Hugging Face's logo

AutoTrain documentation

Seq2Seq

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v0.8.24).
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

Seq2Seq

Seq2Seq is a task that involves converting a sequence of words into another sequence of words. It is used in machine translation, text summarization, and question answering.

Data Format

You can have the dataset as a CSV file:

text,target
"this movie is great","dieser Film ist großartig"
"this movie is bad","dieser Film ist schlecht"
.
.
.

Or as a JSONL file:

{"text": "this movie is great", "target": "dieser Film ist großartig"}
{"text": "this movie is bad", "target": "dieser Film ist schlecht"}
.
.
.

Columns

Your CSV/JSONL dataset must have two columns: text and target.

Parameters

class autotrain.trainers.seq2seq.params.Seq2SeqParams

< >

( data_path: str = None model: str = 'google/flan-t5-base' username: typing.Optional[str] = None seed: int = 42 train_split: str = 'train' valid_split: typing.Optional[str] = None project_name: str = 'project-name' token: typing.Optional[str] = None push_to_hub: bool = False text_column: str = 'text' target_column: str = 'target' lr: float = 5e-05 epochs: int = 3 max_seq_length: int = 128 max_target_length: int = 128 batch_size: int = 2 warmup_ratio: float = 0.1 gradient_accumulation: int = 1 optimizer: str = 'adamw_torch' scheduler: str = 'linear' weight_decay: float = 0.0 max_grad_norm: float = 1.0 logging_steps: int = -1 eval_strategy: str = 'epoch' auto_find_batch_size: bool = False mixed_precision: typing.Optional[str] = None save_total_limit: int = 1 peft: bool = False quantization: typing.Optional[str] = 'int8' lora_r: int = 16 lora_alpha: int = 32 lora_dropout: float = 0.05 target_modules: str = 'all-linear' log: str = 'none' early_stopping_patience: int = 5 early_stopping_threshold: float = 0.01 )

Parameters

  • data_path (str) — Path to the dataset.
  • model (str) — Name of the model to be used. Default is “google/flan-t5-base”.
  • username (Optional[str]) — Hugging Face Username.
  • seed (int) — Random seed for reproducibility. Default is 42.
  • train_split (str) — Name of the training data split. Default is “train”.
  • valid_split (Optional[str]) — Name of the validation data split.
  • project_name (str) — Name of the project or output directory. Default is “project-name”.
  • token (Optional[str]) — Hub Token for authentication.
  • push_to_hub (bool) — Whether to push the model to the Hugging Face Hub. Default is False.
  • text_column (str) — Name of the text column in the dataset. Default is “text”.
  • target_column (str) — Name of the target text column in the dataset. Default is “target”.
  • lr (float) — Learning rate for training. Default is 5e-5.
  • epochs (int) — Number of training epochs. Default is 3.
  • max_seq_length (int) — Maximum sequence length for input text. Default is 128.
  • max_target_length (int) — Maximum sequence length for target text. Default is 128.
  • batch_size (int) — Training batch size. Default is 2.
  • warmup_ratio (float) — Proportion of warmup steps. Default is 0.1.
  • gradient_accumulation (int) — Number of gradient accumulation steps. Default is 1.
  • optimizer (str) — Optimizer to be used. Default is “adamw_torch”.
  • scheduler (str) — Learning rate scheduler to be used. Default is “linear”.
  • weight_decay (float) — Weight decay for the optimizer. Default is 0.0.
  • max_grad_norm (float) — Maximum gradient norm for clipping. Default is 1.0.
  • logging_steps (int) — Number of steps between logging. Default is -1 (disabled).
  • eval_strategy (str) — Evaluation strategy. Default is “epoch”.
  • auto_find_batch_size (bool) — Whether to automatically find the batch size. Default is False.
  • mixed_precision (Optional[str]) — Mixed precision training mode (fp16, bf16, or None).
  • save_total_limit (int) — Maximum number of checkpoints to save. Default is 1.
  • peft (bool) — Whether to use Parameter-Efficient Fine-Tuning (PEFT). Default is False.
  • quantization (Optional[str]) — Quantization mode (int4, int8, or None). Default is “int8”.
  • lora_r (int) — LoRA-R parameter for PEFT. Default is 16.
  • lora_alpha (int) — LoRA-Alpha parameter for PEFT. Default is 32.
  • lora_dropout (float) — LoRA-Dropout parameter for PEFT. Default is 0.05.
  • target_modules (str) — Target modules for PEFT. Default is “all-linear”.
  • log (str) — Logging method for experiment tracking. Default is “none”.
  • early_stopping_patience (int) — Patience for early stopping. Default is 5.
  • early_stopping_threshold (float) — Threshold for early stopping. Default is 0.01.

Seq2SeqParams is a configuration class for sequence-to-sequence training parameters.

< > Update on GitHub

←DreamBooth Token Classification→