Commit
β’
2388064
1
Parent(s):
540b78e
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,323 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
## Overview
|
| 2 |
+
|
| 3 |
+
T5-Base v1.1 model trained to generate hypotheses given a premise and a label. Below the settings used to train it.
|
| 4 |
+
|
| 5 |
+
```yaml
|
| 6 |
+
Experiment configurations
|
| 7 |
+
βββ datasets
|
| 8 |
+
β βββ snli_train:
|
| 9 |
+
β dataset_name: snli
|
| 10 |
+
β dataset_config_name: null
|
| 11 |
+
β cache_dir: null
|
| 12 |
+
β input_fields:
|
| 13 |
+
β - premise
|
| 14 |
+
β - hypothesis
|
| 15 |
+
β target_field: label
|
| 16 |
+
β train_subset_names: null
|
| 17 |
+
β val_subset_names: validation
|
| 18 |
+
β test_subset_names: none
|
| 19 |
+
β train_val_split: null
|
| 20 |
+
β limit_train_samples: null
|
| 21 |
+
β limit_val_samples: null
|
| 22 |
+
β limit_test_samples: null
|
| 23 |
+
β sampling_kwargs:
|
| 24 |
+
β sampling_strategy: random
|
| 25 |
+
β seed: 42
|
| 26 |
+
β replace: false
|
| 27 |
+
β align_labels_with_mapping: null
|
| 28 |
+
β avoid_consistency_check: false
|
| 29 |
+
β predict_label_mapping: null
|
| 30 |
+
β anli_train:
|
| 31 |
+
β dataset_name: anli
|
| 32 |
+
β dataset_config_name: null
|
| 33 |
+
β cache_dir: null
|
| 34 |
+
β input_fields:
|
| 35 |
+
β - premise
|
| 36 |
+
β - hypothesis
|
| 37 |
+
β target_field: label
|
| 38 |
+
β train_subset_names:
|
| 39 |
+
β - train_r1
|
| 40 |
+
β - train_r2
|
| 41 |
+
β - train_r3
|
| 42 |
+
β val_subset_names:
|
| 43 |
+
β - dev_r1
|
| 44 |
+
β - dev_r2
|
| 45 |
+
β - dev_r3
|
| 46 |
+
β test_subset_names: none
|
| 47 |
+
β train_val_split: null
|
| 48 |
+
β limit_train_samples: null
|
| 49 |
+
β limit_val_samples: null
|
| 50 |
+
β limit_test_samples: null
|
| 51 |
+
β sampling_kwargs:
|
| 52 |
+
β sampling_strategy: random
|
| 53 |
+
β seed: 42
|
| 54 |
+
β replace: false
|
| 55 |
+
β align_labels_with_mapping: null
|
| 56 |
+
β avoid_consistency_check: false
|
| 57 |
+
β predict_label_mapping: null
|
| 58 |
+
β mnli_train:
|
| 59 |
+
β dataset_name: multi_nli
|
| 60 |
+
β dataset_config_name: null
|
| 61 |
+
β cache_dir: null
|
| 62 |
+
β input_fields:
|
| 63 |
+
β - premise
|
| 64 |
+
β - hypothesis
|
| 65 |
+
β target_field: label
|
| 66 |
+
β train_subset_names: null
|
| 67 |
+
β val_subset_names: validation_matched
|
| 68 |
+
β test_subset_names: none
|
| 69 |
+
β train_val_split: null
|
| 70 |
+
β limit_train_samples: null
|
| 71 |
+
β limit_val_samples: null
|
| 72 |
+
β limit_test_samples: null
|
| 73 |
+
β sampling_kwargs:
|
| 74 |
+
β sampling_strategy: random
|
| 75 |
+
β seed: 42
|
| 76 |
+
β replace: false
|
| 77 |
+
β align_labels_with_mapping: null
|
| 78 |
+
β avoid_consistency_check: false
|
| 79 |
+
β predict_label_mapping: null
|
| 80 |
+
β snli:
|
| 81 |
+
β dataset_name: snli
|
| 82 |
+
β dataset_config_name: null
|
| 83 |
+
β cache_dir: null
|
| 84 |
+
β input_fields:
|
| 85 |
+
β - premise
|
| 86 |
+
β - hypothesis
|
| 87 |
+
β target_field: label
|
| 88 |
+
β train_subset_names: none
|
| 89 |
+
β val_subset_names: none
|
| 90 |
+
β test_subset_names: null
|
| 91 |
+
β train_val_split: null
|
| 92 |
+
β limit_train_samples: null
|
| 93 |
+
β limit_val_samples: null
|
| 94 |
+
β limit_test_samples: null
|
| 95 |
+
β sampling_kwargs:
|
| 96 |
+
β sampling_strategy: random
|
| 97 |
+
β seed: 42
|
| 98 |
+
β replace: false
|
| 99 |
+
β align_labels_with_mapping: null
|
| 100 |
+
β avoid_consistency_check: false
|
| 101 |
+
β predict_label_mapping: null
|
| 102 |
+
β anli:
|
| 103 |
+
β dataset_name: anli
|
| 104 |
+
β dataset_config_name: null
|
| 105 |
+
β cache_dir: null
|
| 106 |
+
β input_fields:
|
| 107 |
+
β - premise
|
| 108 |
+
β - hypothesis
|
| 109 |
+
β target_field: label
|
| 110 |
+
β train_subset_names: none
|
| 111 |
+
β val_subset_names: none
|
| 112 |
+
β test_subset_names:
|
| 113 |
+
β - test_r1
|
| 114 |
+
β - test_r2
|
| 115 |
+
β - test_r3
|
| 116 |
+
β train_val_split: null
|
| 117 |
+
β limit_train_samples: null
|
| 118 |
+
β limit_val_samples: null
|
| 119 |
+
β limit_test_samples: null
|
| 120 |
+
β sampling_kwargs:
|
| 121 |
+
β sampling_strategy: random
|
| 122 |
+
β seed: 42
|
| 123 |
+
β replace: false
|
| 124 |
+
β align_labels_with_mapping: null
|
| 125 |
+
οΏ½οΏ½οΏ½ avoid_consistency_check: false
|
| 126 |
+
β predict_label_mapping: null
|
| 127 |
+
β mnli:
|
| 128 |
+
β dataset_name: multi_nli
|
| 129 |
+
β dataset_config_name: null
|
| 130 |
+
β cache_dir: null
|
| 131 |
+
β input_fields:
|
| 132 |
+
β - premise
|
| 133 |
+
β - hypothesis
|
| 134 |
+
β target_field: label
|
| 135 |
+
β train_subset_names: none
|
| 136 |
+
β val_subset_names: none
|
| 137 |
+
β test_subset_names: validation_mismatched
|
| 138 |
+
β train_val_split: null
|
| 139 |
+
β limit_train_samples: null
|
| 140 |
+
β limit_val_samples: null
|
| 141 |
+
β limit_test_samples: null
|
| 142 |
+
β sampling_kwargs:
|
| 143 |
+
β sampling_strategy: random
|
| 144 |
+
β seed: 42
|
| 145 |
+
β replace: false
|
| 146 |
+
β align_labels_with_mapping: null
|
| 147 |
+
β avoid_consistency_check: false
|
| 148 |
+
β predict_label_mapping: null
|
| 149 |
+
β
|
| 150 |
+
βββ data
|
| 151 |
+
β βββ _target_: src.task.nli.data.NLIGenerationData.from_config
|
| 152 |
+
β main_dataset_name: null
|
| 153 |
+
β use_additional_as_test: null
|
| 154 |
+
β dataloader:
|
| 155 |
+
β batch_size: 96
|
| 156 |
+
β eval_batch_size: 96
|
| 157 |
+
β num_workers: 8
|
| 158 |
+
β pin_memory: true
|
| 159 |
+
β drop_last: false
|
| 160 |
+
β persistent_workers: false
|
| 161 |
+
β shuffle: true
|
| 162 |
+
β seed_dataloader: 42
|
| 163 |
+
β replacement: false
|
| 164 |
+
β processing:
|
| 165 |
+
β preprocessing_num_workers: 8
|
| 166 |
+
β preprocessing_batch_size: 1000
|
| 167 |
+
β load_from_cache_file: true
|
| 168 |
+
β padding: longest
|
| 169 |
+
β truncation: longest_first
|
| 170 |
+
β max_source_length: 128
|
| 171 |
+
β max_target_length: 128
|
| 172 |
+
β template: 'premise: $premise $label hypothesis: '
|
| 173 |
+
β tokenizer:
|
| 174 |
+
β _target_: transformers.AutoTokenizer.from_pretrained
|
| 175 |
+
β pretrained_model_name_or_path: pietrolesci/t5-v1_1-base_nli_gen
|
| 176 |
+
β use_fast: true
|
| 177 |
+
β
|
| 178 |
+
βββ task
|
| 179 |
+
β βββ optimizer:
|
| 180 |
+
β name: Adafactor
|
| 181 |
+
β lr: 0.001
|
| 182 |
+
β weight_decay: 0.0
|
| 183 |
+
β no_decay:
|
| 184 |
+
β - bias
|
| 185 |
+
β - LayerNorm.weight
|
| 186 |
+
β decay_rate: -0.8
|
| 187 |
+
β clip_threshold: 1.0
|
| 188 |
+
β relative_step: false
|
| 189 |
+
β scale_parameter: false
|
| 190 |
+
β warmup_init: false
|
| 191 |
+
β scheduler:
|
| 192 |
+
β name: constant_schedule
|
| 193 |
+
β model:
|
| 194 |
+
β model_name_or_path: pietrolesci/t5-v1_1-base_nli_gen
|
| 195 |
+
β checkpoint_path: null
|
| 196 |
+
β freeze: false
|
| 197 |
+
β seed_init_weight: 42
|
| 198 |
+
β _target_: src.task.nli.NLIGenerationTask.from_config
|
| 199 |
+
β generation:
|
| 200 |
+
β generation_max_length: 128
|
| 201 |
+
β generation_min_length: 3
|
| 202 |
+
β do_sample: true
|
| 203 |
+
β early_stopping: false
|
| 204 |
+
β num_beams: 1
|
| 205 |
+
β temperature: 1.0
|
| 206 |
+
β top_k: 50
|
| 207 |
+
β top_p: 0.95
|
| 208 |
+
β repetition_penalty: null
|
| 209 |
+
β length_penalty: null
|
| 210 |
+
β no_repeat_ngram_size: null
|
| 211 |
+
β encoder_no_repeat_ngram_size: null
|
| 212 |
+
β num_return_sequences: 1
|
| 213 |
+
β max_time: null
|
| 214 |
+
β max_new_tokens: null
|
| 215 |
+
β decoder_start_token_id: null
|
| 216 |
+
β use_cache: null
|
| 217 |
+
β num_beam_groups: null
|
| 218 |
+
β diversity_penalty: null
|
| 219 |
+
β
|
| 220 |
+
βββ trainer
|
| 221 |
+
β βββ _target_: pytorch_lightning.Trainer
|
| 222 |
+
β callbacks:
|
| 223 |
+
β lr_monitor:
|
| 224 |
+
β _target_: pytorch_lightning.callbacks.LearningRateMonitor
|
| 225 |
+
β logging_interval: step
|
| 226 |
+
β log_momentum: false
|
| 227 |
+
β model_checkpoint:
|
| 228 |
+
β _target_: pytorch_lightning.callbacks.ModelCheckpoint
|
| 229 |
+
β dirpath: ./checkpoints/
|
| 230 |
+
β filename: nli_generator_sma-epoch={epoch:02d}-val_loss={val/aggregat
|
| 231 |
+
β monitor: val/aggregated_loss
|
| 232 |
+
β mode: min
|
| 233 |
+
β verbose: false
|
| 234 |
+
β save_last: true
|
| 235 |
+
β save_top_k: 1
|
| 236 |
+
β auto_insert_metric_name: false
|
| 237 |
+
β save_on_train_epoch_end: false
|
| 238 |
+
β rich_model_summary:
|
| 239 |
+
β _target_: pytorch_lightning.callbacks.RichModelSummary
|
| 240 |
+
β max_depth: 1
|
| 241 |
+
β log_grad_norm:
|
| 242 |
+
β _target_: src.core.callbacks.LogGradNorm
|
| 243 |
+
β norm_type: 2
|
| 244 |
+
β group_separator: /
|
| 245 |
+
β only_total: true
|
| 246 |
+
β on_step: true
|
| 247 |
+
β on_epoch: false
|
| 248 |
+
β prog_bar: true
|
| 249 |
+
β log_generated_text:
|
| 250 |
+
β _target_: src.core.callbacks.GenerateAndLogText
|
| 251 |
+
β dirpath: ./generated_text
|
| 252 |
+
β type: generated_text
|
| 253 |
+
β pop_keys_after_logging: true
|
| 254 |
+
β on_train: false
|
| 255 |
+
β on_validation: false
|
| 256 |
+
β on_test: true
|
| 257 |
+
β log_to_wandb: true
|
| 258 |
+
β wandb_log_dataset_sizes:
|
| 259 |
+
β _target_: src.core.callbacks.WandbLogDatasetSizes
|
| 260 |
+
β logger:
|
| 261 |
+
β wandb:
|
| 262 |
+
β _target_: pytorch_lightning.loggers.WandbLogger
|
| 263 |
+
β project: nli_debiasing
|
| 264 |
+
β entity: team_brushino
|
| 265 |
+
β name: nli_generator_sma
|
| 266 |
+
β save_dir: ./
|
| 267 |
+
β offline: false
|
| 268 |
+
β log_model: false
|
| 269 |
+
β group: generator
|
| 270 |
+
β job_type: genearator_training
|
| 271 |
+
β tags:
|
| 272 |
+
β - nli_generator_sma
|
| 273 |
+
β - seed=42
|
| 274 |
+
β - seed_dataloader=42
|
| 275 |
+
β notes: nli_generator_sma_time=01-37-04
|
| 276 |
+
β enable_checkpointing: true
|
| 277 |
+
β enable_progress_bar: true
|
| 278 |
+
β enable_model_summary: true
|
| 279 |
+
β gradient_clip_val: 6
|
| 280 |
+
β gradient_clip_algorithm: null
|
| 281 |
+
β accelerator: gpu
|
| 282 |
+
β devices: auto
|
| 283 |
+
β gpus: null
|
| 284 |
+
β auto_select_gpus: true
|
| 285 |
+
β accumulate_grad_batches: 1
|
| 286 |
+
β max_epochs: 2
|
| 287 |
+
β min_epochs: 1
|
| 288 |
+
β max_steps: -1
|
| 289 |
+
β min_steps: null
|
| 290 |
+
β max_time: null
|
| 291 |
+
β num_sanity_val_steps: 2
|
| 292 |
+
β overfit_batches: 0.0
|
| 293 |
+
β fast_dev_run: false
|
| 294 |
+
β limit_train_batches: 1.0
|
| 295 |
+
β limit_val_batches: 1.0
|
| 296 |
+
β limit_test_batches: 1.0
|
| 297 |
+
β profiler: null
|
| 298 |
+
β detect_anomaly: false
|
| 299 |
+
β deterministic: false
|
| 300 |
+
β check_val_every_n_epoch: 1
|
| 301 |
+
β val_check_interval: 0.5
|
| 302 |
+
β log_every_n_steps: 1
|
| 303 |
+
β move_metrics_to_cpu: false
|
| 304 |
+
β
|
| 305 |
+
βββ training
|
| 306 |
+
βββ run_val_before_fit: false
|
| 307 |
+
run_val_after_fit: false
|
| 308 |
+
run_test_before_fit: false
|
| 309 |
+
run_test_after_fit: true
|
| 310 |
+
lr: 0.001
|
| 311 |
+
seed: 42
|
| 312 |
+
show_batch: false
|
| 313 |
+
batch_size: 96
|
| 314 |
+
eval_batch_size: 96
|
| 315 |
+
num_workers: 8
|
| 316 |
+
pin_memory: true
|
| 317 |
+
drop_last: false
|
| 318 |
+
persistent_workers: false
|
| 319 |
+
shuffle: true
|
| 320 |
+
seed_dataloader: 42
|
| 321 |
+
ignore_warnings: true
|
| 322 |
+
experiment_name: nli_generator_sma
|
| 323 |
+
```
|