vllm.model_executor.layers ¶

Modules:

• activation –

  Custom activation functions.

• attention –
• attention_layer_base –

  Base class for attention-like layers.

• batch_invariant –
• conv –

  Conv Layer Class.

• fla –
• fused_moe –
• layernorm –

  Custom normalization layers.

• lightning_attn –
• linear –
• logits_processor –

  A layer that compute logits from hidden_stats.

• mamba –
• mhc –
• minimax_rms_norm –
• mla –
• pooler –
• quantization –
• resampler –

  Shared resampler perceiver network used in multimodal models and

• rotary_embedding –

  Rotary Positional Embeddings.

• sparse_attn_indexer –

  Custom Sparse Attention Indexer layers.

• utils –

  Utility methods for model layers.

• vocab_parallel_embedding –