llmcompressor.modifiers.transform.utils.hybrid_attention
Functions:
-
get_config_layer_types–Extract layer_types from a model config, checking text_config first when present.
-
get_hybrid_attention_config–Extract layer_types and num_hidden_layers from a model with hybrid attention.
get_config_layer_types
Extract layer_types from a model config, checking text_config first when present.
get_hybrid_attention_config
Extract layer_types and num_hidden_layers from a model with hybrid attention.
Checks both top-level config and text_config (for VL models like Qwen3.5).
Returns (layer_types, num_hidden_layers) or None if the model does not
expose the metadata needed to identify hybrid attention layers.