olmo2¶
OLMo 2 on vLLM, through its transformers backend (Olmo2ForCausalLM -> TransformersForCausalLM).
vLLM has no OLMo 2 of its own: it runs transformers' modules, so the tree
and the block are transformers' (see nnterp.components.vllm.transformers_backend),
with Llama's names, the stream [1, tokens, hidden_size] and the engine's
attention layer mounted as the attention's attn child. The block is
post-norm, x + post_attention_layernorm(attn(x)) and then +
post_feedforward_layernorm(mlp(x)), so the contributions point at the
sibling norms, as on transformers. The attention norms its queries and keys
over every head at once before splitting them.