nemo_curator.models.qwen_lm

Module Contents

Classes

Name	Description
`LLM`	-
`QwenLM`	Qwen language model.
`SamplingParams`	-

Functions

Name	Description
`_weights_complete`	-

Data

VLLM_AVAILABLE

_QWEN_LM_VARIANTS_INFO

API

class nemo_curator.models.qwen_lm.LLM()

class nemo_curator.models.qwen_lm.QwenLM(
    model_dir: str = '',
    model_variant: str = 'qwen2.5',
    caption_batch_size: int = 1,
    fp8: bool = False,
    max_output_tokens: int = 512,
    vllm_kwargs = {}
)

Bases: ModelInterface

Qwen language model.

nemo_curator.models.qwen_lm.QwenLM.download_weights_on_node(
    model_dir: str,
    variant: str = 'qwen2.5'
) -> None

classmethod

Download the weights for the QwenLM model on the node.

nemo_curator.models.qwen_lm.QwenLM.generate(
    inputs: list[dict[str, typing.Any]]
) -> list[str]

nemo_curator.models.qwen_lm.QwenLM.model_id_names() -> list[str]

nemo_curator.models.qwen_lm.QwenLM.setup() -> None

class nemo_curator.models.qwen_lm.SamplingParams()

nemo_curator.models.qwen_lm._weights_complete(
    model_dir_path: pathlib.Path
) -> bool

nemo_curator.models.qwen_lm.VLLM_AVAILABLE = True

nemo_curator.models.qwen_lm._QWEN_LM_VARIANTS_INFO: Final = {'qwen2.5': ('Qwen/Qwen2.5-14B-Instruct', 'cf98f3b'), 'qwen3': ('Qwen/Qwen3-14B'...