Skip to content

llmcompressor.args.model_arguments

Model argument classes for LLM compression workflows.

This module defines dataclass-based argument containers for configuring model loading, tokenization, and preprocessing parameters. Supports various model sources including HuggingFace model hub, local paths, and custom configurations for compression workflows.

Classes:

  • ModelArguments –

    Model variables used for oneshot calibration, finetuning and

ModelArguments dataclass 

ModelArguments(
    model: str,
    distill_teacher: Optional[str] = None,
    config_name: Optional[str] = None,
    tokenizer: Optional[str] = None,
    processor: Optional[str] = None,
    cache_dir: Optional[str] = None,
    use_auth_token: bool = False,
    precision: str = "auto",
    tie_word_embeddings: bool = False,
    trust_remote_code_model: bool = False,
    save_compressed: Optional[bool] = True,
    model_revision: str = "main",
)

Model variables used for oneshot calibration, finetuning and stage runners (sequential run of oneshot and finetune).