跳到内容

llmcompressor.modeling.deepseek_v3

CalibrationDeepseekV3MoE

CalibrationDeepseekV3MoE(
    original: DeepseekV3MoE,
    config: DeepseekV3Config,
    calibrate_all_experts: bool = True,
)

基类: MoECalibrationModule

CalibrationDeepseekV3MoE 的校准版本,它将所有 token 发送到所有专家。

源代码在 llmcompressor/modeling/deepseek_v3.py
def __init__(
    self,
    original: OriginalDeepseekV3MoE,
    config: DeepseekV3Config,
    calibrate_all_experts: bool = True,
):
    super().__init__()
    self.config = config
    self.experts = original.experts
    self.gate = original.gate
    self.shared_experts = original.shared_experts
    self.calibrate_all_experts = calibrate_all_experts