opt-2.7b.yaml

# Path to pretrained model or model identifier from huggingface.co/models.
model_name_or_path: 'facebook/opt-2.7b'

# If passed, will use this index to ignore in the loss.
# If you're doing an evaluation, be sure to check the pad_token_id of the model you're using and pass it as an argument.
ignore_index: &PAD_TOKEN_ID 1

# If true, will call `model.parallelize` which splits the model on all GPUs available when applicable (model parallelism). "
# Note that this feature is still experimental in HF Transformers."
parallelize: false

# Batch size (per device) for the evaluation dataloader.
per_device_eval_batch_size: 8

# Pretrained config name or path if not the same as model_name
config_name: null

# Pretrained tokenizer name or path if not the same as model_name
tokenizer_name: null

# If passed, will use a slow tokenizer (not backed by the 🤗 Tokenizers library).
use_slow_tokenizer: false

# The maximum total input sequence length after tokenization. Sequences longer than this will be truncated,
# sequences shorter will be padded if `--pad_to_max_lengh` is passed.
max_length: 1024