Loading the LM will be faster if you build a binary file. Reading ../standard/nc-wmt11.en.srilm.gz ----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100 **************************************************************************************************** Example feature: Shape_S00000_T00000 dtrain Parameters: k 100 N 4 T 1 learning rate 0.0001 margin 1 l1 reg 0 decoder conf 'cdec.ini' input 'work/shard.0.0' output 'work/weights.0.1' weights in 'work/weights.0' (1 dot per processed input) Iteration #1 of 1. ... 3 WEIGHTS Glue = +0.08417 WordPenalty = +0.1674 LanguageModel = +0.5448 LanguageModel_OOV = -0.6311 PhraseModel_0 = -0.2931 PhraseModel_1 = +1.173 PhraseModel_2 = +1.369 PhraseModel_3 = +0.1819 PhraseModel_4 = -0.8439 PhraseModel_5 = +0.02055 PhraseModel_6 = -0.3794 PassThrough = -0.6827 --- 1best avg score: 16.83 (+16.83) 1best avg model score: +66.56 avg # updates: 2090 non-0 feature count: 2218 avg f count: 142.5 avg list sz: 100 (time 0.1833 min, 3 s/S) --- Best iteration: 1 [GOLD = 16.83]. This took 0.1833 min.