bicleaner-ai-full-en-hbs / metadata.yaml
ZJaume's picture
LM filter with better support for cyrillic
279baef
bicleaner_ai_version: 2.0
precision_score: 0.892
recall_score: 0.912
f1_score: 0.902
matthews_corr_coef: 0.891
source_lang: en
target_lang: hbs
porn_removal_file: porn_removal.ftz
porn_removal_side: sl
source_lm: lm.en
target_lm: lm.hbs
lm_type: CHARACTER
clean_mean_perp: -1.116365617208272
clean_stddev_perp: 0.1687608181840407
noisy_mean_perp: -4.013399029709761
noisy_stddev_perp: 0.15687448432446158
classifier_type: xlmr
classifier_settings:
activation: relu
base_model: jplu/tf-xlm-roberta-base
batch_size: 96
calibration_params: [0.96121657, -0.41749883]
epochs: 15
loss: binary_crossentropy
maxlen: 150
model_name: bitextor/bicleaner-ai-full-en-hbs
n_classes: 2
n_hidden: 2048
patience: 3
steps_per_epoch: 3000
warmup_steps: 1000