Introduce code from Rust monorepo. This is a mess

2026-10-27 08:58:13 +00:00 · 2022-02-28 05:43:54 -05:00
parent 921fb49f9a
commit ba326bfcfd
1058 changed files with 192192 additions and 0 deletions
@@ -0,0 +1,135 @@
+# Question Answering with SQUAD
+name: &name QA
+
+pretrained_model: null # pretrained QAModel model from list_available_models()
+do_training: true # training mode, for testing change to false
+trainer:
+  gpus: 1 # the number of gpus, 0 for CPU, or list with gpu indices
+  num_nodes: 1
+  max_epochs: 2 # the number of training epochs
+  max_steps: null # precedence over max_epochs
+  accumulate_grad_batches: 1 # accumulates grads every k batches
+  precision: 16 # 16 to use AMP
+  amp_level: O1 # O1 or O2 if using AMP
+  accelerator: ddp
+  gradient_clip_val: 0.0
+  val_check_interval: 1.0 # check once per epoch .25 for 4 times per epoch
+  checkpoint_callback: false # provided by exp_manager
+  logger: false # provided by exp_manager
+  num_sanity_val_steps: 0
+  log_every_n_steps: 1  # Interval of logging.
+
+model:
+  nemo_path: null # exported .nemo path
+  dataset:
+    version_2_with_negative: false
+    # If true, the examples contain some that do not have an answer.
+    doc_stride: 128
+    # When splitting up a long document into chunks,
+    # how much stride to take between chunks.
+    max_query_length: 64
+    # The maximum number of tokens for the question.
+    # Questions longer than this will be truncated to
+    # this length.
+    max_seq_length: 384
+    # The maximum total input sequence length after
+    # WordPiece tokenization. Sequences longer than this
+    # will be truncated, and sequences shorter than this
+    # will be padded.
+    max_answer_length: 30
+    # The maximum length of an answer that can be
+    # generated. This is needed because the start
+    # and end predictions are not conditioned
+    # on one another.
+    null_score_diff_threshold: 0.0
+    # If null_score - best_non_null is greater than the threshold predict null.
+    n_best_size: 20
+    # The total number of n-best predictions to generate at testing.
+    use_cache: true
+    do_lower_case: true
+    # if true does lower case
+
+    num_workers:  2
+    pin_memory: false
+    drop_last: false
+
+  train_ds:
+    file: null # .json file
+    batch_size: 24 # per GPU
+    shuffle: true
+    num_samples: -1
+    # Default values for the following params are retrieved from dataset config section, but you may override them
+    num_workers: ${model.dataset.num_workers}
+    drop_last: ${model.dataset.drop_last}
+    pin_memory: ${model.dataset.pin_memory}
+
+  validation_ds:
+    file: null # .json file
+    batch_size: 24 # per GPU
+    shuffle: false
+    num_samples: -1    
+    # Default values for the following params are retrieved from dataset config section, but you may override them
+    num_workers: ${model.dataset.num_workers}
+    drop_last: ${model.dataset.drop_last}
+    pin_memory: ${model.dataset.pin_memory}
+
+  test_ds:
+    file: null # .json file
+    batch_size: 24 # per GPU
+    shuffle: false
+    num_samples: -1
+    # Default values for the following params are retrieved from dataset config section, but you may override them
+    num_workers: ${model.dataset.num_workers}
+    drop_last: ${model.dataset.drop_last}
+    pin_memory: ${model.dataset.pin_memory}
+    
+  tokenizer:
+    tokenizer_name: ${model.language_model.pretrained_model_name} # tokenizer that inherits from TokenizerSpec
+    vocab_file: null # path to vocab file
+    tokenizer_model: null # only used if tokenizer is sentencepiece
+    special_tokens: null # expand the following to a dictionary if special tokens need to be added.
+    #  only necessary for adding transformer/bert-specific special tokens to tokenizer if the tokenizer does not already have these inherently.
+
+  language_model:
+    pretrained_model_name: bert-base-uncased # BERT-like model name
+    lm_checkpoint: null
+    config_file: null # json file, precedence over config
+    config: null # if specified initializes model from scratch
+
+  token_classifier:
+    num_layers: 1
+    dropout: 0.0
+    num_classes: 2
+    activation: relu
+    log_softmax: false
+    use_transformer_init: true
+
+
+  optim:
+    name: adamw
+    lr: 3e-5
+    weight_decay: 0.0
+    sched:
+      name: SquareRootAnnealing
+
+      # pytorch lightning args
+      monitor: val_loss
+      reduce_on_plateau: false
+
+      # scheduler config override
+      warmup_steps: null
+      warmup_ratio: 0.0
+      last_epoch: -1
+
+exp_manager:
+  exp_dir: null # where to store logs and checkpoints
+  name: *name # name of experiment
+  create_tensorboard_logger: True
+  create_checkpoint_callback: True
+
+hydra:
+  run:
+    dir: .
+  job_logging:
+    root:
+      handlers: null