olmoe

olmoe

OLMoE on vLLM (vllm.model_executor.models.olmoe).

Llama's names and Llama's fused block; the attention norms its queries and keys, and every block's mlp is a mixture of experts, whose output is what the block adds. The routing is inside vLLM's fused MoE kernel and has no values here.

RENAME module-attribute

RENAME = {'model.embed_tokens': 'embed_tokens', 'model.layers': 'layers', 'model.norm': 'norm'}

ENVOYS module-attribute

ENVOYS = {OlmoeDecoderLayer: Layer, OlmoeAttention: Attention, OlmoeMoE: Mlp}

Layer

Layer(*args: Any, **kwargs: Any)

Bases: FusedLayer

The decoder block: forward(positions, hidden_states, residual) -> (hidden_states, residual).