Skip to content

`vllm.model_executor.models.sarvam` ¶

Classes:

SarvamMoEForCausalLM –

Same as BailingMoeForCausalLM, but normalizes gate expert_bias pre-load.

`SarvamMoEForCausalLM` ¶

Bases: BailingMoeForCausalLM

Same as BailingMoeForCausalLM, but normalizes gate expert_bias pre-load.

Source code in vllm/model_executor/models/sarvam.py

class SarvamMoEForCausalLM(BailingMoeForCausalLM):
    """Same as BailingMoeForCausalLM, but normalizes gate expert_bias pre-load."""

    def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
        return super().load_weights(_normalized_weights(weights))