Views
No views yet
1
2Experiment configurations
3├── datasets
4│ └── mnli_train:
5│ dataset_name: multi_nli
6│ dataset_config_name: null
7│ cache_dir: null
8│ input_fields:
9│ - premise
10│ - hypothesis
11│ target_field: label
12│ train_subset_names: null
13│ val_subset_names: validation_matched
14│ test_subset_names: none
15│ train_val_split: null
16│ limit_train_samples: null
17│ limit_val_samples: null
18│ limit_test_samples: null
19│ sampling_kwargs:
20│ sampling_strategy: random
21│ seed: 42
22│ replace: false
23│ align_labels_with_mapping: null
24│ avoid_consistency_check: false
25│ predict_label_mapping: null
26│ mnli:
27│ dataset_name: multi_nli
28│ dataset_config_name: null
29│ cache_dir: null
30│ input_fields:
31│ - premise
32│ - hypothesis
33│ target_field: label
34│ train_subset_names: none
35│ val_subset_names: none
36│ test_subset_names: validation_mismatched
37│ train_val_split: null
38│ limit_train_samples: null
39│ limit_val_samples: null
40│ limit_test_samples: null
41│ sampling_kwargs:
42│ sampling_strategy: random
43│ seed: 42
44│ replace: false
45│ align_labels_with_mapping: null
46│ avoid_consistency_check: false
47│ predict_label_mapping: null
48│
49├── data
50│ └── _target_: src.task.nli.data.NLIGenerationData.from_config
51│ main_dataset_name: null
52│ use_additional_as_test: null
53│ dataloader:
54│ batch_size: 64
55│ eval_batch_size: 100
56│ num_workers: 16
57│ pin_memory: true
58│ drop_last: false
59│ persistent_workers: false
60│ shuffle: true
61│ seed_dataloader: 42
62│ replacement: false
63│ processing:
64│ preprocessing_num_workers: 16
65│ preprocessing_batch_size: 1000
66│ load_from_cache_file: true
67│ padding: longest
68│ truncation: longest_first
69│ max_source_length: 128
70│ max_target_length: 128
71│ template: 'premise: $premise $label hypothesis: '
72│ tokenizer:
73│ _target_: transformers.AutoTokenizer.from_pretrained
74│ pretrained_model_name_or_path: google/t5-v1_1-base
75│ use_fast: true
76│
77├── task
78│ └── optimizer:
79│ name: Adafactor
80│ lr: 0.001
81│ weight_decay: 0.0
82│ no_decay:
83│ - bias
84│ - LayerNorm.weight
85│ decay_rate: -0.8
86│ clip_threshold: 1.0
87│ relative_step: false
88│ scale_parameter: false
89│ warmup_init: false
90│ scheduler:
91│ name: constant_schedule
92│ model:
93│ model_name_or_path: google/t5-v1_1-base
94│ checkpoint_path: null
95│ freeze: false
96│ seed_init_weight: 42
97│ _target_: src.task.nli.NLIGenerationTask.from_config
98│ generation:
99│ max_length: 128
100│ min_length: 3
101│ do_sample: true
102│ early_stopping: false
103│ num_beams: 1
104│ temperature: 1.0
105│ top_k: 50
106│ top_p: 0.95
107│ repetition_penalty: null
108│ length_penalty: null
109│ no_repeat_ngram_size: null
110│ encoder_no_repeat_ngram_size: null
111│ num_return_sequences: 1
112│ max_time: null
113│ max_new_tokens: null
114│ decoder_start_token_id: null
115│ use_cache: null
116│ num_beam_groups: null
117│ diversity_penalty: null
118│
119├── trainer
120│ └── _target_: pytorch_lightning.Trainer
121│ callbacks:
122│ lr_monitor:
123│ _target_: pytorch_lightning.callbacks.LearningRateMonitor
124│ logging_interval: step
125│ log_momentum: false
126│ model_checkpoint:
127│ _target_: pytorch_lightning.callbacks.ModelCheckpoint
128│ dirpath: ./checkpoints/
129│ filename: nli_generator_mnli-epoch={epoch:02d}-val_loss={val/aggregated_loss:.2f}
130│ monitor: val/aggregated_loss
131│ mode: min
132│ verbose: false
133│ save_last: true
134│ save_top_k: 1
135│ auto_insert_metric_name: false
136│ save_on_train_epoch_end: false
137│ rich_model_summary:
138│ _target_: pytorch_lightning.callbacks.RichModelSummary
139│ max_depth: 1
140│ log_grad_norm:
141│ _target_: src.core.callbacks.LogGradNorm
142│ norm_type: 2
143│ group_separator: /
144│ only_total: true
145│ on_step: true
146│ on_epoch: false
147│ prog_bar: true
148│ log_generated_text:
149│ _target_: src.core.callbacks.GenerateAndLogText
150│ dirpath: ./generated_text
151│ type: generated_text
152│ pop_keys_after_logging: true
153│ on_train: false
154│ on_validation: false
155│ on_test: true
156│ log_to_wandb: true
157│ wandb_log_dataset_sizes:
158│ _target_: src.core.callbacks.WandbLogDatasetSizes
159│ logger:
160│ wandb:
161│ _target_: pytorch_lightning.loggers.WandbLogger
162│ project: nli_debiasing
163│ entity: team_brushino
164│ name: nli_generator_mnli
165│ save_dir: ./
166│ offline: false
167│ log_model: false
168│ group: mnli
169│ job_type: generator
170│ tags:
171│ - nli_generator_mnli
172│ - seed=42
173│ - seed_dataloader=42
174│ notes: nli_generator_mnli_time=02-24-53
175│ enable_checkpointing: true
176│ enable_progress_bar: true
177│ enable_model_summary: true
178│ gradient_clip_val: 0.0
179│ gradient_clip_algorithm: null
180│ accelerator: gpu
181│ devices: auto
182│ gpus: null
183│ auto_select_gpus: true
184│ accumulate_grad_batches: 1
185│ max_epochs: 3
186│ min_epochs: 1
187│ max_steps: -1
188│ min_steps: null
189│ max_time: null
190│ num_sanity_val_steps: 2
191│ overfit_batches: 0.0
192│ fast_dev_run: false
193│ limit_train_batches: 1.0
194│ limit_val_batches: 1.0
195│ limit_test_batches: 1.0
196│ profiler: null
197│ detect_anomaly: false
198│ deterministic: false
199│ check_val_every_n_epoch: 1
200│ val_check_interval: 0.1
201│ log_every_n_steps: 10
202│ move_metrics_to_cpu: false
203│
204└── training
205 └── run_val_before_fit: false
206 run_val_after_fit: false
207 run_test_before_fit: false
208 run_test_after_fit: true
209 lr: 0.001
210 seed: 42
211 show_batch: false
212 batch_size: 64
213 eval_batch_size: 100
214 num_workers: 16
215 pin_memory: true
216 drop_last: false
217 persistent_workers: false
218 shuffle: true
219 seed_dataloader: 42
220 ignore_warnings: true
221 experiment_name: nli_generator_mnli
222