Skip to content

vllm.entrypoints.pooling.utils

Functions:

encode_pooling_output_float_or_ndarray(output)

Return an ndarray when the response renderer can serialize NumPy.

Source code in vllm/entrypoints/pooling/utils.py 
75
76
77
78
79
80
81
82
83
def encode_pooling_output_float_or_ndarray(output: PoolingRequestOutput) -> Any:
    """Return an ndarray when the response renderer can serialize NumPy."""
    try:
        data = output.outputs.data
        if not data.is_contiguous():
            data = data.contiguous()
        return data.numpy()
    except (RuntimeError, TypeError):
        return output.outputs.data.tolist()