nemo_curator.models.qwen_vl

Module Contents

Classes

Name	Description
`LLM`	-
`QwenVL`	-
`SamplingParams`	-

Data

VLLM_AVAILABLE

_QWEN2_5_VL_MODEL_ID

_QWEN2_5_VL_MODEL_REVISION

_QWEN3_VL_MODEL_ID

_QWEN3_VL_MODEL_REVISION

_QWEN_REVISION_INFO

_QWEN_VARIANTS_INFO

_QWEN_VL_PIXEL_PARAMS

API

class nemo_curator.models.qwen_vl.LLM()

class nemo_curator.models.qwen_vl.QwenVL(
    model_dir: str,
    model_variant: str,
    caption_batch_size: int,
    fp8: bool = True,
    max_output_tokens: int = 512,
    model_does_preprocess: bool = False,
    disable_mmcache: bool = False,
    stage2_prompt_text: str | None = None,
    verbose: bool = False,
    vllm_kwargs = {}
)

Bases: ModelInterface

model_id_names

list[str]

pattern

= '(.*)(user_prompt)(.*)'

stage2_prompt

weight_file

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

classmethod

Download the weights for the QwenVL model on the node.

nemo_curator.models.qwen_vl.QwenVL.generate(
    videos: list[dict[str, typing.Any]],
    generate_stage2_caption: bool = False,
    batch_size: int = 16
) -> list[str]

nemo_curator.models.qwen_vl.QwenVL.setup() -> None

class nemo_curator.models.qwen_vl.SamplingParams()

nemo_curator.models.qwen_vl.VLLM_AVAILABLE = True

nemo_curator.models.qwen_vl._QWEN2_5_VL_MODEL_ID = 'Qwen/Qwen2.5-VL-7B-Instruct'

nemo_curator.models.qwen_vl._QWEN2_5_VL_MODEL_REVISION = 'cc59489'

nemo_curator.models.qwen_vl._QWEN3_VL_MODEL_ID = 'Qwen/Qwen3-VL-8B-Instruct'

nemo_curator.models.qwen_vl._QWEN3_VL_MODEL_REVISION = '0c351dd'

nemo_curator.models.qwen_vl._QWEN_REVISION_INFO: Final = {'qwen2.5': _QWEN2_5_VL_MODEL_REVISION, 'qwen3': _QWEN3_VL_MODEL_REVISION}

nemo_curator.models.qwen_vl._QWEN_VARIANTS_INFO: Final = {'qwen2.5': _QWEN2_5_VL_MODEL_ID, 'qwen3': _QWEN3_VL_MODEL_ID}

nemo_curator.models.qwen_vl._QWEN_VL_PIXEL_PARAMS: Final = {'qwen2.5': {'image_factor': 28, 'min_pixels': 4 * 28 * 28, 'max_pixels': 16384 ...