002278333849972114 loss 7.751 = 7.657 + 0.037 + 0.057 avg prob of [ fantasy] 0.001048888429068029 loss 6.024 = 5.827 + 0.125 + 0.072 avg prob of [ fantasy] 0.00844492670148611 loss 3.257 = 3.017 + 0.156 + 0.084 avg prob of [ fantasy] 0.0861474871635437 loss 0.817 = 0.539 + 0.182 + 0.096 avg prob of [ fantasy] 0.6086218953132629 loss 0.429 = 0.117 + 0.209 + 0.102 avg prob of [ fantasy] 0.8911901116371155 loss 0.344 = 0.059 + 0.182 + 0.102 avg prob of [ fantasy] 0.9428513646125793 loss 0.207 = 0.039 + 0.066 + 0.102 avg prob of [ fantasy] 0.962523877620697 loss 0.15 = 0.03 + 0.018 + 0.102 avg prob of [ fantasy] 0.970583438873291 loss 0.15 = 0.025 + 0.023 + 0.102 avg prob of [ fantasy] 0.9750663042068481 loss 0.149 = 0.022 + 0.025 + 0.102 avg prob of [ fantasy] 0.978421688079834 loss 0.146 = 0.019 + 0.025 + 0.102 avg prob of [ fantasy] 0.9813145399093628 loss 0.142 = 0.016 + 0.023 + 0.102 avg prob of [ fantasy] 0.9838903546333313 loss 0.135 = 0.014 + 0.019 + 0.102 avg prob of [ fantasy] 0.9861571788787842 loss 0.128 = 0.012 + 0.014 + 0.102 avg prob of [ fantasy] 0.9880988001823425 loss 0.124 = 0.01 + 0.011 + 0.102 avg prob of [ fantasy] 0.9897048473358154 loss 0.128 = 0.009 + 0.016 + 0.102 avg prob of [ fantasy] 0.9909929633140564 loss 0.128 = 0.008 + 0.018 + 0.102 avg prob of [ fantasy] 0.9920353889465332 Delta norm: 78.26398468017578 Change in target norm: 19.565994262695312 to 80.30353546142578 => 60.73754119873047 Division Factor: 9.951940536499023 Right vector norm: 7.864193439483643 seq_released {"n_edits": 20, "retention": 1.0, "unedited_true": 0.8, "max_update_norm": 10.766031265258789, "median_update_norm": 8.015984535217285} (594s) 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 4.723 = 4.723 + 0.0 + 0.0 avg prob of [ soccer] 0.016667446121573448 loss 1.504 = 1.472 + 0.009 + 0.023 avg prob of [ soccer] 0.32274213433265686 loss 1.023 = 0.972 + 0.014 + 0.036 avg prob of [ soccer] 0.520860493183136 loss 0.46 = 0.398 + 0.014 + 0.047 avg prob of [ soccer] 0.7666803002357483 loss 0.277 = 0.206 + 0.014 + 0.057 avg prob of [ soccer] 0.8495830297470093 loss 0.197 = 0.117 + 0.014 + 0.066 avg prob of [ soccer] 0.900708794593811 loss 0.162 = 0.074 + 0.014 + 0.073 avg prob of [ soccer] 0.9320288300514221 loss 0.147 = 0.052 + 0.015 + 0.08 avg prob of [ soccer] 0.9509032964706421 loss 0.14 = 0.038 + 0.015 + 0.087 avg prob of [ soccer] 0.9632847309112549 loss 0.137 = 0.029 + 0.016 + 0.093 avg prob of [ soccer] 0.9717714190483093 loss 0.134 = 0.023 + 0.016 + 0.095 avg prob of [ soccer] 0.9772998094558716 loss 0.13 = 0.019 + 0.016 + 0.095 avg prob of [ soccer] 0.9810778498649597 loss 0.127 = 0.016 + 0.015 + 0.095 avg prob of [ soccer] 0.9839826226234436 loss 0.124 = 0.014 + 0.015 + 0.095 avg prob of [ soccer] 0.9862712621688843 loss 0.122 = 0.012 + 0.014 + 0.095 avg prob of [ soccer] 0.988121509552002 loss 0.119 = 0.01 + 0.014 + 0.095 avg prob of [ soccer] 0.9896504282951355 loss 0.118 = 0.009 + 0.013 + 0.095 avg prob of [ soccer] 0.9909353852272034 loss 0.116 = 0.008 + 0.013 + 0.095 avg prob of [ soccer] 0.9920290112495422 loss 0.114 = 0.007 + 0.012 + 0.095 avg prob of [ soccer] 0.9929698705673218 loss 0.113 = 0.006 + 0.011 + 0.095 avg prob of [ soccer] 0.9937869310379028 Delta norm: 83.99517822265625 Change in target norm: 20.998794555664062 to 87.31066131591797 => 66.3118667602539 Division Factor: 11.23317813873291 Right vector norm: 7.477419853210449 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 8.314 = 8.314 + 0.0 + 0.0 avg prob of [ Malaysia] 0.00035233228118158877 loss 4.385 = 4.366 + 0.005 + 0.015 avg prob of [ Malaysia] 0.025288132950663567 loss 0.48 = 0.441 + 0.014 + 0.025 avg prob of [ Malaysia] 0.659459114074707 loss 0.098 = 0.045 + 0.02 + 0.034 avg prob of [ Malaysia] 0.9565187096595764 loss 0.08 = 0.015 + 0.023 + 0.042 avg prob of [ Malaysia] 0.985166072845459 loss 0.083 = 0.01 + 0.025 + 0.048 avg prob of [ Malaysia] 0.9905522465705872 loss 0.089 = 0.008 + 0.027 + 0.054 avg prob of [ Malaysia] 0.9918943047523499 loss 0.095 = 0.008 + 0.027 + 0.059 avg prob of [ Malaysia] 0.9921654462814331 loss 0.1 = 0.008 + 0.028 + 0.064 avg prob of [ Malaysia] 0.9921323657035828 loss 0.104 = 0.008 + 0.028 + 0.068 avg prob of [ Malaysia] 0.9920580983161926 loss 0.108 = 0.008 + 0.028 + 0.072 avg prob of [ Malaysia] 0.9920327067375183 loss 0.111 = 0.008 + 0.028 + 0.075 avg prob of [ Malaysia] 0.9920796155929565 loss 0.114 = 0.008 + 0.028 + 0.078 avg prob of [ Malaysia] 0.9921973347663879 loss 0.113 = 0.007 + 0.028 + 0.078 avg prob of [ Malaysia] 0.9928992986679077 loss 0.112 = 0.006 + 0.027 + 0.078 avg prob of [ Malaysia] 0.9936012029647827 loss 0.11 = 0.006 + 0.026 + 0.078 avg prob of [ Malaysia] 0.9942276477813721 loss 0.109 = 0.005 + 0.026 + 0.078 avg prob of [ Malaysia] 0.9947844743728638 loss 0.108 = 0.005 + 0.025 + 0.078 avg prob of [ Malaysia] 0.9952782988548279 loss 0.106 = 0.004 + 0.024 + 0.078 avg prob of [ Malaysia] 0.9957154393196106 loss 0.105 = 0.004 + 0.023 + 0.078 avg prob of [ Malaysia] 0.9961025714874268 Delta norm: 102.54942321777344 Change in target norm: 25.63735580444336 to 107.01641845703125 => 81.37905883789062 Division Factor: 9.42315673828125 Right vector norm: 10.88270378112793 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 13.852 = 13.852 + 0.0 + 0.0 avg prob of [ Leicester] 2.3545794647361618e-06 loss 9.434 = 9.403 + 0.01 + 0.022 avg prob of [ Leicester] 0.00010605313582345843 loss 7.485 = 7.42 + 0.028 + 0.036 avg prob of [ Leicester] 0.0007282186998054385 loss 5.822 = 5.724 + 0.049 + 0.048 avg prob of [ Leicester] 0.003730202093720436 loss 4.698 = 4.573 + 0.065 + 0.06 avg prob of [ Leicester] 0.011112923733890057 loss 3.729 = 3.584 + 0.075 + 0.07 avg prob of [ Leicester] 0.029283437877893448 loss 2.732 = 2.575 + 0.078 + 0.079 avg prob of [ Leicester] 0.07995311170816422 loss 1.785 = 1.619 + 0.078 + 0.088 avg prob of [ Leicester] 0.20595332980155945 loss 1.074 = 0.905 + 0.076 + 0.093 avg prob of [ Leicester] 0.4137507379055023 loss 0.647 = 0.482 + 0.073 + 0.093 avg prob of [ Leicester] 0.6246881484985352 loss 0.41 = 0.247 + 0.07 + 0.093 avg prob of [ Leicester] 0.7848911285400391 loss 0.292 = 0.13 + 0.069 + 0.093 avg prob of [ Leicester] 0.8796820640563965 loss 0.234 = 0.073 + 0.068 + 0.093 avg prob of [ Leicester] 0.9302250742912292 loss 0.205 = 0.044 + 0.067 + 0.093 avg prob of [ Leicester] 0.9568339586257935 loss 0.189 = 0.029 + 0.067 + 0.093 avg prob of [ Leicester] 0.9712954759597778 loss 0.18 = 0.021 + 0.067 + 0.093 avg prob of [ Leicester] 0.9795669317245483 loss 0.175 = 0.016 + 0.066 + 0.093 avg prob of [ Leicester] 0.984578549861908 loss 0.171 = 0.012 + 0.066 + 0.093 avg prob of [ Leicester] 0.9877957701683044 loss 0.169 = 0.01 + 0.065 + 0.093 avg prob of [ Leicester] 0.9899782538414001 loss 0.166 = 0.009 + 0.065 + 0.093 avg prob of [ Leicester] 0.9915357232093811 Delta norm: 85.98028564453125 Change in target norm: 21.495071411132812 to 89.05180358886719 => 67.55673217773438 Division Factor: 10.056056022644043 Right vector norm: 8.55009937286377 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 13.797 = 13.797 + 0.0 + 0.0 avg prob of [ politician] 1.4979542584114824e-06 loss 12.67 = 12.649 + 0.003 + 0.017 avg prob of [ politician] 4.546676791505888e-06 loss 10.567 = 10.502 + 0.036 + 0.029 avg prob of [ politician] 3.352490966790356e-05 loss 9.459 = 9.374 + 0.047 + 0.038 avg prob of [ politician] 9.038593270815909e-05 loss 8.099 = 7.997 + 0.056 + 0.046 avg prob of [ politician] 0.0003561117046047002 loss 6.645 = 6.53 + 0.062 + 0.053 avg prob of [ politician] 0.001538944779895246 loss 5.089 = 4.967 + 0.063 + 0.06 avg prob of [ politician] 0.007340540643781424 loss 3.644 = 3.513 + 0.064 + 0.066 avg prob of [ politician] 0.03175711631774902 loss 1.916 = 1.778 + 0.065 + 0.073 avg prob of [ politician] 0.17464369535446167 loss 0.768 = 0.627 + 0.062 + 0.079 avg prob of [ politician] 0.5386337041854858 loss 0.331 = 0.188 + 0.059 + 0.083 avg prob of [ politician] 0.8293354511260986 loss 0.215 = 0.074 + 0.057 + 0.083 avg prob of [ politician] 0.9288402795791626 loss 0.176 = 0.036 + 0.057 + 0.083 avg prob of [ politician] 0.9644331336021423 loss 0.161 = 0.022 + 0.056 + 0.083 avg prob of [ politician] 0.9782052636146545 loss 0.154 = 0.016 + 0.054 + 0.083 avg prob of [ politician] 0.9842849969863892 loss 0.149 = 0.013 + 0.053 + 0.083 avg prob of [ politician] 0.9872404932975769 loss 0.147 = 0.011 + 0.052 + 0.083 avg prob of [ politician] 0.9887570142745972 loss 0.146 = 0.01 + 0.052 + 0.083 avg prob of [ politician] 0.9895813465118408 loss 0.146 = 0.01 + 0.052 + 0.083 avg prob of [ politician] 0.9901081919670105 loss 0.145 = 0.009 + 0.052 + 0.083 avg prob of [ politician] 0.990566611289978 Delta norm: 95.9278564453125 Change in target norm: 23.981962203979492 to 99.08032989501953 => 75.0983657836914 Division Factor: 13.112835884094238 Right vector norm: 7.315568923950195 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 8.156 = 8.156 + 0.0 + 0.0 avg prob of [ Boston] 0.0004016328603029251 loss 6.514 = 6.485 + 0.002 + 0.027 avg prob of [ Boston] 0.0023312093690037727 loss 4.533 = 4.483 + 0.005 + 0.045 avg prob of [ Boston] 0.015570245683193207 loss 2.877 = 2.804 + 0.012 + 0.062 avg prob of [ Boston] 0.0673055574297905 loss 1.878 = 1.78 + 0.023 + 0.076 avg prob of [ Boston] 0.1892002820968628 loss 0.616 = 0.502 + 0.025 + 0.088 avg prob of [ Boston] 0.6269094347953796 loss 0.165 = 0.037 + 0.028 + 0.101 avg prob of [ Boston] 0.9642106890678406 loss 0.138 = 0.003 + 0.03 + 0.104 avg prob of [ Boston] 0.99675452709198 loss 0.137 = 0.001 + 0.032 + 0.104 avg prob of [ Boston] 0.9987549781799316 loss 0.138 = 0.001 + 0.032 + 0.104 avg prob of [ Boston] 0.998856246471405 loss 0.138 = 0.001 + 0.032 + 0.104 avg prob of [ Boston] 0.9986966252326965 loss 0.137 = 0.002 + 0.031 + 0.104 avg prob of [ Boston] 0.9984328150749207 loss 0.136 = 0.002 + 0.03 + 0.104 avg prob of [ Boston] 0.9981338977813721 loss 0.135 = 0.002 + 0.029 + 0.104 avg prob of [ Boston] 0.9978480339050293 loss 0.135 = 0.002 + 0.028 + 0.104 avg prob of [ Boston] 0.9976111054420471 loss 0.134 = 0.003 + 0.027 + 0.104 avg prob of [ Boston] 0.9974457025527954 loss 0.133 = 0.003 + 0.026 + 0.104 avg prob of [ Boston] 0.9973616600036621 loss 0.131 = 0.003 + 0.024 + 0.104 avg prob of [ Boston] 0.9973582029342651 loss 0.13 = 0.003 + 0.023 + 0.104 avg prob of [ Boston] 0.9974255561828613 loss 0.129 = 0.002 + 0.022 + 0.104 avg prob of [ Boston] 0.9975492358207703 Delta norm: 76.66880798339844 Change in target norm: 19.167200088500977 to 80.5097427368164 => 61.34254455566406 Division Factor: 8.687148094177246 Right vector norm: 8.825543403625488 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 13.009 = 13.009 + 0.0 + 0.0 avg prob of [ Prague] 4.678810000768863e-06 loss 10.324 = 10.287 + 0.003 + 0.034 avg prob of [ Prague] 6.904991460032761e-05 loss 7.947 = 7.876 + 0.009 + 0.062 avg prob of [ Prague] 0.0007692746585235 loss 5.28 = 5.177 + 0.018 + 0.086 avg prob of [ Prague] 0.007210295181721449 loss 3.627 = 3.495 + 0.025 + 0.107 avg prob of [ Prague] 0.03148888796567917 loss 2.579 = 2.436 + 0.027 + 0.116 avg prob of [ Prague] 0.0913630872964859 loss 1.755 = 1.613 + 0.026 + 0.116 avg prob of [ Prague] 0.20832079648971558 loss 0.969 = 0.828 + 0.025 + 0.116 avg prob of [ Prague] 0.4462737441062927 loss 0.468 = 0.328 + 0.024 + 0.116 avg prob of [ Prague] 0.7236290574073792 loss 0.274 = 0.134 + 0.024 + 0.116 avg prob of [ Prague] 0.8752986192703247 loss 0.204 = 0.065 + 0.023 + 0.116 avg prob of [ Prague] 0.9368435144424438 loss 0.175 = 0.037 + 0.022 + 0.116 avg prob of [ Prague] 0.9638376832008362 loss 0.161 = 0.023 + 0.022 + 0.116 avg prob of [ Prague] 0.977249026298523 loss 0.152 = 0.016 + 0.021 + 0.116 avg prob of [ Prague] 0.984585702419281 loss 0.147 = 0.011 + 0.02 + 0.116 avg prob of [ Prague] 0.988908052444458 loss 0.143 = 0.008 + 0.019 + 0.116 avg prob of [ Prague] 0.9916133880615234 loss 0.14 = 0.007 + 0.018 + 0.116 avg prob of [ Prague] 0.9933943748474121 loss 0.138 = 0.005 + 0.017 + 0.116 avg prob of [ Prague] 0.9946188926696777 loss 0.136 = 0.005 + 0.016 + 0.116 avg prob of [ Prague] 0.9954930543899536 loss 0.134 = 0.004 + 0.015 + 0.116 avg prob of [ Prague] 0.9961377382278442 Delta norm: 69.00421905517578 Change in target norm: 17.251052856445312 to 72.70392608642578 => 55.45287322998047 Division Factor: 7.0796966552734375 Right vector norm: 9.74677562713623 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 10.933 = 10.933 + 0.0 + 0.0 avg prob of [ Spanish] 3.651340011856519e-05 loss 10.173 = 10.114 + 0.034 + 0.025 avg prob of [ Spanish] 8.524079021299258e-05 loss 9.069 = 9.014 + 0.015 + 0.04 avg prob of [ Spanish] 0.00029106016154401004 loss 6.292 = 6.18 + 0.058 + 0.054 avg prob of [ Spanish] 0.007064725738018751 loss 2.146 = 1.984 + 0.096 + 0.066 avg prob of [ Spanish] 0.2584550678730011 loss 0.562 = 0.35 + 0.135 + 0.077 avg prob of [ Spanish] 0.7217904329299927 loss 0.396 = 0.136 + 0.171 + 0.089 avg prob of [ Spanish] 0.8734997510910034 loss 0.365 = 0.084 + 0.181 + 0.1 avg prob of [ Spanish] 0.9194576740264893 loss 0.304 = 0.046 + 0.158 + 0.1 avg prob of [ Spanish] 0.9548506140708923 loss 0.26 = 0.025 + 0.134 + 0.1 avg prob of [ Spanish] 0.9750217199325562 loss 0.237 = 0.015 + 0.122 + 0.1 avg prob of [ Spanish] 0.9852740168571472 loss 0.224 = 0.009 + 0.114 + 0.1 avg prob of [ Spanish] 0.9908406138420105 loss 0.212 = 0.006 + 0.106 + 0.1 avg prob of [ Spanish] 0.9939312934875488 loss 0.204 = 0.004 + 0.099 + 0.1 avg prob of [ Spanish] 0.9956629872322083 loss 0.2 = 0.003 + 0.096 + 0.1 avg prob of [ Spanish] 0.9966789484024048 loss 0.199 = 0.003 + 0.096 + 0.1 avg prob of [ Spanish] 0.9973164796829224 loss 0.199 = 0.002 + 0.097 + 0.1 avg prob of [ Spanish] 0.9977423548698425 loss 0.199 = 0.002 + 0.097 + 0.1 avg prob of [ Spanish] 0.9980425238609314 loss 0.198 = 0.002 + 0.096 + 0.1 avg prob of [ Spanish] 0.9982649683952332 loss 0.196 = 0.002 + 0.095 + 0.1 avg prob of [ Spanish] 0.998437762260437 Delta norm: 79.86036682128906 Change in target norm: 19.9650936126709 to 82.59099578857422 => 62.62590026855469 Division Factor: 10.259485244750977 Right vector norm: 7.78405237197876 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 7.775 = 7.775 + 0.0 + 0.0 avg prob of [ Berlin] 0.0005322839133441448 loss 7.229 = 7.211 + 0.001 + 0.017 avg prob of [ Berlin] 0.0009145914809778333 loss 6.581 = 6.548 + 0.004 + 0.029 avg prob of [ Berlin] 0.0017792497528716922 loss 5.257 = 5.198 + 0.021 + 0.038 avg prob of [ Berlin] 0.006650027818977833 loss 3.016 = 2.913 + 0.057 + 0.046 avg prob of [ Berlin] 0.06741897016763687 loss 0.607 = 0.443 + 0.111 + 0.053 avg prob of [ Berlin] 0.6904650926589966 loss 0.259 = 0.06 + 0.14 + 0.059 avg prob of [ Berlin] 0.9442043304443359 loss 0.202 = 0.018 + 0.119 + 0.066 avg prob of [ Berlin] 0.9830402135848999 loss 0.181 = 0.008 + 0.101 + 0.072 avg prob of [ Berlin] 0.9921395182609558 loss 0.164 = 0.005 + 0.081 + 0.078 avg prob of [ Berlin] 0.9950941205024719 loss 0.151 = 0.004 + 0.063 + 0.084 avg prob of [ Berlin] 0.9964278936386108 loss 0.138 = 0.003 + 0.052 + 0.084 avg prob of [ Berlin] 0.9974976778030396 loss 0.133 = 0.002 + 0.048 + 0.084 avg prob of [ Berlin] 0.9980404376983643 loss 0.128 = 0.002 + 0.043 + 0.084 avg prob of [ Berlin] 0.9983050227165222 loss 0.119 = 0.002 + 0.034 + 0.084 avg prob of [ Berlin] 0.9984474182128906 loss 0.109 = 0.001 + 0.024 + 0.084 avg prob of [ Berlin] 0.9985802173614502 loss 0.106 = 0.001 + 0.021 + 0.084 avg prob of [ Berlin] 0.9987545013427734 loss 0.107 = 0.001 + 0.023 + 0.084 avg prob of [ Berlin] 0.9989365339279175 loss 0.108 = 0.001 + 0.023 + 0.084 avg prob of [ Berlin] 0.9990927577018738 loss 0.106 = 0.001 + 0.021 + 0.084 avg prob of [ Berlin] 0.9992192387580872 Delta norm: 95.78473663330078 Change in target norm: 23.946184158325195 to 98.92008972167969 => 74.97390747070312 Division Factor: 12.65417766571045 Right vector norm: 7.569416046142578 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 5.382 = 5.382 + 0.0 + 0.0 avg prob of [ quarterback] 0.008063904941082 loss 4.396 = 4.376 + 0.005 + 0.015 avg prob of [ quarterback] 0.01967504993081093 loss 3.886 = 3.857 + 0.005 + 0.024 avg prob of [ quarterback] 0.032925836741924286 loss 2.965 = 2.882 + 0.052 + 0.031 avg prob of [ quarterback] 0.08082684129476547 loss 1.586 = 1.482 + 0.067 + 0.037 avg prob of [ quarterback] 0.2451331913471222 loss 0.901 = 0.783 + 0.075 + 0.043 avg prob of [ quarterback] 0.464819997549057 loss 0.612 = 0.478 + 0.084 + 0.049 avg prob of [ quarterback] 0.6262181401252747 loss 0.447 = 0.298 + 0.094 + 0.055 avg prob of [ quarterback] 0.7459214329719543 loss 0.346 = 0.184 + 0.102 + 0.06 avg prob of [ quarterback] 0.8333715796470642 loss 0.291 = 0.119 + 0.107 + 0.065 avg prob of [ quarterback] 0.8888121247291565 loss 0.26 = 0.082 + 0.108 + 0.069 avg prob of [ quarterback] 0.9215903282165527 loss 0.241 = 0.062 + 0.106 + 0.074 avg prob of [ quarterback] 0.9404288530349731 loss 0.228 = 0.05 + 0.101 + 0.076 avg prob of [ quarterback] 0.9514951109886169 loss 0.213 = 0.042 + 0.095 + 0.076 avg prob of [ quarterback] 0.9590171575546265 loss 0.202 = 0.036 + 0.09 + 0.076 avg prob of [ quarterback] 0.9647624492645264 loss 0.193 = 0.031 + 0.085 + 0.076 avg prob of [ quarterback] 0.9693616032600403 loss 0.185 = 0.027 + 0.081 + 0.076 avg prob of [ quarterback] 0.9731587767601013 loss 0.178 = 0.024 + 0.078 + 0.076 avg prob of [ quarterback] 0.976351261138916 loss 0.172 = 0.021 + 0.075 + 0.076 avg prob of [ quarterback] 0.9790652394294739 loss 0.168 = 0.019 + 0.073 + 0.076 avg prob of [ quarterback] 0.9813892841339111 Delta norm: 104.62002563476562 Change in target norm: 26.15500831604004 to 109.47386169433594 => 83.31885528564453 Division Factor: 13.095809936523438 Right vector norm: 7.98881721496582 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 8.65 = 8.65 + 0.0 + 0.0 avg prob of [ Asia] 0.00024065164325293154 loss 6.403 = 6.377 + 0.002 + 0.024 avg prob of [ Asia] 0.0021829893812537193 loss 4.431 = 4.385 + 0.005 + 0.041 avg prob of [ Asia] 0.016069743782281876 loss 2.318 = 2.251 + 0.011 + 0.056 avg prob of [ Asia] 0.1270274966955185 loss 0.949 = 0.86 + 0.018 + 0.071 avg prob of [ Asia] 0.4559960663318634 loss 0.426 = 0.316 + 0.026 + 0.085 avg prob of [ Asia] 0.7407849431037903 loss 0.261 = 0.132 + 0.032 + 0.097 avg prob of [ Asia] 0.878284752368927 loss 0.208 = 0.079 + 0.031 + 0.098 avg prob of [ Asia] 0.9242375493049622 loss 0.179 = 0.052 + 0.029 + 0.098 avg prob of [ Asia] 0.9491774439811707 loss 0.162 = 0.037 + 0.027 + 0.098 avg prob of [ Asia] 0.9638200402259827 loss 0.15 = 0.028 + 0.025 + 0.098 avg prob of [ Asia] 0.9729291200637817 loss 0.142 = 0.021 + 0.023 + 0.098 avg prob of [ Asia] 0.9789078235626221 loss 0.137 = 0.017 + 0.022 + 0.098 avg prob of [ Asia] 0.9830082654953003 loss 0.133 = 0.014 + 0.021 + 0.098 avg prob of [ Asia] 0.9859241247177124 loss 0.13 = 0.012 + 0.02 + 0.098 avg prob of [ Asia] 0.988064169883728 loss 0.127 = 0.01 + 0.019 + 0.098 avg prob of [ Asia] 0.9896796345710754 loss 0.125 = 0.009 + 0.018 + 0.098 avg prob of [ Asia] 0.9909313917160034 loss 0.123 = 0.008 + 0.017 + 0.098 avg prob of [ Asia] 0.9919244050979614 loss 0.122 = 0.007 + 0.017 + 0.098 avg prob of [ Asia] 0.9927297234535217 loss 0.121 = 0.007 + 0.016 + 0.098 avg prob of [ Asia] 0.9933956861495972 Delta norm: 81.64818572998047 Change in target norm: 20.412046432495117 to 83.13941955566406 => 62.72737121582031 Division Factor: 7.756814956665039 Right vector norm: 10.525994300842285 seq_released {"n_edits": 30, "retention": 1.0, "unedited_true": 0.81, "max_update_norm": 10.874397277832031, "median_update_norm": 8.015984535217285} (868s) 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 5.905 = 5.905 + 0.0 + 0.0 avg prob of [ Austria] 0.0031034734565764666 loss 4.24 = 4.215 + 0.002 + 0.024 avg prob of [ Austria] 0.01721792295575142 loss 3.181 = 3.14 + 0.003 + 0.038 avg prob of [ Austria] 0.04636726155877113 loss 2.779 = 2.724 + 0.005 + 0.05 avg prob of [ Austria] 0.06968716531991959 loss 2.306 = 2.238 + 0.007 + 0.061 avg prob of [ Austria] 0.11327019333839417 loss 1.687 = 1.605 + 0.01 + 0.071 avg prob of [ Austria] 0.21225111186504364 loss 1.038 = 0.943 + 0.014 + 0.081 avg prob of [ Austria] 0.40217891335487366 loss 0.569 = 0.463 + 0.017 + 0.09 avg prob of [ Austria] 0.6347709894180298 loss 0.327 = 0.211 + 0.018 + 0.098 avg prob of [ Austria] 0.8114914894104004 loss 0.217 = 0.1 + 0.018 + 0.098 avg prob of [ Austria] 0.905227541923523 loss 0.163 = 0.047 + 0.018 + 0.098 avg prob of [ Austria] 0.9540915489196777 loss 0.14 = 0.024 + 0.018 + 0.098 avg prob of [ Austria] 0.9765477776527405 loss 0.129 = 0.013 + 0.018 + 0.098 avg prob of [ Austria] 0.9868234992027283 loss 0.124 = 0.008 + 0.017 + 0.098 avg prob of [ Austria] 0.9918254613876343 loss 0.12 = 0.006 + 0.017 + 0.098 avg prob of [ Austria] 0.9944621324539185 loss 0.118 = 0.004 + 0.016 + 0.098 avg prob of [ Austria] 0.9959616661071777 loss 0.117 = 0.003 + 0.016 + 0.098 avg prob of [ Austria] 0.9968735575675964 loss 0.115 = 0.003 + 0.015 + 0.098 avg prob of [ Austria] 0.9974608421325684 loss 0.114 = 0.002 + 0.014 + 0.098 avg prob of [ Austria] 0.9978575706481934 loss 0.113 = 0.002 + 0.014 + 0.098 avg prob of [ Austria] 0.9981368780136108 Delta norm: 81.7357177734375 Change in target norm: 20.433929443359375 to 86.1610107421875 => 65.72708129882812 Division Factor: 8.76500129699707 Right vector norm: 9.325237274169922 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 5.282 = 5.282 + 0.0 + 0.0 avg prob of [ Dutch] 0.005435590166598558 loss 2.985 = 2.962 + 0.002 + 0.02 avg prob of [ Dutch] 0.05857866257429123 loss 1.726 = 1.691 + 0.003 + 0.032 avg prob of [ Dutch] 0.2055658996105194 loss 0.737 = 0.69 + 0.004 + 0.043 avg prob of [ Dutch] 0.5300022959709167 loss 0.388 = 0.33 + 0.006 + 0.052 avg prob of [ Dutch] 0.7332921624183655 loss 0.266 = 0.198 + 0.007 + 0.061 avg prob of [ Dutch] 0.8270195722579956 loss 0.214 = 0.135 + 0.009 + 0.07 avg prob of [ Dutch] 0.8761937022209167 loss 0.189 = 0.1 + 0.012 + 0.077 avg prob of [ Dutch] 0.9058892130851746 loss 0.175 = 0.078 + 0.014 + 0.084 avg prob of [ Dutch] 0.925298810005188 loss 0.168 = 0.064 + 0.014 + 0.09 avg prob of [ Dutch] 0.9383988380432129 loss 0.157 = 0.053 + 0.015 + 0.09 avg prob of [ Dutch] 0.9490289092063904 loss 0.148 = 0.043 + 0.014 + 0.09 avg prob of [ Dutch] 0.9576384425163269 loss 0.14 = 0.036 + 0.014 + 0.09 avg prob of [ Dutch] 0.9645032286643982 loss 0.134 = 0.031 + 0.013 + 0.09 avg prob of [ Dutch] 0.9699797630310059 loss 0.128 = 0.026 + 0.012 + 0.09 avg prob of [ Dutch] 0.9743736982345581 loss 0.124 = 0.022 + 0.011 + 0.09 avg prob of [ Dutch] 0.9779304265975952 loss 0.12 = 0.019 + 0.01 + 0.09 avg prob of [ Dutch] 0.9808396100997925 loss 0.117 = 0.017 + 0.009 + 0.09 avg prob of [ Dutch] 0.9832434058189392 loss 0.114 = 0.015 + 0.009 + 0.09 avg prob of [ Dutch] 0.9852482080459595 loss 0.112 = 0.013 + 0.008 + 0.09 avg prob of [ Dutch] 0.9869341850280762 Delta norm: 88.65489959716797 Change in target norm: 22.16372299194336 to 92.11139678955078 => 69.94767761230469 Division Factor: 8.683645248413086 Right vector norm: 10.209409713745117 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 8.906 = 8.906 + 0.0 + 0.0 avg prob of [ French] 0.00014543018187396228 loss 6.589 = 6.559 + 0.01 + 0.019 avg prob of [ French] 0.0015507963253185153 loss 4.979 = 4.932 + 0.015 + 0.033 avg prob of [ French] 0.008052059449255466 loss 2.819 = 2.756 + 0.02 + 0.043 avg prob of [ French] 0.07247310876846313 loss 1.687 = 1.613 + 0.021 + 0.053 avg prob of [ French] 0.21650448441505432 loss 1.157 = 1.072 + 0.024 + 0.062 avg prob of [ French] 0.3556235730648041 loss 0.855 = 0.759 + 0.026 + 0.07 avg prob of [ French] 0.4761631488800049 loss 0.64 = 0.534 + 0.027 + 0.078 avg prob of [ French] 0.5899430513381958 loss 0.477 = 0.363 + 0.028 + 0.085 avg prob of [ French] 0.6975107789039612 loss 0.345 = 0.229 + 0.028 + 0.088 avg prob of [ French] 0.7960925102233887 loss 0.252 = 0.137 + 0.027 + 0.088 avg prob of [ French] 0.8723633885383606 loss 0.196 = 0.082 + 0.027 + 0.088 avg prob of [ French] 0.9218688011169434 loss 0.164 = 0.05 + 0.026 + 0.088 avg prob of [ French] 0.9511839747428894 loss 0.146 = 0.032 + 0.026 + 0.088 avg prob of [ French] 0.9681422710418701 loss 0.136 = 0.022 + 0.026 + 0.088 avg prob of [ French] 0.978101909160614 loss 0.13 = 0.016 + 0.026 + 0.088 avg prob of [ French] 0.9841561317443848 loss 0.126 = 0.012 + 0.027 + 0.088 avg prob of [ French] 0.9880012273788452 loss 0.124 = 0.009 + 0.027 + 0.088 avg prob of [ French] 0.9905610680580139 loss 0.123 = 0.008 + 0.027 + 0.088 avg prob of [ French] 0.9923466444015503 loss 0.122 = 0.006 + 0.028 + 0.088 avg prob of [ French] 0.9936467409133911 Delta norm: 91.37677764892578 Change in target norm: 22.844194412231445 to 95.07585144042969 => 72.23165893554688 Division Factor: 10.15747356414795 Right vector norm: 8.996014595031738 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 10.665 = 10.665 + 0.0 + 0.0 avg prob of [ Sydney] 3.373892468516715e-05 loss 7.317 = 7.283 + 0.005 + 0.029 avg prob of [ Sydney] 0.0008137390250340104 loss 5.319 = 5.259 + 0.011 + 0.049 avg prob of [ Sydney] 0.00694240489974618 loss 3.174 = 3.089 + 0.018 + 0.067 avg prob of [ Sydney] 0.05960610508918762 loss 1.829 = 1.716 + 0.03 + 0.083 avg prob of [ Sydney] 0.1979583501815796 loss 1.107 = 0.969 + 0.04 + 0.099 avg prob of [ Sydney] 0.3962697386741638 loss 0.61 = 0.459 + 0.044 + 0.108 avg prob of [ Sydney] 0.6438463926315308 loss 0.37 = 0.218 + 0.044 + 0.108 avg prob of [ Sydney] 0.8090858459472656 loss 0.26 = 0.108 + 0.043 + 0.108 avg prob of [ Sydney] 0.898813784122467 loss 0.21 = 0.06 + 0.042 + 0.108 avg prob of [ Sydney] 0.9425374865531921 loss 0.184 = 0.036 + 0.04 + 0.108 avg prob of [ Sydney] 0.964259922504425 loss 0.17 = 0.025 + 0.037 + 0.108 avg prob of [ Sydney] 0.9758272767066956 loss 0.16 = 0.018 + 0.034 + 0.108 avg prob of [ Sydney] 0.9825595617294312 loss 0.153 = 0.013 + 0.032 + 0.108 avg prob of [ Sydney] 0.986785888671875 loss 0.148 = 0.01 + 0.029 + 0.108 avg prob of [ Sydney] 0.9895920753479004 loss 0.144 = 0.009 + 0.027 + 0.108 avg prob of [ Sydney] 0.9915362000465393 loss 0.141 = 0.007 + 0.026 + 0.108 avg prob of [ Sydney] 0.9929322004318237 loss 0.138 = 0.006 + 0.024 + 0.108 avg prob of [ Sydney] 0.9939682483673096 loss 0.136 = 0.005 + 0.023 + 0.108 avg prob of [ Sydney] 0.9947625398635864 loss 0.134 = 0.005 + 0.021 + 0.108 avg prob of [ Sydney] 0.9953906536102295 Delta norm: 74.00859832763672 Change in target norm: 18.502151489257812 to 77.6585922241211 => 59.15644073486328 Division Factor: 7.50040340423584 Right vector norm: 9.86728286743164 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 8.865 = 8.865 + 0.0 + 0.0 avg prob of [ Minneapolis] 0.00016050992417149246 loss 6.927 = 6.894 + 0.008 + 0.025 avg prob of [ Minneapolis] 0.0012620938941836357 loss 5.321 = 5.257 + 0.023 + 0.04 avg prob of [ Minneapolis] 0.006084189750254154 loss 3.358 = 3.284 + 0.021 + 0.054 avg prob of [ Minneapolis] 0.03860463574528694 loss 1.358 = 1.26 + 0.032 + 0.067 avg prob of [ Minneapolis] 0.2889515459537506 loss 0.485 = 0.362 + 0.044 + 0.079 avg prob of [ Minneapolis] 0.7017151713371277 loss 0.281 = 0.139 + 0.051 + 0.091 avg prob of [ Minneapolis] 0.8722919225692749 loss 0.217 = 0.063 + 0.053 + 0.101 avg prob of [ Minneapolis] 0.9391428828239441 loss 0.185 = 0.031 + 0.054 + 0.101 avg prob of [ Minneapolis] 0.9700856804847717 loss 0.171 = 0.016 + 0.054 + 0.101 avg prob of [ Minneapolis] 0.9838511943817139 loss 0.163 = 0.01 + 0.052 + 0.101 avg prob of [ Minneapolis] 0.9902235865592957 loss 0.157 = 0.007 + 0.05 + 0.101 avg prob of [ Minneapolis] 0.9934185743331909 loss 0.152 = 0.005 + 0.046 + 0.101 avg prob of [ Minneapolis] 0.9951580762863159 loss 0.147 = 0.004 + 0.042 + 0.101 avg prob of [ Minneapolis] 0.9961777925491333 loss 0.143 = 0.003 + 0.039 + 0.101 avg prob of [ Minneapolis] 0.9968146085739136 loss 0.139 = 0.003 + 0.036 + 0.101 avg prob of [ Minneapolis] 0.9972348809242249 loss 0.137 = 0.002 + 0.034 + 0.101 avg prob of [ Minneapolis] 0.9975259900093079 loss 0.134 = 0.002 + 0.032 + 0.101 avg prob of [ Minneapolis] 0.9977373480796814 loss 0.133 = 0.002 + 0.03 + 0.101 avg prob of [ Minneapolis] 0.9978975653648376 loss 0.131 = 0.002 + 0.029 + 0.101 avg prob of [ Minneapolis] 0.9980245232582092 Delta norm: 79.49658203125 Change in target norm: 19.8741455078125 to 82.75811004638672 => 62.88396453857422 Division Factor: 9.177937507629395 Right vector norm: 8.661704063415527 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 10.976 = 10.976 + 0.0 + 0.0 avg prob of [ cardinal] 2.17902816075366e-05 loss 10.498 = 10.47 + 0.01 + 0.019 avg prob of [ cardinal] 3.846253821393475e-05 loss 9.994 = 9.949 + 0.015 + 0.031 avg prob of [ cardinal] 6.746770668542013e-05 loss 7.989 = 7.933 + 0.014 + 0.041 avg prob of [ cardinal] 0.00047515874030068517 loss 5.669 = 5.596 + 0.023 + 0.05 avg prob of [ cardinal] 0.004322332330048084 loss 4.201 = 4.112 + 0.03 + 0.058 avg prob of [ cardinal] 0.018341559916734695 loss 2.962 = 2.855 + 0.04 + 0.067 avg prob of [ cardinal] 0.06208644434809685 loss 1.915 = 1.79 + 0.05 + 0.075 avg prob of [ cardinal] 0.17738260328769684 loss 0.897 = 0.751 + 0.063 + 0.083 avg prob of [ cardinal] 0.4867148697376251 loss 0.361 = 0.153 + 0.121 + 0.087 avg prob of [ cardinal] 0.8606131672859192 loss 0.297 = 0.052 + 0.158 + 0.087 avg prob of [ cardinal] 0.9494658708572388 loss 0.178 = 0.022 + 0.069 + 0.087 avg prob of [ cardinal] 0.9780701398849487 loss 0.146 = 0.014 + 0.044 + 0.087 avg prob of [ cardinal] 0.9862135648727417 loss 0.139 = 0.011 + 0.04 + 0.087 avg prob of [ cardinal] 0.9889686107635498 loss 0.137 = 0.01 + 0.039 + 0.087 avg prob of [ cardinal] 0.9899923205375671 loss 0.135 = 0.009 + 0.039 + 0.087 avg prob of [ cardinal] 0.9906775951385498 loss 0.134 = 0.009 + 0.037 + 0.087 avg prob of [ cardinal] 0.9914037585258484 loss 0.131 = 0.008 + 0.036 + 0.087 avg prob of [ cardinal] 0.992220938205719 loss 0.129 = 0.007 + 0.035 + 0.087 avg prob of [ cardinal] 0.9931089282035828 loss 0.126 = 0.006 + 0.033 + 0.087 avg prob of [ cardinal] 0.9940266609191895 Delta norm: 91.464599609375 Change in target norm: 22.866147994995117 to 95.44527435302734 => 72.5791244506836 Division Factor: 9.570873260498047 Right vector norm: 9.556556701660156 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 16.462 = 16.462 + 0.0 + 0.0 avg prob of [ Idaho] 8.456085964780868e-08 loss 13.023 = 13.008 + 0.002 + 0.013 avg prob of [ Idaho] 2.6202103526884457e-06 loss 10.183 = 10.156 + 0.006 + 0.022 avg prob of [ Idaho] 4.995512426830828e-05 loss 7.882 = 7.837 + 0.016 + 0.029 avg prob of [ Idaho] 0.0005021520191803575 loss 6.113 = 6.037 + 0.041 + 0.036 avg prob of [ Idaho] 0.002865681890398264 loss 4.169 = 4.049 + 0.079 + 0.041 avg prob of [ Idaho] 0.01940777525305748 loss 1.475 = 1.343 + 0.086 + 0.046 avg prob of [ Idaho] 0.3080064356327057 loss 0.461 = 0.337 + 0.073 + 0.051 avg prob of [ Idaho] 0.7450700402259827 loss 0.151 = 0.025 + 0.071 + 0.055 avg prob of [ Idaho] 0.9755574464797974 loss 0.132 = 0.004 + 0.069 + 0.059 avg prob of [ Idaho] 0.9962748289108276 loss 0.134 = 0.002 + 0.069 + 0.063 avg prob of [ Idaho] 0.9978690147399902 loss 0.13 = 0.002 + 0.062 + 0.066 avg prob of [ Idaho] 0.9978945851325989 loss 0.121 = 0.002 + 0.05 + 0.069 avg prob of [ Idaho] 0.997586190700531 loss 0.113 = 0.003 + 0.038 + 0.071 avg prob of [ Idaho] 0.9970959424972534 loss 0.106 = 0.004 + 0.031 + 0.071 avg prob of [ Idaho] 0.9963776469230652 loss 0.104 = 0.004 + 0.028 + 0.071 avg prob of [ Idaho] 0.9957294464111328 loss 0.103 = 0.005 + 0.027 + 0.071 avg prob of [ Idaho] 0.9952815771102905 loss 0.102 = 0.005 + 0.026 + 0.071 avg prob of [ Idaho] 0.9950766563415527 loss 0.102 = 0.005 + 0.025 + 0.071 avg prob of [ Idaho] 0.995071530342102 loss 0.101 = 0.005 + 0.025 + 0.071 avg prob of [ Idaho] 0.9951876401901245 Delta norm: 112.10794067382812 Change in target norm: 28.026981353759766 to 116.3491439819336 => 88.32215881347656 Division Factor: 11.328210830688477 Right vector norm: 9.89634895324707 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 10.323 = 10.323 + 0.0 + 0.0 avg prob of [ baseball] 4.678437835536897e-05 loss 8.632 = 8.614 + 0.003 + 0.015 avg prob of [ baseball] 0.0002548349439166486 loss 5.704 = 5.67 + 0.008 + 0.026 avg prob of [ baseball] 0.004536791238933802 loss 2.771 = 2.724 + 0.013 + 0.035 avg prob of [ baseball] 0.07952576875686646 loss 1.387 = 1.324 + 0.02 + 0.043 avg prob of [ baseball] 0.29915890097618103 loss 0.405 = 0.329 + 0.026 + 0.05 avg prob of [ baseball] 0.7348836064338684 loss 0.17 = 0.084 + 0.029 + 0.056 avg prob of [ baseball] 0.9203552007675171 loss 0.141 = 0.044 + 0.034 + 0.063 avg prob of [ baseball] 0.9570364952087402 loss 0.14 = 0.034 + 0.038 + 0.068 avg prob of [ baseball] 0.9672843217849731 loss 0.144 = 0.028 + 0.042 + 0.073 avg prob of [ baseball] 0.9721528887748718 loss 0.146 = 0.024 + 0.044 + 0.078 avg prob of [ baseball] 0.9765635132789612 loss 0.136 = 0.019 + 0.04 + 0.078 avg prob of [ baseball] 0.9814903736114502 loss 0.128 = 0.014 + 0.036 + 0.078 avg prob of [ baseball] 0.9861468076705933 loss 0.121 = 0.01 + 0.033 + 0.078 avg prob of [ baseball] 0.989827036857605 loss 0.116 = 0.008 + 0.031 + 0.078 avg prob of [ baseball] 0.9924246668815613 loss 0.113 = 0.006 + 0.029 + 0.078 avg prob of [ baseball] 0.9941706657409668 loss 0.11 = 0.005 + 0.028 + 0.078 avg prob of [ baseball] 0.995336651802063 loss 0.108 = 0.004 + 0.026 + 0.078 avg prob of [ baseball] 0.9961279034614563 loss 0.106 = 0.003 + 0.024 + 0.078 avg prob of [ baseball] 0.9966791272163391 loss 0.103 = 0.003 + 0.023 + 0.078 avg prob of [ baseball] 0.9970743060112 Delta norm: 102.87789916992188 Change in target norm: 25.71947479248047 to 107.58776092529297 => 81.8682861328125 Division Factor: 11.037250518798828 Right vector norm: 9.320971488952637 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 8.52 = 8.52 + 0.0 + 0.0 avg prob of [ linebacker] 0.00029923944384790957 loss 8.376 = 8.163 + 0.197 + 0.016 avg prob of [ linebacker] 0.00042520827264524996 loss 8.12 = 8.085 + 0.012 + 0.023 avg prob of [ linebacker] 0.00045689434045925736 loss 7.969 = 7.933 + 0.007 + 0.029 avg prob of [ linebacker] 0.0005198147264309227 loss 7.593 = 7.55 + 0.007 + 0.036 avg prob of [ linebacker] 0.0007317524868994951 loss 6.562 = 6.511 + 0.01 + 0.041 avg prob of [ linebacker] 0.0019800972659140825 loss 5.201 = 5.136 + 0.018 + 0.047 avg prob of [ linebacker] 0.007318190298974514 loss 4.556 = 4.464 + 0.039 + 0.053 avg prob of [ linebacker] 0.013077129609882832 loss 4.285 = 4.161 + 0.065 + 0.059 avg prob of [ linebacker] 0.016629908233880997 loss 3.927 = 3.78 + 0.082 + 0.065 avg prob of [ linebacker] 0.02420816756784916 loss 3.383 = 3.226 + 0.086 + 0.071 avg prob of [ linebacker] 0.04215541481971741 loss 2.731 = 2.577 + 0.077 + 0.077 avg prob of [ linebacker] 0.07987438887357712 loss 2.141 = 1.996 + 0.066 + 0.079 avg prob of [ linebacker] 0.14066269993782043 loss 1.644 = 1.509 + 0.057 + 0.079 avg prob of [ linebacker] 0.22708450257778168 loss 1.177 = 1.048 + 0.05 + 0.079 avg prob of [ linebacker] 0.357351154088974 loss 0.759 = 0.635 + 0.045 + 0.079 avg prob of [ linebacker] 0.5356730818748474 loss 0.47 = 0.349 + 0.042 + 0.079 avg prob of [ linebacker] 0.7085986137390137 loss 0.309 = 0.191 + 0.039 + 0.079 avg prob of [ linebacker] 0.8275208473205566 loss 0.223 = 0.109 + 0.036 + 0.079 avg prob of [ linebacker] 0.8980052471160889 loss 0.176 = 0.065 + 0.033 + 0.079 avg prob of [ linebacker] 0.9377838969230652 Delta norm: 101.4195327758789 Change in target norm: 25.354881286621094 to 104.5794677734375 => 79.2245864868164 Division Factor: 11.886787414550781 Right vector norm: 8.532122611999512 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 9.548 = 9.548 + 0.0 + 0.0 avg prob of [ Italian] 8.462194819003344e-05 loss 4.861 = 4.84 + 0.012 + 0.01 avg prob of [ Italian] 0.008870101533830166 loss 1.863 = 1.825 + 0.023 + 0.016 avg prob of [ Italian] 0.17681336402893066 loss 0.97 = 0.913 + 0.036 + 0.021 avg prob of [ Italian] 0.4161852300167084 loss 0.653 = 0.583 + 0.044 + 0.026 avg prob of [ Italian] 0.5682665705680847 loss 0.426 = 0.353 + 0.043 + 0.03 avg prob of [ Italian] 0.7073866724967957 loss 0.289 = 0.217 + 0.038 + 0.034 avg prob of [ Italian] 0.8073043823242188 loss 0.213 = 0.141 + 0.034 + 0.037 avg prob of [ Italian] 0.8696913719177246 loss 0.169 = 0.097 + 0.032 + 0.041 avg prob of [ Italian] 0.9085404276847839 loss 0.143 = 0.069 + 0.031 + 0.043 avg prob of [ Italian] 0.9332762956619263 loss 0.128 = 0.052 + 0.03 + 0.046 avg prob of [ Italian] 0.9493966102600098 loss 0.119 = 0.041 + 0.03 + 0.048 avg prob of [ Italian] 0.9601617455482483 loss 0.113 = 0.033 + 0.03 + 0.05 avg prob of [ Italian] 0.9675483703613281 loss 0.11 = 0.028 + 0.03 + 0.052 avg prob of [ Italian] 0.9727728962898254 loss 0.108 = 0.024 + 0.03 + 0.054 avg prob of [ Italian] 0.9765942096710205 loss 0.106 = 0.021 + 0.03 + 0.056 avg prob of [ Italian] 0.9794861674308777 loss 0.105 = 0.018 + 0.03 + 0.057 avg prob of [ Italian] 0.9817466735839844 loss 0.105 = 0.017 + 0.03 + 0.059 avg prob of [ Italian] 0.9835658073425293 loss 0.104 = 0.015 + 0.029 + 0.06 avg prob of [ Italian] 0.9850677251815796 loss 0.104 = 0.014 + 0.029 + 0.061 avg prob of [ Italian] 0.9863343834877014 Delta norm: 127.32780456542969 Change in target norm: 32.351341247558594 to 131.9698944091797 => 99.6185531616211 Division Factor: 10.029826164245605 Right vector norm: 12.694917678833008 seq_released {"n_edits": 40, "retention": 1.0, "unedited_true": 0.82, "max_update_norm": 12.684980392456055, "median_update_norm": 8.625029563903809} (1154s) 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 12.231 = 12.231 + 0.0 + 0.0 avg prob of [ jazz] 6.426288109651068e-06 loss 9.467 = 9.438 + 0.01 + 0.018 avg prob of [ jazz] 9.592840069672093e-05 loss 8.013 = 7.965 + 0.019 + 0.03 avg prob of [ jazz] 0.0003912821994163096 loss 6.481 = 6.411 + 0.029 + 0.041 avg prob of [ jazz] 0.0018494079122319818 loss 4.729 = 4.646 + 0.032 + 0.051 avg prob of [ jazz] 0.010751467198133469 loss 3.183 = 3.088 + 0.035 + 0.06 avg prob of [ jazz] 0.04869040846824646 loss 2.037 = 1.927 + 0.042 + 0.068 avg prob of [ jazz] 0.15003931522369385 loss 1.117 = 0.989 + 0.051 + 0.077 avg prob of [ jazz] 0.37605342268943787 loss 0.558 = 0.414 + 0.059 + 0.085 avg prob of [ jazz] 0.6626328229904175 loss 0.34 = 0.203 + 0.051 + 0.086 avg prob of [ jazz] 0.8167018890380859 loss 0.241 = 0.115 + 0.04 + 0.086 avg prob of [ jazz] 0.8914214372634888 loss 0.192 = 0.075 + 0.031 + 0.086 avg prob of [ jazz] 0.9278569221496582 loss 0.165 = 0.054 + 0.025 + 0.086 avg prob of [ jazz] 0.9472318291664124 loss 0.15 = 0.042 + 0.022 + 0.086 avg prob of [ jazz] 0.9587138891220093 loss 0.14 = 0.034 + 0.02 + 0.086 avg prob of [ jazz] 0.966268002986908 loss 0.133 = 0.029 + 0.019 + 0.086 avg prob of [ jazz] 0.9717283844947815 loss 0.128 = 0.024 + 0.018 + 0.086 avg prob of [ jazz] 0.9759791493415833 loss 0.123 = 0.021 + 0.017 + 0.086 avg prob of [ jazz] 0.9794476628303528 loss 0.12 = 0.018 + 0.016 + 0.086 avg prob of [ jazz] 0.9823392033576965 loss 0.117 = 0.015 + 0.016 + 0.086 avg prob of [ jazz] 0.9847519397735596 Delta norm: 93.54818725585938 Change in target norm: 23.387046813964844 to 98.09980010986328 => 74.71275329589844 Division Factor: 10.181941986083984 Right vector norm: 9.187657356262207 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 12.383 = 12.383 + 0.0 + 0.0 avg prob of [ Hebrew] 5.535132368095219e-05 loss 7.518 = 7.485 + 0.018 + 0.015 avg prob of [ Hebrew] 0.0013686249731108546 loss 1.884 = 1.837 + 0.023 + 0.024 avg prob of [ Hebrew] 0.17246316373348236 loss 0.729 = 0.661 + 0.036 + 0.032 avg prob of [ Hebrew] 0.5261754393577576 loss 0.327 = 0.247 + 0.041 + 0.039 avg prob of [ Hebrew] 0.7843031883239746 loss 0.185 = 0.108 + 0.031 + 0.046 avg prob of [ Hebrew] 0.898270845413208 loss 0.133 = 0.056 + 0.025 + 0.051 avg prob of [ Hebrew] 0.9456570148468018 loss 0.114 = 0.032 + 0.025 + 0.057 avg prob of [ Hebrew] 0.9681828022003174 loss 0.107 = 0.021 + 0.025 + 0.061 avg prob of [ Hebrew] 0.9795052409172058 loss 0.105 = 0.014 + 0.025 + 0.066 avg prob of [ Hebrew] 0.9856364727020264 loss 0.105 = 0.011 + 0.025 + 0.069 avg prob of [ Hebrew] 0.9892413020133972 loss 0.105 = 0.009 + 0.024 + 0.073 avg prob of [ Hebrew] 0.9915258288383484 loss 0.106 = 0.007 + 0.023 + 0.076 avg prob of [ Hebrew] 0.9930673241615295 loss 0.106 = 0.006 + 0.022 + 0.078 avg prob of [ Hebrew] 0.9941437244415283 loss 0.104 = 0.005 + 0.021 + 0.078 avg prob of [ Hebrew] 0.9948920011520386 loss 0.102 = 0.005 + 0.019 + 0.078 avg prob of [ Hebrew] 0.995494544506073 loss 0.1 = 0.004 + 0.018 + 0.078 avg prob of [ Hebrew] 0.9959874749183655 loss 0.099 = 0.004 + 0.017 + 0.078 avg prob of [ Hebrew] 0.9963952898979187 loss 0.098 = 0.003 + 0.016 + 0.078 avg prob of [ Hebrew] 0.9967360496520996 loss 0.097 = 0.003 + 0.016 + 0.078 avg prob of [ Hebrew] 0.9970229268074036 Delta norm: 102.37065887451172 Change in target norm: 25.59266471862793 to 108.45658111572266 => 82.8639144897461 Division Factor: 10.492090225219727 Right vector norm: 9.756937980651855 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 14.799 = 14.799 + 0.0 + 0.0 avg prob of [ Bihar] 3.9947536834006314e-07 loss 12.834 = 12.814 + 0.007 + 0.013 avg prob of [ Bihar] 3.112594413323677e-06 loss 9.092 = 9.058 + 0.014 + 0.02 avg prob of [ Bihar] 0.00025576542248018086 loss 2.752 = 2.706 + 0.02 + 0.026 avg prob of [ Bihar] 0.10183995217084885 loss 1.185 = 1.117 + 0.036 + 0.031 avg prob of [ Bihar] 0.407859742641449 loss 0.621 = 0.529 + 0.056 + 0.037 avg prob of [ Bihar] 0.6340775489807129 loss 0.343 = 0.231 + 0.071 + 0.041 avg prob of [ Bihar] 0.8053826093673706 loss 0.23 = 0.105 + 0.08 + 0.046 avg prob of [ Bihar] 0.9031862616539001 loss 0.188 = 0.055 + 0.083 + 0.05 avg prob of [ Bihar] 0.9472885131835938 loss 0.169 = 0.033 + 0.082 + 0.054 avg prob of [ Bihar] 0.9674245715141296 loss 0.158 = 0.023 + 0.079 + 0.057 avg prob of [ Bihar] 0.9774990081787109 loss 0.152 = 0.017 + 0.074 + 0.06 avg prob of [ Bihar] 0.9830461144447327 loss 0.147 = 0.014 + 0.07 + 0.063 avg prob of [ Bihar] 0.9863659143447876 loss 0.144 = 0.012 + 0.067 + 0.066 avg prob of [ Bihar] 0.9884957671165466 loss 0.142 = 0.01 + 0.064 + 0.068 avg prob of [ Bihar] 0.9899448156356812 loss 0.141 = 0.009 + 0.062 + 0.07 avg prob of [ Bihar] 0.9909809827804565 loss 0.139 = 0.008 + 0.059 + 0.071 avg prob of [ Bihar] 0.9918530583381653 loss 0.136 = 0.007 + 0.057 + 0.071 avg prob of [ Bihar] 0.9927082061767578 loss 0.133 = 0.007 + 0.055 + 0.071 avg prob of [ Bihar] 0.9934009313583374 loss 0.131 = 0.006 + 0.053 + 0.071 avg prob of [ Bihar] 0.9939723014831543 Delta norm: 112.17367553710938 Change in target norm: 28.043420791625977 to 118.53560638427734 => 90.4921875 Division Factor: 6.916705131530762 Right vector norm: 16.217788696289062 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 10.371 = 10.371 + 0.0 + 0.0 avg prob of [ Lion] 4.490087667363696e-05 loss 7.165 = 7.139 + 0.015 + 0.011 avg prob of [ Lion] 0.0012237762566655874 loss 3.192 = 3.138 + 0.035 + 0.019 avg prob of [ Lion] 0.05584895610809326 loss 0.629 = 0.554 + 0.05 + 0.025 avg prob of [ Lion] 0.6013291478157043 loss 0.145 = 0.058 + 0.057 + 0.031 avg prob of [ Lion] 0.9445226788520813 loss 0.111 = 0.015 + 0.059 + 0.037 avg prob of [ Lion] 0.9851253032684326 loss 0.111 = 0.007 + 0.062 + 0.041 avg prob of [ Lion] 0.9927647113800049 loss 0.116 = 0.005 + 0.066 + 0.046 avg prob of [ Lion] 0.995226263999939 loss 0.12 = 0.004 + 0.067 + 0.05 avg prob of [ Lion] 0.9963358044624329 loss 0.12 = 0.003 + 0.064 + 0.053 avg prob of [ Lion] 0.9969242811203003 loss 0.118 = 0.003 + 0.059 + 0.056 avg prob of [ Lion] 0.9972416162490845 loss 0.117 = 0.003 + 0.055 + 0.059 avg prob of [ Lion] 0.9974001049995422 loss 0.116 = 0.003 + 0.052 + 0.062 avg prob of [ Lion] 0.997465968132019 loss 0.116 = 0.003 + 0.05 + 0.064 avg prob of [ Lion] 0.9974827766418457 loss 0.117 = 0.003 + 0.048 + 0.066 avg prob of [ Lion] 0.9974786043167114 loss 0.116 = 0.003 + 0.046 + 0.067 avg prob of [ Lion] 0.9974722862243652 loss 0.114 = 0.003 + 0.044 + 0.067 avg prob of [ Lion] 0.9974774718284607 loss 0.111 = 0.003 + 0.042 + 0.067 avg prob of [ Lion] 0.9974973797798157 loss 0.109 = 0.002 + 0.04 + 0.067 avg prob of [ Lion] 0.9975315928459167 loss 0.107 = 0.002 + 0.038 + 0.067 avg prob of [ Lion] 0.9975790977478027 Delta norm: 119.57958221435547 Change in target norm: 29.894895553588867 to 127.23020935058594 => 97.33531188964844 Division Factor: 8.519647598266602 Right vector norm: 14.03574275970459 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 6.531 = 6.531 + 0.0 + 0.0 avg prob of [ Belgium] 0.0027056192047894 loss 5.959 = 5.915 + 0.033 + 0.011 avg prob of [ Belgium] 0.004049048759043217 loss 5.432 = 5.382 + 0.032 + 0.018 avg prob of [ Belgium] 0.006281957030296326 loss 4.483 = 4.41 + 0.049 + 0.024 avg prob of [ Belgium] 0.015733759850263596 loss 3.148 = 3.062 + 0.057 + 0.029 avg prob of [ Belgium] 0.05501076579093933 loss 2.067 = 1.974 + 0.059 + 0.035 avg prob of [ Belgium] 0.14868581295013428 loss 1.189 = 1.085 + 0.064 + 0.04 avg prob of [ Belgium] 0.346853643655777 loss 0.535 = 0.416 + 0.074 + 0.045 avg prob of [ Belgium] 0.6641955375671387 loss 0.268 = 0.135 + 0.083 + 0.05 avg prob of [ Belgium] 0.874483585357666 loss 0.201 = 0.059 + 0.087 + 0.055 avg prob of [ Belgium] 0.942782998085022 loss 0.182 = 0.035 + 0.088 + 0.059 avg prob of [ Belgium] 0.9661215543746948 loss 0.175 = 0.024 + 0.088 + 0.064 avg prob of [ Belgium] 0.9767065644264221 loss 0.17 = 0.018 + 0.086 + 0.065 avg prob of [ Belgium] 0.9820435047149658 loss 0.165 = 0.015 + 0.086 + 0.065 avg prob of [ Belgium] 0.9854928851127625 loss 0.163 = 0.012 + 0.086 + 0.065 avg prob of [ Belgium] 0.9881460070610046 loss 0.161 = 0.01 + 0.086 + 0.065 avg prob of [ Belgium] 0.9901995062828064 loss 0.158 = 0.008 + 0.085 + 0.065 avg prob of [ Belgium] 0.991794764995575 loss 0.155 = 0.007 + 0.083 + 0.065 avg prob of [ Belgium] 0.9930365681648254 loss 0.152 = 0.006 + 0.08 + 0.065 avg prob of [ Belgium] 0.9940043687820435 loss 0.149 = 0.005 + 0.078 + 0.065 avg prob of [ Belgium] 0.9947603940963745 Delta norm: 122.4149169921875 Change in target norm: 30.603729248046875 to 129.3820037841797 => 98.77827453613281 Division Factor: 6.9614057540893555 Right vector norm: 17.584796905517578 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 5.527 = 5.527 + 0.0 + 0.0 avg prob of [ Russian] 0.0043369485065341 loss 1.024 = 1.001 + 0.008 + 0.015 avg prob of [ Russian] 0.38506990671157837 loss 0.437 = 0.399 + 0.013 + 0.025 avg prob of [ Russian] 0.6758978962898254 loss 0.23 = 0.182 + 0.014 + 0.033 avg prob of [ Russian] 0.8348459601402283 loss 0.148 = 0.093 + 0.014 + 0.041 avg prob of [ Russian] 0.9110163450241089 loss 0.116 = 0.056 + 0.013 + 0.047 avg prob of [ Russian] 0.9455772638320923 loss 0.105 = 0.039 + 0.012 + 0.053 avg prob of [ Russian] 0.9615085124969482 loss 0.103 = 0.031 + 0.013 + 0.058 avg prob of [ Russian] 0.9696221351623535 loss 0.104 = 0.026 + 0.015 + 0.063 avg prob of [ Russian] 0.9745491743087769 loss 0.105 = 0.022 + 0.016 + 0.067 avg prob of [ Russian] 0.9780892729759216 loss 0.106 = 0.019 + 0.016 + 0.071 avg prob of [ Russian] 0.9809091687202454 loss 0.107 = 0.017 + 0.016 + 0.074 avg prob of [ Russian] 0.9832600355148315 loss 0.107 = 0.015 + 0.015 + 0.077 avg prob of [ Russian] 0.9852484464645386 loss 0.104 = 0.013 + 0.014 + 0.077 avg prob of [ Russian] 0.9870527982711792 loss 0.102 = 0.012 + 0.013 + 0.077 avg prob of [ Russian] 0.9885011315345764 loss 0.1 = 0.01 + 0.012 + 0.077 avg prob of [ Russian] 0.9896560907363892 loss 0.098 = 0.009 + 0.012 + 0.077 avg prob of [ Russian] 0.9905833005905151 loss 0.097 = 0.009 + 0.011 + 0.077 avg prob of [ Russian] 0.9913373589515686 loss 0.096 = 0.008 + 0.011 + 0.077 avg prob of [ Russian] 0.9919606447219849 loss 0.095 = 0.008 + 0.011 + 0.077 avg prob of [ Russian] 0.9924857020378113 Delta norm: 103.70228576660156 Change in target norm: 25.925573348999023 to 108.03996276855469 => 82.11438751220703 Division Factor: 9.267975807189941 Right vector norm: 11.189312934875488 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 9.879 = 9.879 + 0.0 + 0.0 avg prob of [ singing] 0.0001397538580931723 loss 9.396 = 9.362 + 0.002 + 0.032 avg prob of [ singing] 0.0002697824966162443 loss 8.606 = 8.552 + 0.003 + 0.052 avg prob of [ singing] 0.0007882170611992478 loss 5.989 = 5.907 + 0.013 + 0.068 avg prob of [ singing] 0.014933564700186253 loss 2.371 = 2.249 + 0.04 + 0.083 avg prob of [ singing] 0.21726667881011963 loss 0.934 = 0.782 + 0.054 + 0.098 avg prob of [ singing] 0.48831114172935486 loss 0.528 = 0.354 + 0.061 + 0.113 avg prob of [ singing] 0.7102615833282471 loss 0.334 = 0.155 + 0.066 + 0.113 avg prob of [ singing] 0.8576299548149109 loss 0.263 = 0.078 + 0.072 + 0.113 avg prob of [ singing] 0.9248493909835815 loss 0.238 = 0.049 + 0.076 + 0.113 avg prob of [ singing] 0.9522802233695984 loss 0.223 = 0.035 + 0.075 + 0.113 avg prob of [ singing] 0.9654839038848877 loss 0.211 = 0.027 + 0.071 + 0.113 avg prob of [ singing] 0.973146915435791 loss 0.201 = 0.022 + 0.066 + 0.113 avg prob of [ singing] 0.9782410860061646 loss 0.194 = 0.018 + 0.063 + 0.113 avg prob of [ singing] 0.9819339513778687 loss 0.188 = 0.015 + 0.06 + 0.113 avg prob of [ singing] 0.9847483038902283 loss 0.183 = 0.013 + 0.058 + 0.113 avg prob of [ singing] 0.986953854560852 loss 0.18 = 0.011 + 0.056 + 0.113 avg prob of [ singing] 0.9887176752090454 loss 0.177 = 0.01 + 0.054 + 0.113 avg prob of [ singing] 0.9901586174964905 loss 0.175 = 0.009 + 0.053 + 0.113 avg prob of [ singing] 0.9913543462753296 loss 0.172 = 0.008 + 0.052 + 0.113 avg prob of [ singing] 0.9923558831214905 Delta norm: 70.9884262084961 Change in target norm: 17.747106552124023 to 73.90152740478516 => 56.1544189453125 Division Factor: 7.210412502288818 Right vector norm: 9.845266342163086 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 9.917 = 9.917 + 0.0 + 0.0 avg prob of [ HBO] 0.00015314178017433733 loss 9.261 = 9.24 + 0.006 + 0.015 avg prob of [ HBO] 0.00033487711334601045 loss 7.91 = 7.86 + 0.026 + 0.024 avg prob of [ HBO] 0.0013419591123238206 loss 4.799 = 4.708 + 0.061 + 0.031 avg prob of [ HBO] 0.01636204682290554 loss 1.792 = 1.672 + 0.083 + 0.037 avg prob of [ HBO] 0.27329298853874207 loss 0.497 = 0.352 + 0.101 + 0.044 avg prob of [ HBO] 0.7612909078598022 loss 0.21 = 0.051 + 0.109 + 0.05 avg prob of [ HBO] 0.9528597593307495 loss 0.163 = 0.008 + 0.1 + 0.055 avg prob of [ HBO] 0.9925509691238403 loss 0.148 = 0.002 + 0.085 + 0.061 avg prob of [ HBO] 0.9981619715690613 loss 0.141 = 0.001 + 0.074 + 0.066 avg prob of [ HBO] 0.9992476105690002 loss 0.14 = 0.0 + 0.069 + 0.071 avg prob of [ HBO] 0.9995472431182861 loss 0.143 = 0.0 + 0.068 + 0.075 avg prob of [ HBO] 0.9996451139450073 loss 0.144 = 0.0 + 0.067 + 0.077 avg prob of [ HBO] 0.9996767640113831 loss 0.143 = 0.0 + 0.066 + 0.077 avg prob of [ HBO] 0.9996920824050903 loss 0.143 = 0.0 + 0.066 + 0.077 avg prob of [ HBO] 0.9996981620788574 loss 0.142 = 0.0 + 0.065 + 0.077 avg prob of [ HBO] 0.9996986389160156 loss 0.141 = 0.0 + 0.063 + 0.077 avg prob of [ HBO] 0.999695360660553 loss 0.139 = 0.0 + 0.062 + 0.077 avg prob of [ HBO] 0.9996898174285889 loss 0.138 = 0.0 + 0.061 + 0.077 avg prob of [ HBO] 0.9996825456619263 loss 0.137 = 0.0 + 0.06 + 0.077 avg prob of [ HBO] 0.9996740221977234 Delta norm: 104.1490707397461 Change in target norm: 26.037263870239258 to 111.83891296386719 => 85.80165100097656 Division Factor: 9.844855308532715 Right vector norm: 10.579034805297852 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] 4.472856289794436e-06 loss 10.855 = 10.831 + 0.007 + 0.017 avg prob of [ Hindi] 4.29991923738271e-05 loss 7.653 = 7.609 + 0.015 + 0.029 avg prob of [ Hindi] 0.0009175987215712667 loss 3.666 = 3.596 + 0.03 + 0.04 avg prob of [ Hindi] 0.03822670504450798 loss 1.936 = 1.842 + 0.044 + 0.05 avg prob of [ Hindi] 0.17404505610466003 loss 1.181 = 1.076 + 0.046 + 0.059 avg prob of [ Hindi] 0.3488996922969818 loss 0.752 = 0.641 + 0.043 + 0.067 avg prob of [ Hindi] 0.5343066453933716 loss 0.403 = 0.286 + 0.042 + 0.075 avg prob of [ Hindi] 0.7530574202537537 loss 0.264 = 0.139 + 0.043 + 0.082 avg prob of [ Hindi] 0.870917558670044 loss 0.206 = 0.081 + 0.042 + 0.083 avg prob of [ Hindi] 0.9223244786262512 loss 0.176 = 0.053 + 0.04 + 0.083 avg prob of [ Hindi] 0.9480706453323364 loss 0.158 = 0.038 + 0.037 + 0.083 avg prob of [ Hindi] 0.9632166028022766 loss 0.146 = 0.027 + 0.035 + 0.083 avg prob of [ Hindi] 0.9730021357536316 loss 0.137 = 0.021 + 0.033 + 0.083 avg prob of [ Hindi] 0.9796698689460754 loss 0.13 = 0.016 + 0.031 + 0.083 avg prob of [ Hindi] 0.9843327403068542 loss 0.125 = 0.012 + 0.029 + 0.083 avg prob of [ Hindi] 0.9876429438591003 loss 0.121 = 0.01 + 0.027 + 0.083 avg prob of [ Hindi] 0.990023136138916 loss 0.117 = 0.008 + 0.026 + 0.083 avg prob of [ Hindi] 0.9917586445808411 loss 0.114 = 0.007 + 0.024 + 0.083 avg prob of [ Hindi] 0.993044376373291 loss 0.111 = 0.006 + 0.022 + 0.083 avg prob of [ Hindi] 0.9940123558044434 Delta norm: 96.01229095458984 Change in target norm: 24.003076553344727 to 100.79212951660156 => 76.78905487060547 Division Factor: 10.781982421875 Right vector norm: 8.904884338378906 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 13.012 = 13.012 + 0.0 + 0.0 avg prob of [ Dublin] 3.375142114236951e-06 loss 9.348 = 9.322 + 0.012 + 0.014 avg prob of [ Dublin] 0.0001053840242093429 loss 7.854 = 7.817 + 0.014 + 0.022 avg prob of [ Dublin] 0.0005438167718239129 loss 5.794 = 5.737 + 0.028 + 0.03 avg prob of [ Dublin] 0.004960678517818451 loss 2.874 = 2.791 + 0.046 + 0.036 avg prob of [ Dublin] 0.08654595911502838 loss 1.37 = 1.274 + 0.055 + 0.042 avg prob of [ Dublin] 0.3202897310256958 loss 0.694 = 0.584 + 0.063 + 0.047 avg prob of [ Dublin] 0.5808102488517761 loss 0.298 = 0.172 + 0.074 + 0.052 avg prob of [ Dublin] 0.8473354578018188 loss 0.183 = 0.05 + 0.076 + 0.057 avg prob of [ Dublin] 0.9523088335990906 loss 0.153 = 0.022 + 0.07 + 0.062 avg prob of [ Dublin] 0.9789397716522217 loss 0.139 = 0.013 + 0.06 + 0.066 avg prob of [ Dublin] 0.9868834614753723 loss 0.13 = 0.01 + 0.05 + 0.069 avg prob of [ Dublin] 0.9897631406784058 loss 0.125 = 0.009 + 0.043 + 0.073 avg prob of [ Dublin] 0.9909631609916687 loss 0.119 = 0.008 + 0.037 + 0.074 avg prob of [ Dublin] 0.9920799732208252 loss 0.113 = 0.007 + 0.032 + 0.074 avg prob of [ Dublin] 0.9932214617729187 loss 0.108 = 0.006 + 0.029 + 0.074 avg prob of [ Dublin] 0.9942139387130737 loss 0.105 = 0.005 + 0.026 + 0.074 avg prob of [ Dublin] 0.9950726628303528 loss 0.102 = 0.004 + 0.024 + 0.074 avg prob of [ Dublin] 0.9958029389381409 loss 0.099 = 0.004 + 0.022 + 0.074 avg prob of [ Dublin] 0.9964143633842468 loss 0.097 = 0.003 + 0.02 + 0.074 avg prob of [ Dublin] 0.9969210028648376 Delta norm: 108.23479461669922 Change in target norm: 27.058696746826172 to 112.28221893310547 => 85.22352600097656 Division Factor: 9.501596450805664 Right vector norm: 11.391221046447754 seq_released {"n_edits": 50, "retention": 1.0, "unedited_true": 0.83, "max_update_norm": 17.569990158081055, "median_update_norm": 9.013951778411865} (1419s) 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 7.847 = 7.847 + 0.0 + 0.0 avg prob of [ Dublin] 0.0004438208998180926 loss 5.249 = 5.218 + 0.017 + 0.014 avg prob of [ Dublin] 0.0069075291976332664 loss 1.699 = 1.626 + 0.052 + 0.021 avg prob of [ Dublin] 0.22717539966106415 loss 0.783 = 0.684 + 0.071 + 0.028 avg prob of [ Dublin] 0.5234515070915222 loss 0.349 = 0.233 + 0.082 + 0.034 avg prob of [ Dublin] 0.7950999736785889 loss 0.236 = 0.104 + 0.093 + 0.039 avg prob of [ Dublin] 0.902167022228241 loss 0.209 = 0.064 + 0.101 + 0.044 avg prob of [ Dublin] 0.938198983669281 loss 0.197 = 0.046 + 0.102 + 0.049 avg prob of [ Dublin] 0.9551329016685486 loss 0.185 = 0.034 + 0.098 + 0.053 avg prob of [ Dublin] 0.9670916795730591 loss 0.173 = 0.024 + 0.093 + 0.057 avg prob of [ Dublin] 0.9764431715011597 loss 0.163 = 0.017 + 0.086 + 0.06 avg prob of [ Dublin] 0.9832295179367065 loss 0.154 = 0.012 + 0.079 + 0.063 avg prob of [ Dublin] 0.9878363609313965 loss 0.146 = 0.009 + 0.071 + 0.066 avg prob of [ Dublin] 0.9908955693244934 loss 0.139 = 0.007 + 0.064 + 0.068 avg prob of [ Dublin] 0.9929395318031311 loss 0.133 = 0.006 + 0.057 + 0.071 avg prob of [ Dublin] 0.9943308234214783 loss 0.129 = 0.005 + 0.051 + 0.073 avg prob of [ Dublin] 0.9953014850616455 loss 0.126 = 0.004 + 0.047 + 0.075 avg prob of [ Dublin] 0.9960044622421265 loss 0.122 = 0.003 + 0.044 + 0.075 avg prob of [ Dublin] 0.9965468645095825 loss 0.12 = 0.003 + 0.042 + 0.075 avg prob of [ Dublin] 0.9969536066055298 loss 0.118 = 0.003 + 0.04 + 0.075 avg prob of [ Dublin] 0.9972691535949707 Delta norm: 107.27961730957031 Change in target norm: 26.819900512695312 to 114.59424591064453 => 87.77434539794922 Division Factor: 12.548563003540039 Right vector norm: 8.549155235290527 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 10.604 = 10.604 + 0.0 + 0.0 avg prob of [ politician] 4.178657036391087e-05 loss 7.633 = 7.61 + 0.012 + 0.011 avg prob of [ politician] 0.0010846618097275496 loss 5.763 = 5.713 + 0.033 + 0.018 avg prob of [ politician] 0.005463705398142338 loss 4.372 = 4.309 + 0.04 + 0.024 avg prob of [ politician] 0.020872030407190323 loss 3.193 = 3.122 + 0.042 + 0.029 avg prob of [ politician] 0.06423421949148178 loss 2.146 = 2.067 + 0.044 + 0.034 avg prob of [ politician] 0.17290166020393372 loss 1.332 = 1.247 + 0.046 + 0.039 avg prob of [ politician] 0.36587023735046387 loss 0.812 = 0.714 + 0.054 + 0.044 avg prob of [ politician] 0.5843842625617981 loss 0.509 = 0.394 + 0.066 + 0.049 avg prob of [ politician] 0.7589552998542786 loss 0.339 = 0.208 + 0.078 + 0.053 avg prob of [ politician] 0.8647571802139282 loss 0.24 = 0.099 + 0.084 + 0.057 avg prob of [ politician] 0.9238840937614441 loss 0.188 = 0.045 + 0.083 + 0.06 avg prob of [ politician] 0.9599432349205017 loss 0.164 = 0.023 + 0.077 + 0.064 avg prob of [ politician] 0.9776638746261597 loss 0.149 = 0.015 + 0.067 + 0.066 avg prob of [ politician] 0.9853153228759766 loss 0.133 = 0.012 + 0.055 + 0.066 avg prob of [ politician] 0.9887622594833374 loss 0.12 = 0.009 + 0.044 + 0.066 avg prob of [ politician] 0.9910872578620911 loss 0.11 = 0.007 + 0.036 + 0.066 avg prob of [ politician] 0.992766261100769 loss 0.103 = 0.006 + 0.031 + 0.066 avg prob of [ politician] 0.9940448999404907 loss 0.1 = 0.005 + 0.028 + 0.066 avg prob of [ politician] 0.9950459599494934 loss 0.098 = 0.004 + 0.027 + 0.066 avg prob of [ politician] 0.9958376884460449 Delta norm: 120.43208312988281 Change in target norm: 30.108022689819336 to 127.23170471191406 => 97.1236801147461 Division Factor: 8.902196884155273 Right vector norm: 13.52835464477539 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 14.036 = 14.036 + 0.0 + 0.0 avg prob of [ Tamil] 1.5373519772765576e-06 loss 10.03 = 10.02 + 0.006 + 0.005 avg prob of [ Tamil] 6.40240396023728e-05 loss 6.702 = 6.682 + 0.013 + 0.008 avg prob of [ Tamil] 0.0021521463058888912 loss 3.48 = 3.449 + 0.02 + 0.01 avg prob of [ Tamil] 0.05697329342365265 loss 2.286 = 2.247 + 0.026 + 0.013 avg prob of [ Tamil] 0.14019668102264404 loss 1.372 = 1.324 + 0.033 + 0.015 avg prob of [ Tamil] 0.2981800138950348 loss 0.794 = 0.736 + 0.041 + 0.017 avg prob of [ Tamil] 0.500558078289032 loss 0.479 = 0.41 + 0.05 + 0.019 avg prob of [ Tamil] 0.6730212569236755 loss 0.317 = 0.24 + 0.056 + 0.021 avg prob of [ Tamil] 0.7900407910346985 loss 0.229 = 0.148 + 0.058 + 0.022 avg prob of [ Tamil] 0.8637437224388123 loss 0.177 = 0.095 + 0.057 + 0.024 avg prob of [ Tamil] 0.9097622632980347 loss 0.146 = 0.064 + 0.057 + 0.025 avg prob of [ Tamil] 0.9387019276618958 loss 0.126 = 0.044 + 0.055 + 0.026 avg prob of [ Tamil] 0.9573320746421814 loss 0.112 = 0.031 + 0.053 + 0.028 avg prob of [ Tamil] 0.9696059226989746 loss 0.102 = 0.023 + 0.051 + 0.029 avg prob of [ Tamil] 0.9777514934539795 loss 0.095 = 0.017 + 0.049 + 0.03 avg prob of [ Tamil] 0.9831360578536987 loss 0.091 = 0.013 + 0.047 + 0.03 avg prob of [ Tamil] 0.9866688251495361 loss 0.088 = 0.011 + 0.046 + 0.031 avg prob of [ Tamil] 0.9889559745788574 loss 0.087 = 0.01 + 0.045 + 0.032 avg prob of [ Tamil] 0.9903921484947205 loss 0.086 = 0.009 + 0.045 + 0.033 avg prob of [ Tamil] 0.9912340044975281 Delta norm: 139.4416046142578 Change in target norm: 46.23244094848633 to 150.0008544921875 => 103.76841735839844 Division Factor: 11.448381423950195 Right vector norm: 12.180027961730957 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 3.33 = 3.33 + 0.0 + 0.0 avg prob of [ baseball] 0.05053706467151642 loss 0.987 = 0.971 + 0.006 + 0.01 avg prob of [ baseball] 0.4210773706436157 loss 0.084 = 0.05 + 0.019 + 0.016 avg prob of [ baseball] 0.9549140930175781 loss 0.088 = 0.031 + 0.036 + 0.021 avg prob of [ baseball] 0.9702146649360657 loss 0.101 = 0.027 + 0.049 + 0.025 avg prob of [ baseball] 0.9739501476287842 loss 0.105 = 0.024 + 0.052 + 0.028 avg prob of [ baseball] 0.9767053723335266 loss 0.1 = 0.021 + 0.047 + 0.032 avg prob of [ baseball] 0.9797089695930481 loss 0.091 = 0.018 + 0.038 + 0.034 avg prob of [ baseball] 0.9825443029403687 loss 0.083 = 0.016 + 0.031 + 0.037 avg prob of [ baseball] 0.9847593903541565 loss 0.079 = 0.014 + 0.026 + 0.039 avg prob of [ baseball] 0.9862903356552124 loss 0.076 = 0.013 + 0.022 + 0.041 avg prob of [ baseball] 0.987322211265564 loss 0.075 = 0.012 + 0.02 + 0.043 avg prob of [ baseball] 0.9880792498588562 loss 0.075 = 0.011 + 0.019 + 0.045 avg prob of [ baseball] 0.9887312054634094 loss 0.075 = 0.011 + 0.018 + 0.046 avg prob of [ baseball] 0.9893782734870911 loss 0.075 = 0.01 + 0.018 + 0.047 avg prob of [ baseball] 0.9900655150413513 loss 0.075 = 0.009 + 0.017 + 0.049 avg prob of [ baseball] 0.9908003807067871 loss 0.075 = 0.009 + 0.017 + 0.05 avg prob of [ baseball] 0.9915673136711121 loss 0.075 = 0.008 + 0.016 + 0.051 avg prob of [ baseball] 0.9923414587974548 loss 0.074 = 0.007 + 0.016 + 0.051 avg prob of [ baseball] 0.9930960536003113 loss 0.074 = 0.006 + 0.015 + 0.052 avg prob of [ baseball] 0.9938086271286011 Delta norm: 107.03309631347656 Change in target norm: 31.99724006652832 to 112.3260498046875 => 80.32881164550781 Division Factor: 9.607198715209961 Right vector norm: 11.1409273147583 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 10.376 = 10.376 + 0.0 + 0.0 avg prob of [ opera] 5.209146183915436e-05 loss 9.405 = 9.389 + 0.002 + 0.015 avg prob of [ opera] 0.0001518915523774922 loss 8.494 = 8.464 + 0.006 + 0.024 avg prob of [ opera] 0.0003982878988608718 loss 7.059 = 7.013 + 0.013 + 0.033 avg prob of [ opera] 0.0017329475376755 loss 4.95 = 4.883 + 0.026 + 0.041 avg prob of [ opera] 0.012087374925613403 loss 3.127 = 3.04 + 0.038 + 0.049 avg prob of [ opera] 0.05616115406155586 loss 1.365 = 1.261 + 0.048 + 0.056 avg prob of [ opera] 0.29933980107307434 loss 0.392 = 0.26 + 0.069 + 0.063 avg prob of [ opera] 0.7743654847145081 loss 0.254 = 0.079 + 0.105 + 0.069 avg prob of [ opera] 0.9239030480384827 loss 0.239 = 0.043 + 0.121 + 0.076 avg prob of [ opera] 0.9580920338630676 loss 0.196 = 0.034 + 0.086 + 0.076 avg prob of [ opera] 0.9670680165290833 loss 0.169 = 0.028 + 0.065 + 0.076 avg prob of [ opera] 0.9724225401878357 loss 0.155 = 0.024 + 0.055 + 0.076 avg prob of [ opera] 0.9766838550567627 loss 0.146 = 0.02 + 0.049 + 0.076 avg prob of [ opera] 0.9801926612854004 loss 0.139 = 0.017 + 0.045 + 0.076 avg prob of [ opera] 0.9831721782684326 loss 0.133 = 0.014 + 0.042 + 0.076 avg prob of [ opera] 0.985764741897583 loss 0.128 = 0.012 + 0.04 + 0.076 avg prob of [ opera] 0.9880155920982361 loss 0.124 = 0.01 + 0.038 + 0.076 avg prob of [ opera] 0.9899314641952515 loss 0.121 = 0.009 + 0.036 + 0.076 avg prob of [ opera] 0.9915246367454529 loss 0.119 = 0.007 + 0.035 + 0.076 avg prob of [ opera] 0.9928250908851624 Delta norm: 104.82521057128906 Change in target norm: 26.206302642822266 to 109.8839111328125 => 83.6776123046875 Division Factor: 10.637153625488281 Right vector norm: 9.854630470275879 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 7.729 = 7.729 + 0.0 + 0.0 avg prob of [ Nissan] 0.0008056974620558321 loss 2.252 = 2.223 + 0.017 + 0.012 avg prob of [ Nissan] 0.22660984098911285 loss 1.157 = 1.091 + 0.045 + 0.021 avg prob of [ Nissan] 0.45348435640335083 loss 0.587 = 0.503 + 0.056 + 0.027 avg prob of [ Nissan] 0.6527972221374512 loss 0.269 = 0.179 + 0.056 + 0.033 avg prob of [ Nissan] 0.8400477170944214 loss 0.17 = 0.073 + 0.058 + 0.039 avg prob of [ Nissan] 0.9301371574401855 loss 0.14 = 0.037 + 0.06 + 0.044 avg prob of [ Nissan] 0.963988184928894 loss 0.13 = 0.021 + 0.06 + 0.048 avg prob of [ Nissan] 0.9788581132888794 loss 0.126 = 0.014 + 0.06 + 0.052 avg prob of [ Nissan] 0.9862697124481201 loss 0.124 = 0.01 + 0.059 + 0.055 avg prob of [ Nissan] 0.9902971386909485 loss 0.123 = 0.007 + 0.057 + 0.058 avg prob of [ Nissan] 0.9926315546035767 loss 0.123 = 0.006 + 0.055 + 0.061 avg prob of [ Nissan] 0.9940586090087891 loss 0.122 = 0.005 + 0.054 + 0.064 avg prob of [ Nissan] 0.9949734210968018 loss 0.122 = 0.004 + 0.052 + 0.066 avg prob of [ Nissan] 0.9955866932868958 loss 0.122 = 0.004 + 0.05 + 0.068 avg prob of [ Nissan] 0.9960164427757263 loss 0.121 = 0.004 + 0.048 + 0.07 avg prob of [ Nissan] 0.9963311553001404 loss 0.12 = 0.003 + 0.046 + 0.07 avg prob of [ Nissan] 0.9965752363204956 loss 0.118 = 0.003 + 0.045 + 0.07 avg prob of [ Nissan] 0.996784508228302 loss 0.116 = 0.003 + 0.043 + 0.07 avg prob of [ Nissan] 0.9969707727432251 loss 0.115 = 0.003 + 0.042 + 0.07 avg prob of [ Nissan] 0.9971387982368469 Delta norm: 113.7303695678711 Change in target norm: 28.432594299316406 to 119.11705780029297 => 90.68446350097656 Division Factor: 1.5647116899490356 Right vector norm: 72.68455505371094 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 5.988 = 5.988 + 0.0 + 0.0 avg prob of [ piano] 0.004063047468662262 loss 2.86 = 2.838 + 0.009 + 0.013 avg prob of [ piano] 0.06385885179042816 loss 0.953 = 0.882 + 0.05 + 0.021 avg prob of [ piano] 0.4266478717327118 loss 0.468 = 0.34 + 0.1 + 0.028 avg prob of [ piano] 0.7169352769851685 loss 0.269 = 0.144 + 0.091 + 0.034 avg prob of [ piano] 0.867296040058136 loss 0.174 = 0.065 + 0.069 + 0.04 avg prob of [ piano] 0.9374747276306152 loss 0.127 = 0.032 + 0.05 + 0.045 avg prob of [ piano] 0.9685308337211609 loss 0.106 = 0.017 + 0.04 + 0.049 avg prob of [ piano] 0.9828570485115051 loss 0.098 = 0.01 + 0.035 + 0.053 avg prob of [ piano] 0.9898678660392761 loss 0.095 = 0.007 + 0.032 + 0.057 avg prob of [ piano] 0.9935184717178345 loss 0.093 = 0.004 + 0.029 + 0.06 avg prob of [ piano] 0.9955277442932129 loss 0.093 = 0.003 + 0.027 + 0.063 avg prob of [ piano] 0.9966902732849121 loss 0.092 = 0.003 + 0.024 + 0.065 avg prob of [ piano] 0.997399628162384 loss 0.092 = 0.002 + 0.022 + 0.068 avg prob of [ piano] 0.9978575706481934 loss 0.092 = 0.002 + 0.02 + 0.07 avg prob of [ piano] 0.9981703758239746 loss 0.092 = 0.002 + 0.019 + 0.071 avg prob of [ piano] 0.9984026551246643 loss 0.091 = 0.001 + 0.018 + 0.071 avg prob of [ piano] 0.9985796809196472 loss 0.09 = 0.001 + 0.018 + 0.071 avg prob of [ piano] 0.9987105131149292 loss 0.09 = 0.001 + 0.017 + 0.071 avg prob of [ piano] 0.9988076090812683 loss 0.089 = 0.001 + 0.017 + 0.071 avg prob of [ piano] 0.9988790154457092 Delta norm: 112.45140075683594 Change in target norm: 28.112850189208984 to 118.16455841064453 => 90.05171203613281 Division Factor: 11.13074016571045 Right vector norm: 10.102779388427734 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 5.912 = 5.912 + 0.0 + 0.0 avg prob of [ Google] 0.004956049378961325 loss 0.375 = 0.336 + 0.034 + 0.005 avg prob of [ Google] 0.7192598581314087 loss 0.179 = 0.115 + 0.056 + 0.009 avg prob of [ Google] 0.892218291759491 loss 0.138 = 0.071 + 0.055 + 0.012 avg prob of [ Google] 0.9317492246627808 loss 0.119 = 0.054 + 0.051 + 0.014 avg prob of [ Google] 0.947935938835144 loss 0.107 = 0.045 + 0.047 + 0.016 avg prob of [ Google] 0.9567664265632629 loss 0.1 = 0.038 + 0.044 + 0.018 avg prob of [ Google] 0.9629828333854675 loss 0.094 = 0.033 + 0.042 + 0.019 avg prob of [ Google] 0.9680553674697876 loss 0.089 = 0.028 + 0.041 + 0.02 avg prob of [ Google] 0.9723897576332092 loss 0.086 = 0.024 + 0.04 + 0.022 avg prob of [ Google] 0.9761239290237427 loss 0.083 = 0.021 + 0.039 + 0.023 avg prob of [ Google] 0.9793487787246704 loss 0.08 = 0.018 + 0.038 + 0.024 avg prob of [ Google] 0.982136070728302 loss 0.078 = 0.016 + 0.037 + 0.025 avg prob of [ Google] 0.9845358729362488 loss 0.076 = 0.014 + 0.037 + 0.025 avg prob of [ Google] 0.9865838885307312 loss 0.074 = 0.012 + 0.036 + 0.026 avg prob of [ Google] 0.9883115887641907 loss 0.072 = 0.01 + 0.035 + 0.027 avg prob of [ Google] 0.9897521138191223 loss 0.071 = 0.009 + 0.034 + 0.027 avg prob of [ Google] 0.9909420013427734 loss 0.07 = 0.008 + 0.034 + 0.028 avg prob of [ Google] 0.991919994354248 loss 0.069 = 0.007 + 0.033 + 0.028 avg prob of [ Google] 0.992723822593689 loss 0.068 = 0.007 + 0.033 + 0.029 avg prob of [ Google] 0.9933874011039734 Delta norm: 110.32018280029297 Change in target norm: 43.68524932861328 to 119.19844055175781 => 75.51319122314453 Division Factor: 9.331751823425293 Right vector norm: 11.822022438049316 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 5.237 = 5.237 + 0.0 + 0.0 avg prob of [ physics] 0.006062699947506189 loss 2.082 = 2.064 + 0.005 + 0.013 avg prob of [ physics] 0.1379716694355011 loss 1.677 = 1.632 + 0.023 + 0.022 avg prob of [ physics] 0.20613625645637512 loss 0.42 = 0.36 + 0.03 + 0.029 avg prob of [ physics] 0.7007075548171997 loss 0.139 = 0.053 + 0.05 + 0.036 avg prob of [ physics] 0.9490337371826172 loss 0.161 = 0.038 + 0.081 + 0.041 avg prob of [ physics] 0.9630085825920105 loss 0.126 = 0.029 + 0.05 + 0.046 avg prob of [ physics] 0.9718421101570129 loss 0.122 = 0.018 + 0.053 + 0.051 avg prob of [ physics] 0.9826067090034485 loss 0.119 = 0.011 + 0.054 + 0.055 avg prob of [ physics] 0.989504337310791 loss 0.115 = 0.007 + 0.05 + 0.058 avg prob of [ physics] 0.9932367205619812 loss 0.114 = 0.005 + 0.048 + 0.062 avg prob of [ physics] 0.9952829480171204 loss 0.117 = 0.004 + 0.049 + 0.065 avg prob of [ physics] 0.9964687824249268 loss 0.114 = 0.003 + 0.044 + 0.067 avg prob of [ physics] 0.997198224067688 loss 0.11 = 0.002 + 0.038 + 0.07 avg prob of [ physics] 0.99767005443573 loss 0.11 = 0.002 + 0.036 + 0.072 avg prob of [ physics] 0.9979840517044067 loss 0.108 = 0.002 + 0.034 + 0.072 avg prob of [ physics] 0.9981558918952942 loss 0.105 = 0.002 + 0.031 + 0.072 avg prob of [ physics] 0.9982357025146484 loss 0.103 = 0.002 + 0.029 + 0.072 avg prob of [ physics] 0.9982785582542419 loss 0.102 = 0.002 + 0.028 + 0.072 avg prob of [ physics] 0.9983029961585999 loss 0.101 = 0.002 + 0.027 + 0.072 avg prob of [ physics] 0.9983223676681519 Delta norm: 110.35810089111328 Change in target norm: 27.58952522277832 to 118.68294525146484 => 91.09342193603516 Division Factor: 9.32628059387207 Right vector norm: 11.833024978637695 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 7.582 = 7.582 + 0.0 + 0.0 avg prob of [ English] 0.000597747100982815 loss 4.167 = 4.135 + 0.025 + 0.006 avg prob of [ English] 0.019066818058490753 loss 0.793 = 0.755 + 0.028 + 0.01 avg prob of [ English] 0.5252488255500793 loss 0.061 = 0.022 + 0.026 + 0.013 avg prob of [ English] 0.9789601564407349 loss 0.054 = 0.004 + 0.034 + 0.016 avg prob of [ English] 0.9960394501686096 loss 0.06 = 0.004 + 0.037 + 0.018 avg prob of [ English] 0.9959326982498169 loss 0.058 = 0.006 + 0.032 + 0.021 avg prob of [ English] 0.9943591356277466 loss 0.058 = 0.007 + 0.028 + 0.023 avg prob of [ English] 0.9926100373268127 loss 0.059 = 0.009 + 0.025 + 0.024 avg prob of [ English] 0.991014838218689 loss 0.06 = 0.01 + 0.024 + 0.026 avg prob of [ English] 0.989932656288147 loss 0.06 = 0.01 + 0.023 + 0.027 avg prob of [ English] 0.9898924231529236 loss 0.06 = 0.009 + 0.022 + 0.029 avg prob of [ English] 0.9908759593963623 loss 0.059 = 0.008 + 0.021 + 0.03 avg prob of [ English] 0.9923422932624817 loss 0.058 = 0.006 + 0.021 + 0.031 avg prob of [ English] 0.9938021898269653 loss 0.057 = 0.005 + 0.02 + 0.032 avg prob of [ English] 0.9950242042541504 loss 0.056 = 0.004 + 0.019 + 0.033 avg prob of [ English] 0.9959626197814941 loss 0.055 = 0.003 + 0.019 + 0.033 avg prob of [ English] 0.9966540336608887 loss 0.055 = 0.003 + 0.018 + 0.034 avg prob of [ English] 0.9971553087234497 loss 0.054 = 0.002 + 0.017 + 0.035 avg prob of [ English] 0.9975184202194214 loss 0.054 = 0.002 + 0.017 + 0.035 avg prob of [ English] 0.9977843761444092 Delta norm: 113.9834213256836 Change in target norm: 40.23126220703125 to 126.29327392578125 => 86.06201171875 Division Factor: 11.650592803955078 Right vector norm: 9.783486366271973 seq_released {"n_edits": 60, "retention": 1.0, "unedited_true": 0.83, "max_update_norm": 72.61759185791016, "median_update_norm": 9.283258438110352} (1692s) 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 9.25 = 9.25 + 0.0 + 0.0 avg prob of [ Lyon] 0.0003147078095935285 loss 4.521 = 4.512 + 0.004 + 0.005 avg prob of [ Lyon] 0.021502014249563217 loss 1.316 = 1.296 + 0.012 + 0.008 avg prob of [ Lyon] 0.3675330877304077 loss 0.388 = 0.344 + 0.033 + 0.011 avg prob of [ Lyon] 0.7496614456176758 loss 0.136 = 0.073 + 0.049 + 0.013 avg prob of [ Lyon] 0.9335436820983887 loss 0.096 = 0.023 + 0.058 + 0.015 avg prob of [ Lyon] 0.9776244163513184 loss 0.097 = 0.016 + 0.064 + 0.017 avg prob of [ Lyon] 0.9846209287643433 loss 0.098 = 0.011 + 0.068 + 0.019 avg prob of [ Lyon] 0.988619327545166 loss 0.097 = 0.008 + 0.068 + 0.02 avg prob of [ Lyon] 0.9915671348571777 loss 0.093 = 0.006 + 0.064 + 0.022 avg prob of [ Lyon] 0.9935842156410217 loss 0.088 = 0.005 + 0.06 + 0.023 avg prob of [ Lyon] 0.9949541687965393 loss 0.083 = 0.004 + 0.055 + 0.024 avg prob of [ Lyon] 0.9959002733230591 loss 0.08 = 0.003 + 0.051 + 0.025 avg prob of [ Lyon] 0.9965699315071106 loss 0.076 = 0.003 + 0.048 + 0.026 avg prob of [ Lyon] 0.9970564246177673 loss 0.074 = 0.003 + 0.044 + 0.027 avg prob of [ Lyon] 0.9974187016487122 loss 0.071 = 0.002 + 0.042 + 0.027 avg prob of [ Lyon] 0.9976950287818909 loss 0.069 = 0.002 + 0.039 + 0.028 avg prob of [ Lyon] 0.9979100227355957 loss 0.067 = 0.002 + 0.037 + 0.029 avg prob of [ Lyon] 0.9980805516242981 loss 0.066 = 0.002 + 0.035 + 0.029 avg prob of [ Lyon] 0.9982182383537292 loss 0.064 = 0.002 + 0.033 + 0.03 avg prob of [ Lyon] 0.998331606388092 Delta norm: 116.57701873779297 Change in target norm: 44.21839904785156 to 127.31855010986328 => 83.10015106201172 Division Factor: 10.733002662658691 Right vector norm: 10.861547470092773 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 8.133 = 8.133 + 0.0 + 0.0 avg prob of [ Portsmouth] 0.0007176085491664708 loss 7.05 = 7.028 + 0.004 + 0.018 avg prob of [ Portsmouth] 0.0019298276165500283 loss 5.564 = 5.517 + 0.015 + 0.032 avg prob of [ Portsmouth] 0.006253011059015989 loss 3.723 = 3.644 + 0.034 + 0.044 avg prob of [ Portsmouth] 0.03573520481586456 loss 2.062 = 1.959 + 0.047 + 0.056 avg prob of [ Portsmouth] 0.18735095858573914 loss 0.817 = 0.702 + 0.049 + 0.066 avg prob of [ Portsmouth] 0.5659772753715515 loss 0.305 = 0.178 + 0.052 + 0.075 avg prob of [ Portsmouth] 0.8529710173606873 loss 0.184 = 0.04 + 0.061 + 0.084 avg prob of [ Portsmouth] 0.9630882740020752 loss 0.158 = 0.018 + 0.055 + 0.085 avg prob of [ Portsmouth] 0.9827377796173096 loss 0.146 = 0.011 + 0.049 + 0.085 avg prob of [ Portsmouth] 0.9893813133239746 loss 0.14 = 0.008 + 0.047 + 0.085 avg prob of [ Portsmouth] 0.9924898743629456 loss 0.136 = 0.006 + 0.045 + 0.085 avg prob of [ Portsmouth] 0.9941412806510925 loss 0.133 = 0.005 + 0.043 + 0.085 avg prob of [ Portsmouth] 0.9951187968254089 loss 0.13 = 0.004 + 0.041 + 0.085 avg prob of [ Portsmouth] 0.995738685131073 loss 0.129 = 0.004 + 0.04 + 0.085 avg prob of [ Portsmouth] 0.9961420893669128 loss 0.128 = 0.004 + 0.039 + 0.085 avg prob of [ Portsmouth] 0.9964075684547424 loss 0.128 = 0.003 + 0.039 + 0.085 avg prob of [ Portsmouth] 0.9965925812721252 loss 0.127 = 0.003 + 0.038 + 0.085 avg prob of [ Portsmouth] 0.9967398643493652 loss 0.125 = 0.003 + 0.037 + 0.085 avg prob of [ Portsmouth] 0.9968770146369934 loss 0.124 = 0.003 + 0.035 + 0.085 avg prob of [ Portsmouth] 0.9970152974128723 Delta norm: 93.78255462646484 Change in target norm: 23.44563865661621 to 93.79891204833984 => 70.353271484375 Division Factor: 8.347647666931152 Right vector norm: 11.23460865020752 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 11.087 = 11.087 + 0.0 + 0.0 avg prob of [ Italy] 2.4961338567663915e-05 loss 4.55 = 4.526 + 0.019 + 0.005 avg prob of [ Italy] 0.014642614871263504 loss 2.175 = 2.147 + 0.019 + 0.008 avg prob of [ Italy] 0.16416165232658386 loss 1.258 = 1.231 + 0.017 + 0.01 avg prob of [ Italy] 0.3288019597530365 loss 0.905 = 0.875 + 0.017 + 0.013 avg prob of [ Italy] 0.43649473786354065 loss 0.588 = 0.554 + 0.02 + 0.015 avg prob of [ Italy] 0.5874505043029785 loss 0.327 = 0.286 + 0.024 + 0.016 avg prob of [ Italy] 0.757257342338562 loss 0.185 = 0.137 + 0.03 + 0.018 avg prob of [ Italy] 0.8743529319763184 loss 0.124 = 0.069 + 0.035 + 0.02 avg prob of [ Italy] 0.9336463212966919 loss 0.098 = 0.039 + 0.038 + 0.021 avg prob of [ Italy] 0.9618082642555237 loss 0.085 = 0.025 + 0.039 + 0.022 avg prob of [ Italy] 0.9757713079452515 loss 0.078 = 0.017 + 0.038 + 0.023 avg prob of [ Italy] 0.9832413792610168 loss 0.073 = 0.013 + 0.037 + 0.024 avg prob of [ Italy] 0.9875739812850952 loss 0.07 = 0.01 + 0.035 + 0.025 avg prob of [ Italy] 0.9902753233909607 loss 0.068 = 0.008 + 0.034 + 0.026 avg prob of [ Italy] 0.9920641779899597 loss 0.066 = 0.007 + 0.033 + 0.027 avg prob of [ Italy] 0.9933076500892639 loss 0.065 = 0.006 + 0.032 + 0.027 avg prob of [ Italy] 0.9942065477371216 loss 0.065 = 0.005 + 0.032 + 0.028 avg prob of [ Italy] 0.9948773384094238 loss 0.064 = 0.005 + 0.031 + 0.029 avg prob of [ Italy] 0.9953911900520325 loss 0.064 = 0.004 + 0.031 + 0.029 avg prob of [ Italy] 0.9957934021949768 Delta norm: 120.50543975830078 Change in target norm: 45.515499114990234 to 130.0450439453125 => 84.529541015625 Division Factor: 9.443029403686523 Right vector norm: 12.761311531066895 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 2.131 = 2.131 + 0.0 + 0.0 avg prob of [ PBS] 0.14191129803657532 loss 3.752 = 3.726 + 0.022 + 0.004 avg prob of [ PBS] 0.029971225187182426 loss 1.306 = 1.279 + 0.021 + 0.006 avg prob of [ PBS] 0.3238673806190491 loss 0.296 = 0.269 + 0.019 + 0.008 avg prob of [ PBS] 0.811984658241272 loss 0.155 = 0.126 + 0.019 + 0.01 avg prob of [ PBS] 0.911289393901825 loss 0.101 = 0.071 + 0.019 + 0.012 avg prob of [ PBS] 0.9457780718803406 loss 0.073 = 0.041 + 0.019 + 0.014 avg prob of [ PBS] 0.9652706384658813 loss 0.058 = 0.025 + 0.018 + 0.015 avg prob of [ PBS] 0.9775840044021606 loss 0.05 = 0.016 + 0.018 + 0.016 avg prob of [ PBS] 0.9848935008049011 loss 0.046 = 0.011 + 0.017 + 0.018 avg prob of [ PBS] 0.9889534711837769 Delta norm: 88.1240234375 Change in target norm: 50.14441680908203 to 103.46134185791016 => 53.316925048828125 Division Factor: 12.793971061706543 Right vector norm: 6.887933731079102 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 10.187 = 10.187 + 0.0 + 0.0 avg prob of [ IBM] 0.00021101099264342338 loss 2.45 = 2.432 + 0.013 + 0.005 avg prob of [ IBM] 0.10310593247413635 loss 0.689 = 0.665 + 0.016 + 0.008 avg prob of [ IBM] 0.574550449848175 loss 0.256 = 0.229 + 0.017 + 0.01 avg prob of [ IBM] 0.8171835541725159 loss 0.119 = 0.09 + 0.017 + 0.012 avg prob of [ IBM] 0.9177436828613281 loss 0.078 = 0.047 + 0.017 + 0.014 avg prob of [ IBM] 0.9547688364982605 loss 0.064 = 0.031 + 0.018 + 0.015 avg prob of [ IBM] 0.9694562554359436 loss 0.059 = 0.024 + 0.018 + 0.017 avg prob of [ IBM] 0.9763034582138062 loss 0.057 = 0.02 + 0.018 + 0.018 avg prob of [ IBM] 0.9799882769584656 loss 0.056 = 0.018 + 0.019 + 0.019 avg prob of [ IBM] 0.9822367429733276 loss 0.055 = 0.016 + 0.019 + 0.02 avg prob of [ IBM] 0.9837793707847595 loss 0.055 = 0.015 + 0.019 + 0.021 avg prob of [ IBM] 0.9849562644958496 loss 0.055 = 0.014 + 0.019 + 0.022 avg prob of [ IBM] 0.9859366416931152 loss 0.055 = 0.013 + 0.019 + 0.023 avg prob of [ IBM] 0.9868065714836121 loss 0.055 = 0.012 + 0.019 + 0.023 avg prob of [ IBM] 0.9876095652580261 loss 0.055 = 0.012 + 0.019 + 0.024 avg prob of [ IBM] 0.9883677959442139 loss 0.054 = 0.011 + 0.019 + 0.024 avg prob of [ IBM] 0.9890913963317871 loss 0.054 = 0.01 + 0.019 + 0.025 avg prob of [ IBM] 0.9897840619087219 loss 0.054 = 0.01 + 0.019 + 0.025 avg prob of [ IBM] 0.9904464483261108 loss 0.053 = 0.009 + 0.018 + 0.026 avg prob of [ IBM] 0.9910771250724792 Delta norm: 109.91109466552734 Change in target norm: 46.18030548095703 to 118.10823822021484 => 71.92793273925781 Division Factor: 9.68595027923584 Right vector norm: 11.347476959228516 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 7.192 = 7.192 + 0.0 + 0.0 avg prob of [ Italian] 0.0017842070665210485 loss 2.406 = 2.385 + 0.017 + 0.005 avg prob of [ Italian] 0.11740118265151978 loss 0.584 = 0.523 + 0.053 + 0.008 avg prob of [ Italian] 0.629105269908905 loss 0.33 = 0.259 + 0.061 + 0.01 avg prob of [ Italian] 0.7837226390838623 loss 0.206 = 0.132 + 0.062 + 0.013 avg prob of [ Italian] 0.8798956274986267 loss 0.149 = 0.072 + 0.063 + 0.015 avg prob of [ Italian] 0.9318000078201294 loss 0.125 = 0.044 + 0.066 + 0.016 avg prob of [ Italian] 0.9575924873352051 loss 0.115 = 0.03 + 0.067 + 0.018 avg prob of [ Italian] 0.9703452587127686 loss 0.108 = 0.023 + 0.066 + 0.019 avg prob of [ Italian] 0.977023184299469 loss 0.102 = 0.019 + 0.062 + 0.02 avg prob of [ Italian] 0.9809343218803406 loss 0.096 = 0.017 + 0.058 + 0.021 avg prob of [ Italian] 0.983548104763031 loss 0.09 = 0.015 + 0.053 + 0.022 avg prob of [ Italian] 0.9855130314826965 loss 0.086 = 0.013 + 0.05 + 0.023 avg prob of [ Italian] 0.9871189594268799 loss 0.082 = 0.012 + 0.046 + 0.024 avg prob of [ Italian] 0.9884936809539795 loss 0.079 = 0.01 + 0.044 + 0.025 avg prob of [ Italian] 0.9896941184997559 loss 0.076 = 0.009 + 0.041 + 0.025 avg prob of [ Italian] 0.9907479882240295 loss 0.073 = 0.008 + 0.038 + 0.026 avg prob of [ Italian] 0.9916733503341675 loss 0.07 = 0.008 + 0.036 + 0.027 avg prob of [ Italian] 0.9924843907356262 loss 0.067 = 0.007 + 0.033 + 0.027 avg prob of [ Italian] 0.9931957721710205 loss 0.065 = 0.006 + 0.031 + 0.028 avg prob of [ Italian] 0.9938198328018188 Delta norm: 111.69661712646484 Change in target norm: 45.05204772949219 to 126.87650299072266 => 81.82445526123047 Division Factor: 10.049603462219238 Right vector norm: 11.114530563354492 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 11.328 = 11.328 + 0.0 + 0.0 avg prob of [ Indonesia] 1.692485238891095e-05 loss 10.031 = 9.989 + 0.028 + 0.013 avg prob of [ Indonesia] 5.9559562942013144e-05 loss 2.991 = 2.917 + 0.053 + 0.02 avg prob of [ Indonesia] 0.07589620351791382 loss 1.438 = 1.331 + 0.08 + 0.027 avg prob of [ Indonesia] 0.29692190885543823 loss 0.359 = 0.204 + 0.123 + 0.032 avg prob of [ Indonesia] 0.8233004212379456 loss 0.229 = 0.057 + 0.134 + 0.038 avg prob of [ Indonesia] 0.9450669884681702 loss 0.176 = 0.029 + 0.104 + 0.043 avg prob of [ Indonesia] 0.9715710878372192 loss 0.156 = 0.02 + 0.089 + 0.047 avg prob of [ Indonesia] 0.9806511402130127 loss 0.151 = 0.015 + 0.084 + 0.051 avg prob of [ Indonesia] 0.9846808910369873 loss 0.149 = 0.013 + 0.08 + 0.055 avg prob of [ Indonesia] 0.9868386387825012 loss 0.148 = 0.012 + 0.077 + 0.059 avg prob of [ Indonesia] 0.9881782531738281 loss 0.147 = 0.011 + 0.075 + 0.062 avg prob of [ Indonesia] 0.9891020655632019 loss 0.148 = 0.01 + 0.073 + 0.065 avg prob of [ Indonesia] 0.9897898435592651 loss 0.149 = 0.01 + 0.072 + 0.067 avg prob of [ Indonesia] 0.9903387427330017 loss 0.15 = 0.009 + 0.071 + 0.069 avg prob of [ Indonesia] 0.9908089637756348 loss 0.151 = 0.009 + 0.071 + 0.072 avg prob of [ Indonesia] 0.9912393689155579 loss 0.151 = 0.008 + 0.071 + 0.072 avg prob of [ Indonesia] 0.9917117357254028 loss 0.15 = 0.008 + 0.07 + 0.072 avg prob of [ Indonesia] 0.9922393560409546 loss 0.149 = 0.007 + 0.07 + 0.072 avg prob of [ Indonesia] 0.9927629828453064 loss 0.149 = 0.007 + 0.069 + 0.072 avg prob of [ Indonesia] 0.9932746887207031 Delta norm: 110.34695434570312 Change in target norm: 27.586734771728516 to 117.90917205810547 => 90.32243347167969 Division Factor: 11.543368339538574 Right vector norm: 9.559337615966797 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 12.838 = 12.838 + 0.0 + 0.0 avg prob of [ Sega] 3.7667705328203738e-06 loss 10.336 = 10.309 + 0.004 + 0.024 avg prob of [ Sega] 5.7252400438301265e-05 loss 7.126 = 7.075 + 0.01 + 0.041 avg prob of [ Sega] 0.001557464012876153 loss 3.14 = 3.059 + 0.026 + 0.055 avg prob of [ Sega] 0.05325529724359512 loss 1.486 = 1.376 + 0.042 + 0.069 avg prob of [ Sega] 0.25740283727645874 loss 0.999 = 0.866 + 0.052 + 0.081 avg prob of [ Sega] 0.4262484908103943 loss 0.621 = 0.469 + 0.06 + 0.092 avg prob of [ Sega] 0.6291900277137756 loss 0.381 = 0.218 + 0.066 + 0.097 avg prob of [ Sega] 0.8050679564476013 loss 0.268 = 0.104 + 0.067 + 0.097 avg prob of [ Sega] 0.9017127156257629 loss 0.22 = 0.055 + 0.068 + 0.097 avg prob of [ Sega] 0.9463991522789001 loss 0.198 = 0.033 + 0.068 + 0.097 avg prob of [ Sega] 0.9673219323158264 loss 0.187 = 0.022 + 0.067 + 0.097 avg prob of [ Sega] 0.9780699610710144 loss 0.179 = 0.016 + 0.066 + 0.097 avg prob of [ Sega] 0.9841511249542236 loss 0.174 = 0.012 + 0.064 + 0.097 avg prob of [ Sega] 0.987881064414978 loss 0.17 = 0.01 + 0.063 + 0.097 avg prob of [ Sega] 0.9903216361999512 loss 0.167 = 0.008 + 0.061 + 0.097 avg prob of [ Sega] 0.9920035600662231 loss 0.164 = 0.007 + 0.059 + 0.097 avg prob of [ Sega] 0.9932128190994263 loss 0.161 = 0.006 + 0.058 + 0.097 avg prob of [ Sega] 0.9941139221191406 loss 0.159 = 0.005 + 0.056 + 0.097 avg prob of [ Sega] 0.994805634021759 loss 0.156 = 0.005 + 0.054 + 0.097 avg prob of [ Sega] 0.9953505396842957 Delta norm: 82.07894134521484 Change in target norm: 20.519737243652344 to 85.27234649658203 => 64.75260925292969 Division Factor: 6.225778579711914 Right vector norm: 13.183722496032715 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 12.117 = 12.117 + 0.0 + 0.0 avg prob of [ goaltender] 0.00012546006473712623 loss 6.885 = 6.875 + 0.002 + 0.008 avg prob of [ goaltender] 0.0012625411618500948 loss 6.592 = 6.576 + 0.003 + 0.013 avg prob of [ goaltender] 0.0017218717839568853 loss 6.173 = 6.153 + 0.003 + 0.017 avg prob of [ goaltender] 0.0026261149905622005 loss 5.457 = 5.432 + 0.004 + 0.021 avg prob of [ goaltender] 0.005437660496681929 loss 4.264 = 4.234 + 0.006 + 0.024 avg prob of [ goaltender] 0.018010923638939857 loss 3.07 = 3.032 + 0.012 + 0.027 avg prob of [ goaltender] 0.055700402706861496 loss 2.558 = 2.507 + 0.021 + 0.03 avg prob of [ goaltender] 0.08888803422451019 loss 2.309 = 2.245 + 0.032 + 0.032 avg prob of [ goaltender] 0.11394218355417252 loss 2.119 = 2.043 + 0.042 + 0.035 avg prob of [ goaltender] 0.1386280655860901 loss 1.928 = 1.84 + 0.05 + 0.037 avg prob of [ goaltender] 0.16870594024658203 loss 1.702 = 1.604 + 0.058 + 0.039 avg prob of [ goaltender] 0.21209929883480072 loss 1.443 = 1.336 + 0.066 + 0.042 avg prob of [ goaltender] 0.27475711703300476 loss 1.177 = 1.062 + 0.072 + 0.043 avg prob of [ goaltender] 0.3579874336719513 loss 0.936 = 0.816 + 0.075 + 0.045 avg prob of [ goaltender] 0.4540095925331116 loss 0.744 = 0.622 + 0.075 + 0.047 avg prob of [ goaltender] 0.5478419065475464 loss 0.602 = 0.481 + 0.072 + 0.049 avg prob of [ goaltender] 0.6277956962585449 loss 0.498 = 0.383 + 0.064 + 0.05 avg prob of [ goaltender] 0.690269947052002 loss 0.42 = 0.315 + 0.053 + 0.052 avg prob of [ goaltender] 0.7377155423164368 loss 0.357 = 0.263 + 0.04 + 0.053 avg prob of [ goaltender] 0.7753821611404419 Delta norm: 132.49356079101562 Change in target norm: 35.23655700683594 to 133.4499969482422 => 98.21343994140625 Division Factor: 13.095870018005371 Right vector norm: 10.117201805114746 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 10.777 = 10.777 + 0.0 + 0.0 avg prob of [ quarterback] 5.641456664307043e-05 loss 6.704 = 6.664 + 0.033 + 0.007 avg prob of [ quarterback] 0.0018274562899023294 loss 3.287 = 3.212 + 0.065 + 0.011 avg prob of [ quarterback] 0.05357660725712776 loss 1.949 = 1.829 + 0.106 + 0.014 avg prob of [ quarterback] 0.16823868453502655 loss 1.373 = 1.242 + 0.114 + 0.017 avg prob of [ quarterback] 0.29598554968833923 loss 0.978 = 0.863 + 0.095 + 0.02 avg prob of [ quarterback] 0.4294463098049164 loss 0.713 = 0.611 + 0.08 + 0.023 avg prob of [ quarterback] 0.5506410002708435 loss 0.539 = 0.44 + 0.074 + 0.025 avg prob of [ quarterback] 0.6510215401649475 loss 0.421 = 0.321 + 0.073 + 0.027 avg prob of [ quarterback] 0.7309166193008423 loss 0.339 = 0.237 + 0.073 + 0.029 avg prob of [ quarterback] 0.793066680431366 loss 0.279 = 0.177 + 0.072 + 0.031 avg prob of [ quarterback] 0.8410063982009888 loss 0.234 = 0.132 + 0.07 + 0.032 avg prob of [ quarterback] 0.8779315948486328 loss 0.199 = 0.1 + 0.066 + 0.034 avg prob of [ quarterback] 0.9063618779182434 loss 0.172 = 0.075 + 0.061 + 0.035 avg prob of [ quarterback] 0.928215503692627 loss 0.15 = 0.057 + 0.056 + 0.036 avg prob of [ quarterback] 0.9449542164802551 loss 0.132 = 0.043 + 0.052 + 0.037 avg prob of [ quarterback] 0.9577003717422485 loss 0.12 = 0.033 + 0.048 + 0.039 avg prob of [ quarterback] 0.9673281311988831 loss 0.111 = 0.026 + 0.045 + 0.039 avg prob of [ quarterback] 0.9745352864265442 loss 0.104 = 0.02 + 0.043 + 0.04 avg prob of [ quarterback] 0.9798856377601624 loss 0.099 = 0.016 + 0.042 + 0.041 avg prob of [ quarterback] 0.9838318824768066 Delta norm: 114.75473022460938 Change in target norm: 37.328948974609375 to 124.43107604980469 => 87.10212707519531 Division Factor: 8.70467472076416 Right vector norm: 13.18311595916748 seq_released {"n_edits": 70, "retention": 1.0, "unedited_true": 0.83, "max_update_norm": 72.61759185791016, "median_update_norm": 9.550270557403564} (1967s) 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 10.548 = 10.548 + 0.0 + 0.0 avg prob of [ psychology] 4.893040022579953e-05 loss 6.327 = 6.312 + 0.011 + 0.004 avg prob of [ psychology] 0.0027352063916623592 loss 4.366 = 4.345 + 0.014 + 0.007 avg prob of [ psychology] 0.013990218751132488 loss 3.022 = 3.0 + 0.013 + 0.009 avg prob of [ psychology] 0.05098820477724075 loss 1.826 = 1.805 + 0.01 + 0.011 avg prob of [ psychology] 0.1687208116054535 loss 0.962 = 0.937 + 0.013 + 0.013 avg prob of [ psychology] 0.39770492911338806 loss 0.523 = 0.486 + 0.023 + 0.014 avg prob of [ psychology] 0.617703914642334 loss 0.343 = 0.292 + 0.035 + 0.016 avg prob of [ psychology] 0.7474137544631958 loss 0.246 = 0.182 + 0.047 + 0.017 avg prob of [ psychology] 0.834041953086853 loss 0.185 = 0.111 + 0.056 + 0.018 avg prob of [ psychology] 0.895626962184906 loss 0.149 = 0.069 + 0.061 + 0.02 avg prob of [ psychology] 0.9335165619850159 loss 0.126 = 0.047 + 0.059 + 0.021 avg prob of [ psychology] 0.9544646739959717 loss 0.11 = 0.035 + 0.053 + 0.022 avg prob of [ psychology] 0.9659054279327393 loss 0.098 = 0.028 + 0.048 + 0.023 avg prob of [ psychology] 0.9724117517471313 loss 0.091 = 0.024 + 0.044 + 0.023 avg prob of [ psychology] 0.9763381481170654 loss 0.088 = 0.021 + 0.043 + 0.024 avg prob of [ psychology] 0.9788835644721985 loss 0.088 = 0.02 + 0.043 + 0.025 avg prob of [ psychology] 0.9806798100471497 loss 0.087 = 0.018 + 0.044 + 0.025 avg prob of [ psychology] 0.9820663928985596 loss 0.086 = 0.017 + 0.043 + 0.026 avg prob of [ psychology] 0.9832234382629395 loss 0.084 = 0.016 + 0.042 + 0.027 avg prob of [ psychology] 0.9842436909675598 Delta norm: 132.40615844726562 Change in target norm: 49.92937088012695 to 142.9669189453125 => 93.03755187988281 Division Factor: 11.669675827026367 Right vector norm: 11.346174240112305 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 16.105 = 16.105 + 0.0 + 0.0 avg prob of [ politician] 1.0824055607372429e-07 loss 15.246 = 15.233 + 0.005 + 0.008 avg prob of [ politician] 2.817074573613354e-07 loss 14.018 = 13.973 + 0.032 + 0.013 avg prob of [ politician] 1.1852558827740722e-06 loss 13.091 = 13.027 + 0.047 + 0.017 avg prob of [ politician] 5.70719976167311e-06 loss 11.152 = 11.091 + 0.041 + 0.02 avg prob of [ politician] 3.889252548106015e-05 loss 8.022 = 7.971 + 0.028 + 0.023 avg prob of [ politician] 0.0003899675502907485 loss 6.899 = 6.849 + 0.024 + 0.026 avg prob of [ politician] 0.0011344546219334006 loss 5.418 = 5.363 + 0.026 + 0.029 avg prob of [ politician] 0.0049283867701888084 loss 3.439 = 3.374 + 0.034 + 0.031 avg prob of [ politician] 0.03604351729154587 loss 1.876 = 1.799 + 0.043 + 0.034 avg prob of [ politician] 0.1740346997976303 loss 0.682 = 0.594 + 0.051 + 0.037 avg prob of [ politician] 0.56071537733078 loss 0.166 = 0.06 + 0.067 + 0.039 avg prob of [ politician] 0.941787838935852 loss 0.151 = 0.011 + 0.098 + 0.042 avg prob of [ politician] 0.9889453649520874 loss 0.178 = 0.008 + 0.125 + 0.045 avg prob of [ politician] 0.9921472668647766 loss 0.188 = 0.009 + 0.132 + 0.047 avg prob of [ politician] 0.9912157654762268 loss 0.19 = 0.012 + 0.128 + 0.049 avg prob of [ politician] 0.9879918694496155 loss 0.191 = 0.018 + 0.121 + 0.052 avg prob of [ politician] 0.9822618365287781 loss 0.184 = 0.023 + 0.108 + 0.054 avg prob of [ politician] 0.9775241017341614 loss 0.164 = 0.022 + 0.087 + 0.055 avg prob of [ politician] 0.9781447052955627 loss 0.142 = 0.018 + 0.067 + 0.057 avg prob of [ politician] 0.9826446771621704 Delta norm: 138.2867889404297 Change in target norm: 34.81953430175781 to 145.47964477539062 => 110.66011047363281 Division Factor: 12.114548683166504 Right vector norm: 11.414935111999512 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 8.626 = 8.626 + 0.0 + 0.0 avg prob of [ Germany] 0.0004233257204759866 loss 4.993 = 4.979 + 0.004 + 0.011 avg prob of [ Germany] 0.010019445791840553 loss 3.418 = 3.392 + 0.008 + 0.019 avg prob of [ Germany] 0.05156989395618439 loss 2.308 = 2.269 + 0.014 + 0.025 avg prob of [ Germany] 0.1445697546005249 loss 1.921 = 1.873 + 0.017 + 0.03 avg prob of [ Germany] 0.18981479108333588 loss 1.559 = 1.505 + 0.019 + 0.035 avg prob of [ Germany] 0.2600729465484619 loss 1.067 = 1.006 + 0.022 + 0.04 avg prob of [ Germany] 0.4090889096260071 loss 0.589 = 0.519 + 0.027 + 0.044 avg prob of [ Germany] 0.6300350427627563 loss 0.32 = 0.236 + 0.036 + 0.048 avg prob of [ Germany] 0.8046993017196655 loss 0.212 = 0.116 + 0.045 + 0.051 avg prob of [ Germany] 0.8944249153137207 loss 0.169 = 0.066 + 0.048 + 0.055 avg prob of [ Germany] 0.9373559355735779 loss 0.149 = 0.042 + 0.049 + 0.058 avg prob of [ Germany] 0.9592052698135376 loss 0.139 = 0.029 + 0.049 + 0.061 avg prob of [ Germany] 0.9713004231452942 loss 0.134 = 0.022 + 0.048 + 0.064 avg prob of [ Germany] 0.9785252809524536 loss 0.131 = 0.017 + 0.048 + 0.066 avg prob of [ Germany] 0.9831339716911316 loss 0.128 = 0.014 + 0.047 + 0.067 avg prob of [ Germany] 0.9863823652267456 loss 0.124 = 0.011 + 0.046 + 0.067 avg prob of [ Germany] 0.9887747764587402 loss 0.121 = 0.01 + 0.044 + 0.067 avg prob of [ Germany] 0.9905440807342529 loss 0.118 = 0.008 + 0.042 + 0.067 avg prob of [ Germany] 0.9918891191482544 loss 0.115 = 0.007 + 0.041 + 0.067 avg prob of [ Germany] 0.9929384589195251 Delta norm: 119.03401184082031 Change in target norm: 29.758501052856445 to 125.62329864501953 => 95.86479949951172 Division Factor: 8.971134185791016 Right vector norm: 13.268558502197266 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 11.237 = 11.237 + 0.0 + 0.0 avg prob of [ Israel] 2.18749246414518e-05 loss 7.376 = 7.362 + 0.008 + 0.005 avg prob of [ Israel] 0.001425687107257545 loss 4.825 = 4.802 + 0.015 + 0.008 avg prob of [ Israel] 0.01973032020032406 loss 2.369 = 2.34 + 0.018 + 0.011 avg prob of [ Israel] 0.14006073772907257 loss 0.84 = 0.802 + 0.024 + 0.014 avg prob of [ Israel] 0.5061785578727722 loss 0.234 = 0.18 + 0.037 + 0.016 avg prob of [ Israel] 0.8440195918083191 loss 0.134 = 0.065 + 0.051 + 0.019 avg prob of [ Israel] 0.9395651817321777 loss 0.121 = 0.044 + 0.056 + 0.021 avg prob of [ Israel] 0.9590744972229004 loss 0.113 = 0.036 + 0.055 + 0.023 avg prob of [ Israel] 0.9662445783615112 loss 0.105 = 0.031 + 0.05 + 0.024 avg prob of [ Israel] 0.9706234931945801 loss 0.098 = 0.027 + 0.045 + 0.026 avg prob of [ Israel] 0.9742209911346436 loss 0.093 = 0.023 + 0.042 + 0.027 avg prob of [ Israel] 0.9774519801139832 loss 0.089 = 0.02 + 0.04 + 0.029 avg prob of [ Israel] 0.9803383350372314 loss 0.086 = 0.018 + 0.039 + 0.03 avg prob of [ Israel] 0.9828429222106934 loss 0.084 = 0.015 + 0.038 + 0.031 avg prob of [ Israel] 0.9849610924720764 loss 0.083 = 0.013 + 0.038 + 0.032 avg prob of [ Israel] 0.9867247343063354 loss 0.082 = 0.012 + 0.037 + 0.033 avg prob of [ Israel] 0.9881841540336609 loss 0.081 = 0.011 + 0.037 + 0.033 avg prob of [ Israel] 0.9893921613693237 loss 0.08 = 0.01 + 0.036 + 0.034 avg prob of [ Israel] 0.9903963804244995 loss 0.079 = 0.009 + 0.035 + 0.035 avg prob of [ Israel] 0.9912371039390564 Delta norm: 135.95147705078125 Change in target norm: 44.252891540527344 to 144.6611328125 => 100.40824127197266 Division Factor: 13.703455924987793 Right vector norm: 9.920963287353516 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 10.825 = 10.825 + 0.0 + 0.0 avg prob of [ Suzuki] 4.6430544898612425e-05 loss 4.375 = 4.361 + 0.006 + 0.007 avg prob of [ Suzuki] 0.0174533873796463 loss 2.019 = 1.989 + 0.018 + 0.012 avg prob of [ Suzuki] 0.1626463085412979 loss 0.742 = 0.699 + 0.026 + 0.016 avg prob of [ Suzuki] 0.527522087097168 loss 0.382 = 0.326 + 0.036 + 0.02 avg prob of [ Suzuki] 0.7357528209686279 loss 0.244 = 0.176 + 0.044 + 0.024 avg prob of [ Suzuki] 0.8436534404754639 loss 0.189 = 0.112 + 0.051 + 0.027 avg prob of [ Suzuki] 0.8962473273277283 loss 0.165 = 0.082 + 0.054 + 0.029 avg prob of [ Suzuki] 0.9230620861053467 loss 0.152 = 0.064 + 0.056 + 0.032 avg prob of [ Suzuki] 0.9387724995613098 loss 0.144 = 0.053 + 0.057 + 0.034 avg prob of [ Suzuki] 0.9494568109512329 loss 0.137 = 0.044 + 0.057 + 0.036 avg prob of [ Suzuki] 0.957583487033844 loss 0.132 = 0.037 + 0.057 + 0.038 avg prob of [ Suzuki] 0.9641305208206177 loss 0.127 = 0.031 + 0.057 + 0.039 avg prob of [ Suzuki] 0.9695219397544861 loss 0.123 = 0.027 + 0.056 + 0.041 avg prob of [ Suzuki] 0.9739844799041748 loss 0.12 = 0.023 + 0.055 + 0.042 avg prob of [ Suzuki] 0.9776740074157715 loss 0.117 = 0.02 + 0.054 + 0.043 avg prob of [ Suzuki] 0.9807183146476746 loss 0.115 = 0.017 + 0.053 + 0.044 avg prob of [ Suzuki] 0.9832273125648499 loss 0.113 = 0.015 + 0.052 + 0.045 avg prob of [ Suzuki] 0.9852970838546753 loss 0.111 = 0.013 + 0.051 + 0.046 avg prob of [ Suzuki] 0.9870085716247559 loss 0.109 = 0.012 + 0.051 + 0.047 avg prob of [ Suzuki] 0.9884293675422668 Delta norm: 129.90969848632812 Change in target norm: 37.102622985839844 to 135.40577697753906 => 98.30315399169922 Division Factor: 7.831785678863525 Right vector norm: 16.587493896484375 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 5.68 = 5.68 + 0.0 + 0.0 avg prob of [ violin] 0.01965927705168724 loss 5.231 = 5.209 + 0.007 + 0.015 avg prob of [ violin] 0.027404004707932472 loss 4.354 = 4.312 + 0.017 + 0.025 avg prob of [ violin] 0.05018649250268936 loss 2.206 = 2.137 + 0.036 + 0.033 avg prob of [ violin] 0.2070646733045578 loss 0.957 = 0.827 + 0.089 + 0.042 avg prob of [ violin] 0.4423091411590576 loss 0.594 = 0.425 + 0.12 + 0.049 avg prob of [ violin] 0.6558979749679565 loss 0.313 = 0.141 + 0.116 + 0.055 avg prob of [ violin] 0.8693793416023254 loss 0.213 = 0.048 + 0.103 + 0.062 avg prob of [ violin] 0.9528955817222595 loss 0.183 = 0.022 + 0.093 + 0.067 avg prob of [ violin] 0.9779108166694641 loss 0.174 = 0.014 + 0.088 + 0.073 avg prob of [ violin] 0.9863592982292175 loss 0.173 = 0.01 + 0.085 + 0.078 avg prob of [ violin] 0.9898043274879456 loss 0.169 = 0.009 + 0.082 + 0.078 avg prob of [ violin] 0.9909976124763489 loss 0.166 = 0.008 + 0.08 + 0.078 avg prob of [ violin] 0.9922632575035095 loss 0.163 = 0.006 + 0.079 + 0.078 avg prob of [ violin] 0.9938151240348816 loss 0.161 = 0.005 + 0.078 + 0.078 avg prob of [ violin] 0.9953660368919373 loss 0.158 = 0.003 + 0.077 + 0.078 avg prob of [ violin] 0.9966514110565186 loss 0.156 = 0.002 + 0.076 + 0.078 avg prob of [ violin] 0.9975932240486145 loss 0.155 = 0.002 + 0.075 + 0.078 avg prob of [ violin] 0.9982377886772156 loss 0.153 = 0.001 + 0.073 + 0.078 avg prob of [ violin] 0.9986656904220581 loss 0.151 = 0.001 + 0.072 + 0.078 avg prob of [ violin] 0.9989481568336487 Delta norm: 102.25040435791016 Change in target norm: 25.56260108947754 to 106.00999450683594 => 80.44739532470703 Division Factor: 9.505414962768555 Right vector norm: 10.757068634033203 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 7.006 = 7.006 + 0.0 + 0.0 avg prob of [ Polish] 0.0019608584698289633 loss 6.331 = 6.32 + 0.001 + 0.01 avg prob of [ Polish] 0.0036053217481821775 loss 5.342 = 5.32 + 0.005 + 0.016 avg prob of [ Polish] 0.007801392115652561 loss 3.542 = 3.511 + 0.009 + 0.022 avg prob of [ Polish] 0.03826051205396652 loss 1.947 = 1.908 + 0.013 + 0.027 avg prob of [ Polish] 0.17435070872306824 loss 1.085 = 1.04 + 0.014 + 0.032 avg prob of [ Polish] 0.3834150433540344 loss 0.526 = 0.475 + 0.014 + 0.037 avg prob of [ Polish] 0.637047290802002 loss 0.219 = 0.162 + 0.015 + 0.042 avg prob of [ Polish] 0.8533509373664856 loss 0.115 = 0.051 + 0.018 + 0.047 avg prob of [ Polish] 0.9508923888206482 loss 0.09 = 0.018 + 0.021 + 0.051 avg prob of [ Polish] 0.9817928075790405 loss 0.087 = 0.008 + 0.024 + 0.055 avg prob of [ Polish] 0.991831362247467 loss 0.089 = 0.004 + 0.026 + 0.058 avg prob of [ Polish] 0.9956522583961487 loss 0.091 = 0.003 + 0.027 + 0.062 avg prob of [ Polish] 0.9973570704460144 loss 0.089 = 0.002 + 0.024 + 0.063 avg prob of [ Polish] 0.9981141686439514 loss 0.086 = 0.001 + 0.022 + 0.063 avg prob of [ Polish] 0.9985337257385254 loss 0.084 = 0.001 + 0.02 + 0.063 avg prob of [ Polish] 0.998810350894928 loss 0.082 = 0.001 + 0.019 + 0.063 avg prob of [ Polish] 0.998999834060669 loss 0.081 = 0.001 + 0.018 + 0.063 avg prob of [ Polish] 0.9991337060928345 loss 0.08 = 0.001 + 0.017 + 0.063 avg prob of [ Polish] 0.999230682849884 loss 0.079 = 0.001 + 0.016 + 0.063 avg prob of [ Polish] 0.9993026256561279 Delta norm: 127.90155029296875 Change in target norm: 31.975385665893555 to 133.2173614501953 => 101.24197387695312 Division Factor: 11.480807304382324 Right vector norm: 11.14046573638916 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 6.865 = 6.865 + 0.0 + 0.0 avg prob of [ Africa] 0.004296228755265474 loss 0.768 = 0.755 + 0.011 + 0.003 avg prob of [ Africa] 0.5143827199935913 loss 0.161 = 0.137 + 0.019 + 0.005 avg prob of [ Africa] 0.8917235136032104 loss 0.065 = 0.033 + 0.026 + 0.006 avg prob of [ Africa] 0.9683775305747986 loss 0.06 = 0.018 + 0.034 + 0.007 avg prob of [ Africa] 0.9821820259094238 loss 0.063 = 0.015 + 0.04 + 0.008 avg prob of [ Africa] 0.9848474860191345 loss 0.066 = 0.015 + 0.042 + 0.009 avg prob of [ Africa] 0.9855251312255859 loss 0.067 = 0.014 + 0.043 + 0.01 avg prob of [ Africa] 0.9856791496276855 loss 0.067 = 0.014 + 0.042 + 0.011 avg prob of [ Africa] 0.9856778383255005 loss 0.066 = 0.014 + 0.04 + 0.012 avg prob of [ Africa] 0.9856505393981934 loss 0.065 = 0.014 + 0.038 + 0.012 avg prob of [ Africa] 0.9856536388397217 loss 0.063 = 0.014 + 0.036 + 0.013 avg prob of [ Africa] 0.9857149720191956 loss 0.061 = 0.014 + 0.034 + 0.013 avg prob of [ Africa] 0.9858487248420715 loss 0.06 = 0.014 + 0.032 + 0.013 avg prob of [ Africa] 0.9860619306564331 loss 0.058 = 0.014 + 0.031 + 0.014 avg prob of [ Africa] 0.9863551259040833 loss 0.057 = 0.013 + 0.03 + 0.014 avg prob of [ Africa] 0.9867243766784668 loss 0.056 = 0.013 + 0.029 + 0.014 avg prob of [ Africa] 0.9871628880500793 loss 0.055 = 0.012 + 0.028 + 0.015 avg prob of [ Africa] 0.9876598119735718 loss 0.054 = 0.012 + 0.027 + 0.015 avg prob of [ Africa] 0.9882019758224487 loss 0.053 = 0.011 + 0.026 + 0.015 avg prob of [ Africa] 0.9887755513191223 Delta norm: 108.8011474609375 Change in target norm: 59.77522659301758 to 121.80696105957031 => 62.031734466552734 Division Factor: 10.46786117553711 Right vector norm: 10.393828392028809 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 10.426 = 10.426 + 0.0 + 0.0 avg prob of [ Colombia] 4.3122337956447154e-05 loss 4.3 = 4.279 + 0.012 + 0.009 avg prob of [ Colombia] 0.02236171066761017 loss 1.817 = 1.772 + 0.03 + 0.015 avg prob of [ Colombia] 0.20081236958503723 loss 0.558 = 0.496 + 0.042 + 0.02 avg prob of [ Colombia] 0.6260383129119873 loss 0.193 = 0.118 + 0.05 + 0.024 avg prob of [ Colombia] 0.8915001153945923 loss 0.13 = 0.045 + 0.056 + 0.028 avg prob of [ Colombia] 0.9568077325820923 loss 0.116 = 0.025 + 0.059 + 0.032 avg prob of [ Colombia] 0.975470244884491 loss 0.112 = 0.018 + 0.059 + 0.035 avg prob of [ Colombia] 0.9824163317680359 loss 0.11 = 0.015 + 0.057 + 0.038 avg prob of [ Colombia] 0.9855543971061707 loss 0.109 = 0.013 + 0.056 + 0.04 avg prob of [ Colombia] 0.9871817827224731 loss 0.109 = 0.012 + 0.054 + 0.043 avg prob of [ Colombia] 0.9881396293640137 loss 0.109 = 0.011 + 0.053 + 0.045 avg prob of [ Colombia] 0.988797128200531 loss 0.109 = 0.011 + 0.052 + 0.047 avg prob of [ Colombia] 0.9893324971199036 loss 0.109 = 0.01 + 0.051 + 0.048 avg prob of [ Colombia] 0.9898327589035034 loss 0.109 = 0.01 + 0.049 + 0.05 avg prob of [ Colombia] 0.9903345704078674 loss 0.108 = 0.009 + 0.048 + 0.051 avg prob of [ Colombia] 0.9908478260040283 loss 0.108 = 0.009 + 0.046 + 0.052 avg prob of [ Colombia] 0.9913682341575623 loss 0.107 = 0.008 + 0.045 + 0.054 avg prob of [ Colombia] 0.9918862581253052 loss 0.106 = 0.008 + 0.043 + 0.055 avg prob of [ Colombia] 0.9923922419548035 loss 0.105 = 0.007 + 0.042 + 0.055 avg prob of [ Colombia] 0.9928776621818542 Delta norm: 120.16376495361328 Change in target norm: 32.90620803833008 to 129.02752685546875 => 96.12132263183594 Division Factor: 7.91520881652832 Right vector norm: 15.181376457214355 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 11.003 = 11.003 + 0.0 + 0.0 avg prob of [ opera] 5.945572775090113e-05 loss 7.888 = 7.862 + 0.013 + 0.013 avg prob of [ opera] 0.0005566634936258197 loss 5.766 = 5.716 + 0.03 + 0.02 avg prob of [ opera] 0.004649584647268057 loss 3.652 = 3.571 + 0.055 + 0.027 avg prob of [ opera] 0.03620129078626633 loss 1.833 = 1.67 + 0.131 + 0.033 avg prob of [ opera] 0.2073369324207306 loss 0.576 = 0.421 + 0.117 + 0.038 avg prob of [ opera] 0.6635988354682922 loss 0.218 = 0.064 + 0.11 + 0.044 avg prob of [ opera] 0.9391136169433594 loss 0.136 = 0.009 + 0.079 + 0.049 avg prob of [ opera] 0.9910710453987122 loss 0.138 = 0.009 + 0.076 + 0.053 avg prob of [ opera] 0.991165280342102 loss 0.141 = 0.014 + 0.07 + 0.057 avg prob of [ opera] 0.9865826368331909 loss 0.14 = 0.016 + 0.063 + 0.061 avg prob of [ opera] 0.9843876957893372 loss 0.135 = 0.013 + 0.058 + 0.065 avg prob of [ opera] 0.9874615669250488 loss 0.138 = 0.012 + 0.058 + 0.068 avg prob of [ opera] 0.9878625273704529 loss 0.14 = 0.01 + 0.059 + 0.071 avg prob of [ opera] 0.9898259043693542 loss 0.129 = 0.006 + 0.051 + 0.071 avg prob of [ opera] 0.9935517311096191 loss 0.127 = 0.004 + 0.052 + 0.071 avg prob of [ opera] 0.9960025548934937 loss 0.128 = 0.003 + 0.054 + 0.071 avg prob of [ opera] 0.9974520802497864 loss 0.12 = 0.002 + 0.047 + 0.071 avg prob of [ opera] 0.9982898235321045 loss 0.115 = 0.001 + 0.043 + 0.071 avg prob of [ opera] 0.99876868724823 loss 0.115 = 0.001 + 0.042 + 0.071 avg prob of [ opera] 0.9990404844284058 Delta norm: 112.09595489501953 Change in target norm: 28.023990631103516 to 120.22486877441406 => 92.20088195800781 Division Factor: 8.81004524230957 Right vector norm: 12.723653793334961 seq_released {"n_edits": 80, "retention": 1.0, "unedited_true": 0.82, "max_update_norm": 72.61759185791016, "median_update_norm": 9.806334495544434} (2259s) 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 16.183 = 16.183 + 0.0 + 0.0 avg prob of [ goaltender] 1.0101073399937377e-07 loss 5.74 = 5.731 + 0.004 + 0.005 avg prob of [ goaltender] 0.004501196555793285 loss 2.7 = 2.668 + 0.024 + 0.007 avg prob of [ goaltender] 0.07680393010377884 loss 1.774 = 1.715 + 0.049 + 0.01 avg prob of [ goaltender] 0.18725450336933136 loss 1.282 = 1.204 + 0.066 + 0.012 avg prob of [ goaltender] 0.30447620153427124 loss 0.973 = 0.892 + 0.067 + 0.014 avg prob of [ goaltender] 0.4127929210662842 loss 0.742 = 0.671 + 0.055 + 0.016 avg prob of [ goaltender] 0.5133483409881592 loss 0.567 = 0.506 + 0.043 + 0.017 avg prob of [ goaltender] 0.6047266721725464 loss 0.441 = 0.384 + 0.038 + 0.019 avg prob of [ goaltender] 0.6828424334526062 loss 0.35 = 0.294 + 0.036 + 0.02 avg prob of [ goaltender] 0.7465028166770935 loss 0.286 = 0.229 + 0.036 + 0.021 avg prob of [ goaltender] 0.7964061498641968 loss 0.241 = 0.182 + 0.037 + 0.022 avg prob of [ goaltender] 0.8348612785339355 loss 0.207 = 0.146 + 0.037 + 0.023 avg prob of [ goaltender] 0.8646178245544434 loss 0.181 = 0.12 + 0.037 + 0.024 avg prob of [ goaltender] 0.8879352807998657 loss 0.162 = 0.099 + 0.038 + 0.025 avg prob of [ goaltender] 0.9064284563064575 loss 0.146 = 0.082 + 0.037 + 0.026 avg prob of [ goaltender] 0.9212386608123779 loss 0.133 = 0.07 + 0.037 + 0.027 avg prob of [ goaltender] 0.9332011342048645 loss 0.123 = 0.059 + 0.037 + 0.027 avg prob of [ goaltender] 0.9429445862770081 loss 0.115 = 0.051 + 0.036 + 0.028 avg prob of [ goaltender] 0.9509485960006714 loss 0.108 = 0.044 + 0.035 + 0.029 avg prob of [ goaltender] 0.9575772285461426 Delta norm: 124.00716400146484 Change in target norm: 46.62088394165039 to 134.7905731201172 => 88.16969299316406 Division Factor: 12.021038055419922 Right vector norm: 10.315844535827637 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 3.046 = 3.046 + 0.0 + 0.0 avg prob of [ France] 0.09607571363449097 loss 0.189 = 0.178 + 0.007 + 0.004 avg prob of [ France] 0.8461291790008545 loss 0.18 = 0.153 + 0.02 + 0.007 avg prob of [ France] 0.8651211857795715 loss 0.173 = 0.135 + 0.028 + 0.01 avg prob of [ France] 0.8797589540481567 loss 0.161 = 0.117 + 0.032 + 0.011 avg prob of [ France] 0.8955858945846558 loss 0.147 = 0.1 + 0.034 + 0.013 avg prob of [ France] 0.9104770421981812 loss 0.133 = 0.084 + 0.034 + 0.014 avg prob of [ France] 0.9239455461502075 loss 0.119 = 0.07 + 0.034 + 0.016 avg prob of [ France] 0.9360364079475403 loss 0.107 = 0.057 + 0.033 + 0.017 avg prob of [ France] 0.946648359298706 loss 0.097 = 0.047 + 0.032 + 0.018 avg prob of [ France] 0.9556657075881958 loss 0.089 = 0.038 + 0.032 + 0.019 avg prob of [ France] 0.9631014466285706 loss 0.083 = 0.032 + 0.032 + 0.02 avg prob of [ France] 0.969113826751709 loss 0.079 = 0.027 + 0.032 + 0.02 avg prob of [ France] 0.9739381074905396 loss 0.075 = 0.023 + 0.032 + 0.021 avg prob of [ France] 0.9778144359588623 loss 0.072 = 0.019 + 0.031 + 0.022 avg prob of [ France] 0.9809526801109314 loss 0.07 = 0.017 + 0.031 + 0.022 avg prob of [ France] 0.9835184812545776 loss 0.068 = 0.015 + 0.03 + 0.023 avg prob of [ France] 0.9856383204460144 loss 0.066 = 0.013 + 0.03 + 0.024 avg prob of [ France] 0.9874064922332764 loss 0.064 = 0.011 + 0.029 + 0.024 avg prob of [ France] 0.9888944029808044 loss 0.063 = 0.01 + 0.029 + 0.024 avg prob of [ France] 0.9901547431945801 Delta norm: 111.71965789794922 Change in target norm: 47.78934860229492 to 120.35151672363281 => 72.56216430664062 Division Factor: 11.914831161499023 Right vector norm: 9.376520156860352 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 7.47 = 7.47 + 0.0 + 0.0 avg prob of [ Greek] 0.001604758552275598 loss 1.083 = 1.064 + 0.013 + 0.006 avg prob of [ Greek] 0.5387219786643982 loss 0.175 = 0.112 + 0.053 + 0.01 avg prob of [ Greek] 0.89697265625 loss 0.195 = 0.12 + 0.062 + 0.013 avg prob of [ Greek] 0.8902503252029419 loss 0.144 = 0.064 + 0.064 + 0.015 avg prob of [ Greek] 0.9390206336975098 loss 0.107 = 0.029 + 0.06 + 0.018 avg prob of [ Greek] 0.9720413684844971 loss 0.087 = 0.014 + 0.053 + 0.02 avg prob of [ Greek] 0.9864141345024109 loss 0.076 = 0.007 + 0.047 + 0.021 avg prob of [ Greek] 0.9925565123558044 loss 0.069 = 0.005 + 0.042 + 0.023 avg prob of [ Greek] 0.9953826069831848 loss 0.066 = 0.003 + 0.038 + 0.024 avg prob of [ Greek] 0.996798574924469 loss 0.064 = 0.002 + 0.036 + 0.026 avg prob of [ Greek] 0.9975616931915283 loss 0.062 = 0.002 + 0.033 + 0.027 avg prob of [ Greek] 0.9979937672615051 loss 0.061 = 0.002 + 0.032 + 0.028 avg prob of [ Greek] 0.9982424974441528 loss 0.061 = 0.002 + 0.031 + 0.029 avg prob of [ Greek] 0.9983809590339661 loss 0.061 = 0.002 + 0.03 + 0.029 avg prob of [ Greek] 0.9984463453292847 loss 0.061 = 0.002 + 0.029 + 0.03 avg prob of [ Greek] 0.9984604716300964 loss 0.061 = 0.002 + 0.028 + 0.031 avg prob of [ Greek] 0.9984391331672668 loss 0.061 = 0.002 + 0.028 + 0.032 avg prob of [ Greek] 0.9983954429626465 loss 0.061 = 0.002 + 0.027 + 0.032 avg prob of [ Greek] 0.9983407855033875 loss 0.061 = 0.002 + 0.026 + 0.033 avg prob of [ Greek] 0.9982849359512329 Delta norm: 113.674072265625 Change in target norm: 41.73223114013672 to 123.80826568603516 => 82.07603454589844 Division Factor: 10.345783233642578 Right vector norm: 10.987478256225586 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 12.642 = 12.642 + 0.0 + 0.0 avg prob of [ Bethlehem] 1.2424068700056523e-05 loss 8.228 = 8.21 + 0.002 + 0.016 avg prob of [ Bethlehem] 0.00034733559004962444 loss 6.39 = 6.357 + 0.004 + 0.028 avg prob of [ Bethlehem] 0.0021779441740363836 loss 4.39 = 4.345 + 0.007 + 0.038 avg prob of [ Bethlehem] 0.01769358292222023 loss 2.37 = 2.314 + 0.009 + 0.047 avg prob of [ Bethlehem] 0.12366174161434174 loss 0.769 = 0.701 + 0.012 + 0.056 avg prob of [ Bethlehem] 0.5326746106147766 loss 0.199 = 0.118 + 0.016 + 0.065 avg prob of [ Bethlehem] 0.8930103778839111 loss 0.115 = 0.02 + 0.022 + 0.073 avg prob of [ Bethlehem] 0.9800130128860474 loss 0.112 = 0.005 + 0.027 + 0.08 avg prob of [ Bethlehem] 0.9947955012321472 loss 0.111 = 0.003 + 0.028 + 0.08 avg prob of [ Bethlehem] 0.9973279237747192 loss 0.11 = 0.002 + 0.028 + 0.08 avg prob of [ Bethlehem] 0.9983903169631958 loss 0.109 = 0.001 + 0.027 + 0.08 avg prob of [ Bethlehem] 0.9989367723464966 loss 0.107 = 0.001 + 0.026 + 0.08 avg prob of [ Bethlehem] 0.9992460608482361 loss 0.106 = 0.001 + 0.025 + 0.08 avg prob of [ Bethlehem] 0.9994338154792786 loss 0.104 = 0.0 + 0.023 + 0.08 avg prob of [ Bethlehem] 0.9995543360710144 loss 0.102 = 0.0 + 0.022 + 0.08 avg prob of [ Bethlehem] 0.9996348023414612 loss 0.101 = 0.0 + 0.02 + 0.08 avg prob of [ Bethlehem] 0.99969083070755 loss 0.1 = 0.0 + 0.019 + 0.08 avg prob of [ Bethlehem] 0.9997308850288391 loss 0.098 = 0.0 + 0.018 + 0.08 avg prob of [ Bethlehem] 0.9997603893280029 loss 0.097 = 0.0 + 0.017 + 0.08 avg prob of [ Bethlehem] 0.9997824430465698 Delta norm: 99.48509216308594 Change in target norm: 24.871273040771484 to 103.66165161132812 => 78.79037475585938 Division Factor: 5.4251556396484375 Right vector norm: 18.337739944458008 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 6.164 = 6.164 + 0.0 + 0.0 avg prob of [ Paris] 0.003876294707879424 loss 1.3 = 1.291 + 0.005 + 0.005 avg prob of [ Paris] 0.32266268134117126 loss 0.438 = 0.423 + 0.008 + 0.007 avg prob of [ Paris] 0.6690314412117004 loss 0.253 = 0.233 + 0.01 + 0.01 avg prob of [ Paris] 0.800750195980072 loss 0.157 = 0.132 + 0.013 + 0.012 avg prob of [ Paris] 0.8802374005317688 loss 0.11 = 0.08 + 0.016 + 0.014 avg prob of [ Paris] 0.9247807860374451 loss 0.088 = 0.053 + 0.02 + 0.015 avg prob of [ Paris] 0.9487501978874207 loss 0.078 = 0.039 + 0.022 + 0.016 avg prob of [ Paris] 0.9616929888725281 loss 0.074 = 0.032 + 0.024 + 0.018 avg prob of [ Paris] 0.9689660668373108 loss 0.071 = 0.027 + 0.025 + 0.019 avg prob of [ Paris] 0.973328173160553 loss 0.069 = 0.024 + 0.025 + 0.02 avg prob of [ Paris] 0.9761826395988464 loss 0.067 = 0.022 + 0.024 + 0.021 avg prob of [ Paris] 0.9782465100288391 loss 0.065 = 0.02 + 0.023 + 0.021 avg prob of [ Paris] 0.9798873066902161 loss 0.063 = 0.019 + 0.022 + 0.022 avg prob of [ Paris] 0.9812890291213989 loss 0.061 = 0.018 + 0.021 + 0.023 avg prob of [ Paris] 0.9825420379638672 loss 0.06 = 0.016 + 0.02 + 0.023 avg prob of [ Paris] 0.9836888909339905 loss 0.058 = 0.015 + 0.019 + 0.024 avg prob of [ Paris] 0.9847474694252014 loss 0.057 = 0.014 + 0.018 + 0.024 avg prob of [ Paris] 0.9857257008552551 loss 0.056 = 0.013 + 0.018 + 0.025 avg prob of [ Paris] 0.9866262078285217 loss 0.055 = 0.013 + 0.017 + 0.025 avg prob of [ Paris] 0.9874523878097534 Delta norm: 110.11068725585938 Change in target norm: 46.86590576171875 to 124.74586486816406 => 77.87995910644531 Division Factor: 11.198709487915039 Right vector norm: 9.832444190979004 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 6.546 = 6.546 + 0.0 + 0.0 avg prob of [ Nissan] 0.0021350847091525793 loss 5.172 = 5.152 + 0.004 + 0.016 avg prob of [ Nissan] 0.007791806478053331 loss 2.715 = 2.674 + 0.014 + 0.027 avg prob of [ Nissan] 0.08895377814769745 loss 2.345 = 2.269 + 0.039 + 0.037 avg prob of [ Nissan] 0.14625386893749237 loss 0.395 = 0.321 + 0.031 + 0.043 avg prob of [ Nissan] 0.7483636140823364 loss 0.29 = 0.208 + 0.033 + 0.049 avg prob of [ Nissan] 0.8301700353622437 loss 0.21 = 0.117 + 0.039 + 0.054 avg prob of [ Nissan] 0.8986578583717346 loss 0.152 = 0.048 + 0.046 + 0.059 avg prob of [ Nissan] 0.9558266997337341 loss 0.134 = 0.02 + 0.051 + 0.064 avg prob of [ Nissan] 0.9808284044265747 loss 0.133 = 0.011 + 0.053 + 0.068 avg prob of [ Nissan] 0.9890146851539612 loss 0.135 = 0.009 + 0.054 + 0.073 avg prob of [ Nissan] 0.991478681564331 loss 0.137 = 0.008 + 0.053 + 0.076 avg prob of [ Nissan] 0.9923187494277954 loss 0.138 = 0.007 + 0.051 + 0.079 avg prob of [ Nissan] 0.9929077625274658 loss 0.135 = 0.006 + 0.048 + 0.081 avg prob of [ Nissan] 0.994051456451416 loss 0.13 = 0.005 + 0.045 + 0.081 avg prob of [ Nissan] 0.9952687621116638 loss 0.127 = 0.004 + 0.042 + 0.081 avg prob of [ Nissan] 0.9961362481117249 loss 0.124 = 0.003 + 0.04 + 0.081 avg prob of [ Nissan] 0.9967522621154785 loss 0.122 = 0.003 + 0.039 + 0.081 avg prob of [ Nissan] 0.9972068071365356 loss 0.12 = 0.002 + 0.037 + 0.081 avg prob of [ Nissan] 0.9975590705871582 loss 0.118 = 0.002 + 0.035 + 0.081 avg prob of [ Nissan] 0.9978434443473816 Delta norm: 99.04796600341797 Change in target norm: 24.761991500854492 to 103.39791870117188 => 78.63592529296875 Division Factor: 7.615337371826172 Right vector norm: 13.006380081176758 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 4.617 = 4.617 + 0.0 + 0.0 avg prob of [ Chicago] 0.012537561357021332 loss 0.884 = 0.873 + 0.005 + 0.006 avg prob of [ Chicago] 0.5033994913101196 loss 0.519 = 0.495 + 0.014 + 0.01 avg prob of [ Chicago] 0.6338038444519043 loss 0.219 = 0.182 + 0.024 + 0.013 avg prob of [ Chicago] 0.8401082158088684 loss 0.152 = 0.105 + 0.031 + 0.016 avg prob of [ Chicago] 0.902258038520813 loss 0.12 = 0.065 + 0.036 + 0.018 avg prob of [ Chicago] 0.9379170536994934 loss 0.106 = 0.044 + 0.041 + 0.021 avg prob of [ Chicago] 0.9569947123527527 loss 0.099 = 0.033 + 0.043 + 0.022 avg prob of [ Chicago] 0.967383623123169 loss 0.094 = 0.027 + 0.043 + 0.024 avg prob of [ Chicago] 0.9735086560249329 loss 0.09 = 0.023 + 0.042 + 0.025 avg prob of [ Chicago] 0.9774082899093628 loss 0.088 = 0.02 + 0.041 + 0.027 avg prob of [ Chicago] 0.9800912737846375 loss 0.086 = 0.018 + 0.04 + 0.028 avg prob of [ Chicago] 0.9820972681045532 loss 0.084 = 0.016 + 0.039 + 0.029 avg prob of [ Chicago] 0.9837183952331543 loss 0.083 = 0.015 + 0.038 + 0.03 avg prob of [ Chicago] 0.9851096272468567 loss 0.081 = 0.014 + 0.036 + 0.031 avg prob of [ Chicago] 0.986348569393158 loss 0.079 = 0.013 + 0.035 + 0.032 avg prob of [ Chicago] 0.9874735474586487 loss 0.077 = 0.012 + 0.033 + 0.032 avg prob of [ Chicago] 0.988501787185669 loss 0.075 = 0.011 + 0.032 + 0.033 avg prob of [ Chicago] 0.9894418120384216 loss 0.073 = 0.01 + 0.03 + 0.033 avg prob of [ Chicago] 0.9902985095977783 loss 0.071 = 0.009 + 0.028 + 0.034 avg prob of [ Chicago] 0.9910759925842285 Delta norm: 107.55545043945312 Change in target norm: 39.78049087524414 to 117.18111419677734 => 77.40061950683594 Division Factor: 10.085493087768555 Right vector norm: 10.664371490478516 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 9.08 = 9.08 + 0.0 + 0.0 avg prob of [ Italy] 0.00016603512631263584 loss 5.936 = 5.922 + 0.007 + 0.007 avg prob of [ Italy] 0.004691421054303646 loss 3.79 = 3.759 + 0.02 + 0.011 avg prob of [ Italy] 0.03109513595700264 loss 1.781 = 1.726 + 0.041 + 0.015 avg prob of [ Italy] 0.21593354642391205 loss 0.464 = 0.369 + 0.077 + 0.018 avg prob of [ Italy] 0.7294302582740784 loss 0.215 = 0.099 + 0.095 + 0.022 avg prob of [ Italy] 0.9147151708602905 loss 0.156 = 0.041 + 0.09 + 0.025 avg prob of [ Italy] 0.9614545106887817 loss 0.129 = 0.021 + 0.081 + 0.027 avg prob of [ Italy] 0.9797991514205933 loss 0.114 = 0.013 + 0.071 + 0.03 avg prob of [ Italy] 0.9870860576629639 loss 0.108 = 0.01 + 0.066 + 0.032 avg prob of [ Italy] 0.9901017546653748 loss 0.107 = 0.009 + 0.064 + 0.034 avg prob of [ Italy] 0.9914460182189941 loss 0.107 = 0.008 + 0.063 + 0.036 avg prob of [ Italy] 0.9920742511749268 loss 0.107 = 0.008 + 0.062 + 0.037 avg prob of [ Italy] 0.9923662543296814 loss 0.107 = 0.008 + 0.061 + 0.039 avg prob of [ Italy] 0.9924912452697754 loss 0.106 = 0.008 + 0.059 + 0.04 avg prob of [ Italy] 0.9925348162651062 loss 0.106 = 0.008 + 0.057 + 0.041 avg prob of [ Italy] 0.9925455451011658 loss 0.105 = 0.008 + 0.055 + 0.042 avg prob of [ Italy] 0.9925538897514343 loss 0.104 = 0.007 + 0.054 + 0.043 avg prob of [ Italy] 0.9925797581672668 loss 0.104 = 0.007 + 0.052 + 0.044 avg prob of [ Italy] 0.9926356077194214 loss 0.103 = 0.007 + 0.051 + 0.045 avg prob of [ Italy] 0.9927279353141785 Delta norm: 131.89866638183594 Change in target norm: 38.283199310302734 to 140.4503936767578 => 102.16719055175781 Division Factor: 9.958413124084473 Right vector norm: 13.244948387145996 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 11.288 = 11.288 + 0.0 + 0.0 avg prob of [ Dublin] 1.8331116734771058e-05 loss 7.634 = 7.616 + 0.003 + 0.015 avg prob of [ Dublin] 0.0005725047667510808 loss 4.735 = 4.705 + 0.005 + 0.024 avg prob of [ Dublin] 0.011409574188292027 loss 1.832 = 1.786 + 0.015 + 0.032 avg prob of [ Dublin] 0.1952315717935562 loss 1.363 = 1.307 + 0.016 + 0.04 avg prob of [ Dublin] 0.30548346042633057 loss 0.938 = 0.875 + 0.016 + 0.046 avg prob of [ Dublin] 0.44371190667152405 loss 0.611 = 0.542 + 0.017 + 0.053 avg prob of [ Dublin] 0.5933716297149658 loss 0.421 = 0.345 + 0.018 + 0.058 avg prob of [ Dublin] 0.7119237780570984 loss 0.32 = 0.237 + 0.019 + 0.064 avg prob of [ Dublin] 0.7904089093208313 loss 0.262 = 0.171 + 0.022 + 0.068 avg prob of [ Dublin] 0.8434441685676575 loss 0.224 = 0.127 + 0.025 + 0.073 avg prob of [ Dublin] 0.881395161151886 loss 0.199 = 0.095 + 0.027 + 0.076 avg prob of [ Dublin] 0.9093157649040222 loss 0.177 = 0.073 + 0.028 + 0.076 avg prob of [ Dublin] 0.9297932982444763 loss 0.161 = 0.057 + 0.028 + 0.076 avg prob of [ Dublin] 0.9450914263725281 loss 0.148 = 0.044 + 0.027 + 0.076 avg prob of [ Dublin] 0.9565572142601013 loss 0.138 = 0.035 + 0.026 + 0.076 avg prob of [ Dublin] 0.9652006030082703 loss 0.13 = 0.029 + 0.025 + 0.076 avg prob of [ Dublin] 0.971761167049408 loss 0.123 = 0.024 + 0.023 + 0.076 avg prob of [ Dublin] 0.9767798185348511 loss 0.117 = 0.02 + 0.022 + 0.076 avg prob of [ Dublin] 0.9806520342826843 loss 0.113 = 0.016 + 0.02 + 0.076 avg prob of [ Dublin] 0.9836675524711609 Delta norm: 105.00497436523438 Change in target norm: 26.251243591308594 to 109.81632995605469 => 83.5650863647461 Division Factor: 9.388671875 Right vector norm: 11.184220314025879 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 9.446 = 9.446 + 0.0 + 0.0 avg prob of [ French] 9.242963278666139e-05 loss 3.571 = 3.564 + 0.004 + 0.004 avg prob of [ French] 0.03429456427693367 loss 1.5 = 1.472 + 0.021 + 0.007 avg prob of [ French] 0.24239292740821838 loss 0.731 = 0.702 + 0.019 + 0.009 avg prob of [ French] 0.5029215812683105 loss 0.45 = 0.422 + 0.017 + 0.011 avg prob of [ French] 0.6611031889915466 loss 0.303 = 0.271 + 0.019 + 0.013 avg prob of [ French] 0.7654753923416138 loss 0.217 = 0.181 + 0.021 + 0.014 avg prob of [ French] 0.8360378742218018 loss 0.163 = 0.126 + 0.022 + 0.016 avg prob of [ French] 0.8827053308486938 loss 0.128 = 0.091 + 0.02 + 0.017 avg prob of [ French] 0.9135494232177734 loss 0.105 = 0.068 + 0.018 + 0.018 avg prob of [ French] 0.9343615174293518 loss 0.089 = 0.053 + 0.017 + 0.019 avg prob of [ French] 0.9487651586532593 loss 0.078 = 0.042 + 0.016 + 0.02 avg prob of [ French] 0.9589715003967285 loss 0.072 = 0.034 + 0.016 + 0.021 avg prob of [ French] 0.9663756489753723 loss 0.067 = 0.029 + 0.017 + 0.022 avg prob of [ French] 0.9718812108039856 loss 0.065 = 0.024 + 0.018 + 0.023 avg prob of [ French] 0.9760791063308716 loss 0.063 = 0.021 + 0.018 + 0.023 avg prob of [ French] 0.9793556928634644 loss 0.061 = 0.018 + 0.019 + 0.024 avg prob of [ French] 0.9819658994674683 loss 0.06 = 0.016 + 0.02 + 0.025 avg prob of [ French] 0.9840809106826782 loss 0.059 = 0.014 + 0.02 + 0.025 avg prob of [ French] 0.9858189821243286 loss 0.058 = 0.013 + 0.02 + 0.026 avg prob of [ French] 0.9872633814811707 Delta norm: 129.91732788085938 Change in target norm: 50.455894470214844 to 139.9485321044922 => 89.49263763427734 Division Factor: 8.192535400390625 Right vector norm: 15.858014106750488 seq_released {"n_edits": 90, "retention": 1.0, "unedited_true": 0.8, "max_update_norm": 72.61759185791016, "median_update_norm": 9.873079299926758} (2532s) 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 8.19 = 8.19 + 0.0 + 0.0 avg prob of [ Milan] 0.0004928613197989762 loss 4.946 = 4.936 + 0.001 + 0.01 avg prob of [ Milan] 0.015010321512818336 loss 2.293 = 2.276 + 0.002 + 0.015 avg prob of [ Milan] 0.20212328433990479 loss 1.081 = 1.055 + 0.006 + 0.02 avg prob of [ Milan] 0.5176334381103516 loss 0.537 = 0.499 + 0.013 + 0.025 avg prob of [ Milan] 0.7931135892868042 loss 0.336 = 0.285 + 0.023 + 0.029 avg prob of [ Milan] 0.9068118333816528 loss 0.249 = 0.183 + 0.033 + 0.033 avg prob of [ Milan] 0.937620997428894 loss 0.198 = 0.119 + 0.043 + 0.037 avg prob of [ Milan] 0.9492266774177551 loss 0.163 = 0.072 + 0.051 + 0.04 avg prob of [ Milan] 0.959230899810791 loss 0.141 = 0.039 + 0.059 + 0.043 avg prob of [ Milan] 0.9714152216911316 loss 0.13 = 0.02 + 0.065 + 0.045 avg prob of [ Milan] 0.9827048778533936 loss 0.127 = 0.011 + 0.069 + 0.048 avg prob of [ Milan] 0.9898357391357422 loss 0.126 = 0.006 + 0.07 + 0.05 avg prob of [ Milan] 0.9937662482261658 loss 0.124 = 0.004 + 0.069 + 0.052 avg prob of [ Milan] 0.9960554838180542 loss 0.123 = 0.003 + 0.067 + 0.053 avg prob of [ Milan] 0.9973477125167847 loss 0.121 = 0.002 + 0.064 + 0.055 avg prob of [ Milan] 0.9980815649032593 loss 0.119 = 0.001 + 0.061 + 0.056 avg prob of [ Milan] 0.9985224604606628 loss 0.117 = 0.001 + 0.058 + 0.058 avg prob of [ Milan] 0.9988053441047668 loss 0.115 = 0.001 + 0.055 + 0.059 avg prob of [ Milan] 0.9989973902702332 loss 0.113 = 0.001 + 0.052 + 0.06 avg prob of [ Milan] 0.9991341829299927 Delta norm: 125.36557006835938 Change in target norm: 32.37807846069336 to 130.52125549316406 => 98.14317321777344 Division Factor: 6.817209243774414 Right vector norm: 18.389572143554688 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 4.998 = 4.998 + 0.0 + 0.0 avg prob of [ Islam] 0.007247906178236008 loss 1.249 = 1.237 + 0.001 + 0.01 avg prob of [ Islam] 0.31009283661842346 loss 0.31 = 0.287 + 0.005 + 0.018 avg prob of [ Islam] 0.7543907761573792 loss 0.356 = 0.323 + 0.008 + 0.025 avg prob of [ Islam] 0.7273676991462708 loss 0.349 = 0.309 + 0.011 + 0.03 avg prob of [ Islam] 0.7380382418632507 loss 0.314 = 0.268 + 0.012 + 0.034 avg prob of [ Islam] 0.7685642242431641 loss 0.273 = 0.222 + 0.013 + 0.038 avg prob of [ Islam] 0.804134726524353 loss 0.235 = 0.179 + 0.014 + 0.042 avg prob of [ Islam] 0.8383058905601501 loss 0.203 = 0.143 + 0.014 + 0.045 avg prob of [ Islam] 0.8685193657875061 loss 0.177 = 0.114 + 0.015 + 0.048 avg prob of [ Islam] 0.8940678834915161 loss 0.156 = 0.09 + 0.015 + 0.051 avg prob of [ Islam] 0.9150952100753784 loss 0.14 = 0.071 + 0.016 + 0.053 avg prob of [ Islam] 0.932098388671875 loss 0.128 = 0.056 + 0.016 + 0.056 avg prob of [ Islam] 0.9456823468208313 loss 0.119 = 0.045 + 0.016 + 0.058 avg prob of [ Islam] 0.9564449787139893 loss 0.112 = 0.036 + 0.016 + 0.06 avg prob of [ Islam] 0.9649245738983154 loss 0.107 = 0.029 + 0.016 + 0.062 avg prob of [ Islam] 0.9715805053710938 loss 0.103 = 0.024 + 0.016 + 0.063 avg prob of [ Islam] 0.976793646812439 loss 0.099 = 0.02 + 0.015 + 0.063 avg prob of [ Islam] 0.9803855419158936 loss 0.095 = 0.017 + 0.015 + 0.063 avg prob of [ Islam] 0.9832383990287781 loss 0.092 = 0.015 + 0.014 + 0.063 avg prob of [ Islam] 0.9855483770370483 Delta norm: 126.34683990478516 Change in target norm: 31.58670997619629 to 130.01895141601562 => 98.43224334716797 Division Factor: 10.856204986572266 Right vector norm: 11.638214111328125 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 10.177 = 10.177 + 0.0 + 0.0 avg prob of [ Spanish] 4.122341488255188e-05 loss 7.424 = 7.413 + 0.006 + 0.006 avg prob of [ Spanish] 0.0006879510474391282 loss 5.288 = 5.266 + 0.012 + 0.009 avg prob of [ Spanish] 0.006026213523000479 loss 2.847 = 2.817 + 0.017 + 0.013 avg prob of [ Spanish] 0.06594916433095932 loss 1.015 = 0.978 + 0.021 + 0.016 avg prob of [ Spanish] 0.3859206736087799 loss 0.314 = 0.27 + 0.025 + 0.019 avg prob of [ Spanish] 0.7654105424880981 loss 0.16 = 0.11 + 0.029 + 0.022 avg prob of [ Spanish] 0.8963793516159058 loss 0.124 = 0.068 + 0.032 + 0.025 avg prob of [ Spanish] 0.9345881938934326 loss 0.116 = 0.055 + 0.035 + 0.027 avg prob of [ Spanish] 0.9470584392547607 loss 0.111 = 0.045 + 0.037 + 0.029 avg prob of [ Spanish] 0.9558842778205872 loss 0.105 = 0.035 + 0.039 + 0.031 avg prob of [ Spanish] 0.9657394886016846 loss 0.099 = 0.026 + 0.04 + 0.033 avg prob of [ Spanish] 0.9744536876678467 loss 0.096 = 0.019 + 0.042 + 0.035 avg prob of [ Spanish] 0.9808424711227417 loss 0.094 = 0.015 + 0.043 + 0.036 avg prob of [ Spanish] 0.9852263927459717 loss 0.093 = 0.012 + 0.044 + 0.037 avg prob of [ Spanish] 0.9882292151451111 loss 0.093 = 0.01 + 0.045 + 0.038 avg prob of [ Spanish] 0.9903345704078674 loss 0.092 = 0.008 + 0.045 + 0.04 avg prob of [ Spanish] 0.9918563365936279 loss 0.092 = 0.007 + 0.044 + 0.041 avg prob of [ Spanish] 0.9929890632629395 loss 0.091 = 0.006 + 0.044 + 0.041 avg prob of [ Spanish] 0.9938561320304871 loss 0.091 = 0.005 + 0.043 + 0.042 avg prob of [ Spanish] 0.9945355653762817 Delta norm: 149.64088439941406 Change in target norm: 42.0899543762207 to 162.11656188964844 => 120.026611328125 Division Factor: 11.848564147949219 Right vector norm: 12.629454612731934 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 8.908 = 8.908 + 0.0 + 0.0 avg prob of [ Asia] 0.00035160366678610444 loss 3.689 = 3.676 + 0.003 + 0.01 avg prob of [ Asia] 0.03425021842122078 loss 1.677 = 1.654 + 0.006 + 0.016 avg prob of [ Asia] 0.2612590491771698 loss 0.805 = 0.774 + 0.009 + 0.022 avg prob of [ Asia] 0.5879904627799988 loss 0.581 = 0.542 + 0.012 + 0.026 avg prob of [ Asia] 0.7007333040237427 loss 0.448 = 0.401 + 0.016 + 0.03 avg prob of [ Asia] 0.7645646929740906 loss 0.346 = 0.293 + 0.019 + 0.034 avg prob of [ Asia] 0.8171690106391907 loss 0.274 = 0.214 + 0.022 + 0.037 avg prob of [ Asia] 0.8583908081054688 loss 0.225 = 0.16 + 0.025 + 0.04 avg prob of [ Asia] 0.8882019519805908 loss 0.191 = 0.121 + 0.027 + 0.043 avg prob of [ Asia] 0.9097689390182495 loss 0.167 = 0.092 + 0.029 + 0.045 avg prob of [ Asia] 0.9262585639953613 loss 0.15 = 0.072 + 0.031 + 0.047 avg prob of [ Asia] 0.939411461353302 loss 0.139 = 0.057 + 0.033 + 0.049 avg prob of [ Asia] 0.949937105178833 loss 0.131 = 0.046 + 0.034 + 0.051 avg prob of [ Asia] 0.9582330584526062 loss 0.126 = 0.038 + 0.035 + 0.053 avg prob of [ Asia] 0.9647038578987122 loss 0.122 = 0.032 + 0.036 + 0.054 avg prob of [ Asia] 0.9697539806365967 loss 0.119 = 0.027 + 0.036 + 0.055 avg prob of [ Asia] 0.9737261533737183 loss 0.117 = 0.024 + 0.036 + 0.057 avg prob of [ Asia] 0.9768860340118408 loss 0.115 = 0.021 + 0.036 + 0.058 avg prob of [ Asia] 0.9794312715530396 loss 0.114 = 0.019 + 0.036 + 0.059 avg prob of [ Asia] 0.9815077781677246 Delta norm: 116.73371124267578 Change in target norm: 31.488460540771484 to 125.32312774658203 => 93.83467102050781 Division Factor: 6.820245742797852 Right vector norm: 17.11576271057129 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 12.793 = 12.793 + 0.0 + 0.0 avg prob of [ Finnish] 8.235688255808782e-06 loss 10.013 = 9.995 + 0.002 + 0.017 avg prob of [ Finnish] 0.00010643024870660156 loss 6.856 = 6.819 + 0.007 + 0.03 avg prob of [ Finnish] 0.0016385744092985988 loss 4.356 = 4.3 + 0.014 + 0.042 avg prob of [ Finnish] 0.016466494649648666 loss 2.425 = 2.351 + 0.021 + 0.054 avg prob of [ Finnish] 0.11143313348293304 loss 0.945 = 0.854 + 0.027 + 0.064 avg prob of [ Finnish] 0.44865721464157104 loss 0.339 = 0.233 + 0.032 + 0.074 avg prob of [ Finnish] 0.8007500171661377 loss 0.18 = 0.062 + 0.036 + 0.082 avg prob of [ Finnish] 0.9407485723495483 loss 0.143 = 0.025 + 0.036 + 0.082 avg prob of [ Finnish] 0.9754460453987122 loss 0.129 = 0.012 + 0.035 + 0.082 avg prob of [ Finnish] 0.988369882106781 loss 0.123 = 0.007 + 0.035 + 0.082 avg prob of [ Finnish] 0.9935103058815002 loss 0.12 = 0.004 + 0.034 + 0.082 avg prob of [ Finnish] 0.9958828687667847 loss 0.118 = 0.003 + 0.033 + 0.082 avg prob of [ Finnish] 0.9971228241920471 loss 0.116 = 0.002 + 0.032 + 0.082 avg prob of [ Finnish] 0.9978400468826294 loss 0.115 = 0.002 + 0.031 + 0.082 avg prob of [ Finnish] 0.9982897639274597 loss 0.113 = 0.001 + 0.03 + 0.082 avg prob of [ Finnish] 0.9985907673835754 loss 0.112 = 0.001 + 0.029 + 0.082 avg prob of [ Finnish] 0.9988032579421997 loss 0.111 = 0.001 + 0.028 + 0.082 avg prob of [ Finnish] 0.9989594221115112 loss 0.11 = 0.001 + 0.027 + 0.082 avg prob of [ Finnish] 0.9990784525871277 loss 0.109 = 0.001 + 0.026 + 0.082 avg prob of [ Finnish] 0.9991716742515564 Delta norm: 97.87122344970703 Change in target norm: 24.46780776977539 to 100.22090148925781 => 75.75309753417969 Division Factor: 7.282918930053711 Right vector norm: 13.438461303710938 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 10.017 = 10.017 + 0.0 + 0.0 avg prob of [ Dodge] 5.910419349675067e-05 loss 5.516 = 5.508 + 0.004 + 0.004 avg prob of [ Dodge] 0.005621218588203192 loss 3.449 = 3.43 + 0.012 + 0.007 avg prob of [ Dodge] 0.03327992558479309 loss 2.046 = 2.014 + 0.021 + 0.01 avg prob of [ Dodge] 0.1365051567554474 loss 0.709 = 0.664 + 0.033 + 0.012 avg prob of [ Dodge] 0.5222735404968262 loss 0.187 = 0.129 + 0.043 + 0.015 avg prob of [ Dodge] 0.8809620141983032 loss 0.107 = 0.043 + 0.048 + 0.017 avg prob of [ Dodge] 0.9587575197219849 loss 0.093 = 0.023 + 0.051 + 0.019 avg prob of [ Dodge] 0.9775381684303284 loss 0.09 = 0.016 + 0.054 + 0.02 avg prob of [ Dodge] 0.9847286939620972 loss 0.09 = 0.012 + 0.057 + 0.022 avg prob of [ Dodge] 0.9881070256233215 loss 0.092 = 0.01 + 0.058 + 0.023 avg prob of [ Dodge] 0.989989161491394 loss 0.092 = 0.009 + 0.059 + 0.024 avg prob of [ Dodge] 0.9913784265518188 loss 0.092 = 0.007 + 0.059 + 0.025 avg prob of [ Dodge] 0.9926298260688782 loss 0.091 = 0.006 + 0.058 + 0.026 avg prob of [ Dodge] 0.9938132166862488 loss 0.089 = 0.005 + 0.056 + 0.027 avg prob of [ Dodge] 0.9949039220809937 loss 0.086 = 0.004 + 0.054 + 0.028 avg prob of [ Dodge] 0.9958658814430237 loss 0.084 = 0.003 + 0.051 + 0.029 avg prob of [ Dodge] 0.9966788291931152 loss 0.081 = 0.003 + 0.049 + 0.03 avg prob of [ Dodge] 0.9973427653312683 loss 0.079 = 0.002 + 0.047 + 0.03 avg prob of [ Dodge] 0.9978713989257812 loss 0.078 = 0.002 + 0.045 + 0.031 avg prob of [ Dodge] 0.9982851147651672 Delta norm: 144.8577880859375 Change in target norm: 48.50252151489258 to 158.53231811523438 => 110.02980041503906 Division Factor: 9.01163101196289 Right vector norm: 16.074535369873047 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 4.826 = 4.826 + 0.0 + 0.0 avg prob of [ NBC] 0.00930107943713665 loss 2.477 = 2.466 + 0.004 + 0.006 avg prob of [ NBC] 0.10615311563014984 loss 0.754 = 0.727 + 0.016 + 0.011 avg prob of [ NBC] 0.49280503392219543 loss 0.349 = 0.31 + 0.025 + 0.015 avg prob of [ NBC] 0.742483913898468 loss 0.145 = 0.097 + 0.03 + 0.018 avg prob of [ NBC] 0.9094970226287842 loss 0.083 = 0.03 + 0.032 + 0.021 avg prob of [ NBC] 0.970689058303833 loss 0.07 = 0.012 + 0.035 + 0.024 avg prob of [ NBC] 0.9884823560714722 loss 0.07 = 0.006 + 0.038 + 0.027 avg prob of [ NBC] 0.9941306114196777 loss 0.074 = 0.004 + 0.041 + 0.029 avg prob of [ NBC] 0.9963411688804626 loss 0.076 = 0.003 + 0.043 + 0.031 avg prob of [ NBC] 0.9973880052566528 loss 0.078 = 0.002 + 0.043 + 0.032 avg prob of [ NBC] 0.9979566931724548 loss 0.078 = 0.002 + 0.042 + 0.034 avg prob of [ NBC] 0.998298168182373 loss 0.077 = 0.001 + 0.04 + 0.035 avg prob of [ NBC] 0.99852055311203 loss 0.076 = 0.001 + 0.038 + 0.037 avg prob of [ NBC] 0.9986764788627625 loss 0.076 = 0.001 + 0.037 + 0.038 avg prob of [ NBC] 0.9987941980361938 loss 0.076 = 0.001 + 0.036 + 0.039 avg prob of [ NBC] 0.9988894462585449 loss 0.076 = 0.001 + 0.036 + 0.04 avg prob of [ NBC] 0.9989714026451111 loss 0.076 = 0.001 + 0.035 + 0.041 avg prob of [ NBC] 0.9990445971488953 loss 0.076 = 0.001 + 0.034 + 0.041 avg prob of [ NBC] 0.9991113543510437 loss 0.076 = 0.001 + 0.033 + 0.042 avg prob of [ NBC] 0.9991723299026489 Delta norm: 129.39268493652344 Change in target norm: 39.2604866027832 to 139.37765502929688 => 100.11717224121094 Division Factor: 6.96501350402832 Right vector norm: 18.577518463134766 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 12.868 = 12.868 + 0.0 + 0.0 avg prob of [ England] 2.2375714252120815e-05 loss 5.42 = 5.408 + 0.004 + 0.008 avg prob of [ England] 0.009170238859951496 loss 1.014 = 0.985 + 0.018 + 0.012 avg prob of [ England] 0.44288069009780884 loss 0.505 = 0.453 + 0.036 + 0.016 avg prob of [ England] 0.6930747032165527 loss 0.354 = 0.293 + 0.041 + 0.02 avg prob of [ England] 0.7893989086151123 loss 0.274 = 0.207 + 0.045 + 0.023 avg prob of [ England] 0.8402682542800903 loss 0.212 = 0.139 + 0.048 + 0.025 avg prob of [ England] 0.8800713419914246 loss 0.174 = 0.095 + 0.051 + 0.028 avg prob of [ England] 0.9118166565895081 loss 0.158 = 0.074 + 0.054 + 0.03 avg prob of [ England] 0.930151641368866 loss 0.149 = 0.06 + 0.056 + 0.032 avg prob of [ England] 0.9424823522567749 loss 0.142 = 0.05 + 0.058 + 0.034 avg prob of [ England] 0.9517309069633484 loss 0.136 = 0.042 + 0.058 + 0.036 avg prob of [ England] 0.9590321779251099 loss 0.13 = 0.036 + 0.057 + 0.037 avg prob of [ England] 0.9649434089660645 loss 0.125 = 0.031 + 0.056 + 0.038 avg prob of [ England] 0.9697918891906738 loss 0.12 = 0.027 + 0.054 + 0.04 avg prob of [ England] 0.973798394203186 loss 0.115 = 0.023 + 0.051 + 0.041 avg prob of [ England] 0.9771265387535095 loss 0.11 = 0.02 + 0.048 + 0.042 avg prob of [ England] 0.9799038767814636 loss 0.106 = 0.018 + 0.046 + 0.043 avg prob of [ England] 0.982232391834259 loss 0.102 = 0.016 + 0.043 + 0.043 avg prob of [ England] 0.9841945171356201 loss 0.099 = 0.014 + 0.04 + 0.044 avg prob of [ England] 0.9858564734458923 Delta norm: 114.44085693359375 Change in target norm: 35.9973258972168 to 121.3570785522461 => 85.35975646972656 Division Factor: 10.05752182006836 Right vector norm: 11.378632545471191 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 4.653 = 4.653 + 0.0 + 0.0 avg prob of [ Islam] 0.014245958998799324 loss 1.743 = 1.732 + 0.001 + 0.01 avg prob of [ Islam] 0.19371573626995087 loss 0.601 = 0.582 + 0.003 + 0.016 avg prob of [ Islam] 0.5758236646652222 loss 0.226 = 0.201 + 0.004 + 0.022 avg prob of [ Islam] 0.8247491121292114 loss 0.127 = 0.094 + 0.006 + 0.027 avg prob of [ Islam] 0.9117188453674316 loss 0.093 = 0.054 + 0.008 + 0.031 avg prob of [ Islam] 0.9482773542404175 loss 0.08 = 0.035 + 0.01 + 0.035 avg prob of [ Islam] 0.9656699895858765 loss 0.076 = 0.026 + 0.013 + 0.038 avg prob of [ Islam] 0.9748733043670654 loss 0.076 = 0.02 + 0.014 + 0.041 avg prob of [ Islam] 0.9802846312522888 loss 0.076 = 0.016 + 0.016 + 0.044 avg prob of [ Islam] 0.9837634563446045 loss 0.077 = 0.014 + 0.016 + 0.046 avg prob of [ Islam] 0.9861639738082886 loss 0.077 = 0.012 + 0.016 + 0.049 avg prob of [ Islam] 0.987913966178894 loss 0.078 = 0.011 + 0.016 + 0.051 avg prob of [ Islam] 0.9892445802688599 loss 0.078 = 0.01 + 0.016 + 0.052 avg prob of [ Islam] 0.9902891516685486 loss 0.078 = 0.009 + 0.015 + 0.054 avg prob of [ Islam] 0.9911292791366577 loss 0.078 = 0.008 + 0.014 + 0.055 avg prob of [ Islam] 0.9918181300163269 loss 0.078 = 0.008 + 0.014 + 0.057 avg prob of [ Islam] 0.9923924207687378 loss 0.078 = 0.007 + 0.013 + 0.058 avg prob of [ Islam] 0.9928789734840393 loss 0.078 = 0.007 + 0.013 + 0.059 avg prob of [ Islam] 0.9932973980903625 loss 0.078 = 0.006 + 0.012 + 0.06 avg prob of [ Islam] 0.9936632513999939 Delta norm: 122.68952941894531 Change in target norm: 32.05752182006836 to 129.82797241210938 => 97.77044677734375 Division Factor: 8.645427703857422 Right vector norm: 14.191261291503906 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 9.734 = 9.734 + 0.0 + 0.0 avg prob of [ Berlin] 8.519218681612983e-05 loss 4.526 = 4.507 + 0.01 + 0.009 avg prob of [ Berlin] 0.02084771730005741 loss 1.541 = 1.476 + 0.049 + 0.016 avg prob of [ Berlin] 0.2782350182533264 loss 0.41 = 0.322 + 0.066 + 0.021 avg prob of [ Berlin] 0.7597684860229492 loss 0.206 = 0.108 + 0.072 + 0.026 avg prob of [ Berlin] 0.9192303419113159 loss 0.167 = 0.049 + 0.087 + 0.03 avg prob of [ Berlin] 0.9603118896484375 loss 0.159 = 0.024 + 0.101 + 0.034 avg prob of [ Berlin] 0.9788834452629089 loss 0.152 = 0.014 + 0.1 + 0.037 avg prob of [ Berlin] 0.986860990524292 loss 0.141 = 0.013 + 0.087 + 0.04 avg prob of [ Berlin] 0.987777590751648 loss 0.13 = 0.014 + 0.073 + 0.043 avg prob of [ Berlin] 0.9875300526618958 loss 0.122 = 0.013 + 0.063 + 0.045 avg prob of [ Berlin] 0.9876448512077332 loss 0.118 = 0.012 + 0.058 + 0.048 avg prob of [ Berlin] 0.9884898066520691 loss 0.115 = 0.011 + 0.055 + 0.05 avg prob of [ Berlin] 0.9899019598960876 loss 0.112 = 0.009 + 0.052 + 0.051 avg prob of [ Berlin] 0.9915081262588501 loss 0.11 = 0.007 + 0.049 + 0.053 avg prob of [ Berlin] 0.9929916858673096 loss 0.108 = 0.006 + 0.047 + 0.054 avg prob of [ Berlin] 0.9942147135734558 loss 0.106 = 0.005 + 0.045 + 0.056 avg prob of [ Berlin] 0.9951433539390564 loss 0.105 = 0.004 + 0.043 + 0.057 avg prob of [ Berlin] 0.9958220720291138 loss 0.104 = 0.004 + 0.042 + 0.058 avg prob of [ Berlin] 0.9963247179985046 loss 0.104 = 0.003 + 0.041 + 0.059 avg prob of [ Berlin] 0.9967083930969238 Delta norm: 128.66087341308594 Change in target norm: 32.98694610595703 to 137.03465270996094 => 104.0477066040039 Division Factor: 9.282645225524902 Right vector norm: 13.860367774963379 seq_released {"n_edits": 100, "retention": 1.0, "unedited_true": 0.8, "max_update_norm": 72.61759185791016, "median_update_norm": 10.254530429840088} (2805s) total seconds 2804.6