| daza | 8534747 | 2020-11-23 18:43:33 +0100 | [diff] [blame] | 1 | [paths] |
| daza | d7d7075 | 2021-01-12 18:17:49 +0100 | [diff] [blame] | 2 | train = "/vol/netapp/daza/datasets/TIGER_conll/data_splits/train/Tiger.ALL.Orth.train.spacy" |
| 3 | dev = "/vol/netapp/daza/datasets/TIGER_conll/data_splits/test/Tiger.NewOrth.test.spacy" |
| daza | 8534747 | 2020-11-23 18:43:33 +0100 | [diff] [blame] | 4 | vectors = null |
| 5 | init_tok2vec = null |
| 6 | |
| 7 | [system] |
| 8 | gpu_allocator = "pytorch" |
| 9 | seed = 0 |
| 10 | |
| 11 | [nlp] |
| 12 | lang = "de" |
| 13 | pipeline = ["transformer","tagger"] |
| 14 | tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"} |
| 15 | disabled = [] |
| 16 | before_creation = null |
| 17 | after_creation = null |
| 18 | after_pipeline_creation = null |
| 19 | |
| 20 | [components] |
| 21 | |
| 22 | [components.tagger] |
| 23 | factory = "tagger" |
| 24 | |
| 25 | [components.tagger.model] |
| 26 | @architectures = "spacy.Tagger.v1" |
| 27 | nO = null |
| 28 | |
| 29 | [components.tagger.model.tok2vec] |
| 30 | @architectures = "spacy-transformers.TransformerListener.v1" |
| 31 | grad_factor = 1.0 |
| 32 | pooling = {"@layers":"reduce_mean.v1"} |
| 33 | |
| 34 | [components.transformer] |
| 35 | factory = "transformer" |
| 36 | max_batch_items = 4096 |
| 37 | set_extra_annotations = {"@annotation_setters":"spacy-transformers.null_annotation_setter.v1"} |
| 38 | |
| 39 | [components.transformer.model] |
| 40 | @architectures = "spacy-transformers.TransformerModel.v1" |
| 41 | name = "bert-base-german-cased" |
| 42 | |
| 43 | [components.transformer.model.get_spans] |
| 44 | @span_getters = "spacy-transformers.strided_spans.v1" |
| 45 | window = 128 |
| 46 | stride = 96 |
| 47 | |
| 48 | [components.transformer.model.tokenizer_config] |
| 49 | use_fast = true |
| 50 | |
| 51 | [corpora] |
| 52 | |
| 53 | [corpora.dev] |
| 54 | @readers = "spacy.Corpus.v1" |
| 55 | path = ${paths.dev} |
| 56 | max_length = 0 |
| 57 | gold_preproc = false |
| 58 | limit = 0 |
| 59 | augmenter = null |
| 60 | |
| 61 | [corpora.train] |
| 62 | @readers = "spacy.Corpus.v1" |
| 63 | path = ${paths.train} |
| 64 | max_length = 500 |
| 65 | gold_preproc = false |
| 66 | limit = 0 |
| 67 | augmenter = null |
| 68 | |
| 69 | [training] |
| 70 | accumulate_gradient = 3 |
| 71 | dev_corpus = "corpora.dev" |
| 72 | train_corpus = "corpora.train" |
| 73 | seed = ${system.seed} |
| 74 | gpu_allocator = ${system.gpu_allocator} |
| 75 | dropout = 0.1 |
| 76 | patience = 1600 |
| 77 | max_epochs = 0 |
| 78 | max_steps = 20000 |
| 79 | eval_frequency = 200 |
| 80 | frozen_components = [] |
| 81 | before_to_disk = null |
| 82 | |
| 83 | [training.batcher] |
| 84 | @batchers = "spacy.batch_by_padded.v1" |
| 85 | discard_oversize = true |
| 86 | size = 2000 |
| 87 | buffer = 256 |
| 88 | get_length = null |
| 89 | |
| 90 | [training.logger] |
| 91 | @loggers = "spacy.ConsoleLogger.v1" |
| 92 | progress_bar = false |
| 93 | |
| 94 | [training.optimizer] |
| 95 | @optimizers = "Adam.v1" |
| 96 | beta1 = 0.9 |
| 97 | beta2 = 0.999 |
| 98 | L2_is_weight_decay = true |
| 99 | L2 = 0.01 |
| 100 | grad_clip = 1.0 |
| 101 | use_averages = false |
| 102 | eps = 0.00000001 |
| 103 | |
| 104 | [training.optimizer.learn_rate] |
| 105 | @schedules = "warmup_linear.v1" |
| 106 | warmup_steps = 250 |
| 107 | total_steps = 20000 |
| 108 | initial_rate = 0.00005 |
| 109 | |
| 110 | [training.score_weights] |
| 111 | tag_acc = 1.0 |
| 112 | |
| 113 | [pretraining] |
| 114 | |
| 115 | [initialize] |
| 116 | vectors = null |
| 117 | init_tok2vec = ${paths.init_tok2vec} |
| 118 | vocab_data = null |
| 119 | lookups = null |
| 120 | |
| 121 | [initialize.components] |
| 122 | |
| 123 | [initialize.tokenizer] |