neuron.py 744 Bytes
Newer Older
1
2
from typing import TYPE_CHECKING

3
4
from .interface import Platform, PlatformEnum

5
6
7
8
9
if TYPE_CHECKING:
    from vllm.config import VllmConfig
else:
    VllmConfig = None

10
11
12

class NeuronPlatform(Platform):
    _enum = PlatformEnum.NEURON
13
    device_name: str = "neuron"
14
    device_type: str = "neuron"
15
    supported_quantization: list[str] = ["neuron_quant"]
16
17
18
19

    @classmethod
    def get_device_name(cls, device_id: int = 0) -> str:
        return "neuron"
20
21
22
23
24
25
26

    @classmethod
    def check_and_update_config(cls, vllm_config: VllmConfig) -> None:
        parallel_config = vllm_config.parallel_config
        if parallel_config.worker_cls == "auto":
            parallel_config.worker_cls = \
                "vllm.worker.neuron_worker.NeuronWorker"