neural_compressor.jax.quantization.layers_dynamic

Dynamic quantized layer implementations for JAX-backed Keras models.

Classes

DynamicQDQLayer

Layer that applies dynamic quantize-dequantize to activations.

QDynamicDenseMixin

Mixin that adds dynamic quantization to dense-like layers.

QDynamicDense

Dynamically quantized Dense layer.

QDynamicEinsumDense

Dynamically quantized EinsumDense layer.

QDynamicMultiHeadAttention

Dynamically quantized MultiHeadAttention layer.

QDynamicCachedGemma3Attention

Dynamically quantized CachedGemma3Attention layer.

QDynamicGemma3VisionAttention

Dynamically quantized Gemma3VisionAttention layer.

QDynamicReversibleEmbedding

Dynamically quantized ReversibleEmbedding layer.

Functions

register_dynamic_quantized_layer(clso)

Register quantized layer class for an original layer class.

Module Contents

neural_compressor.jax.quantization.layers_dynamic.register_dynamic_quantized_layer(clso)[source]

Register quantized layer class for an original layer class.

Parameters:

clso (type) – Original layer class to map to a quantized implementation.

Returns:

Decorator that registers the quantized class.

Return type:

Callable

class neural_compressor.jax.quantization.layers_dynamic.DynamicQDQLayer(name, activation_dtype, asymmetric=False)[source]

Layer that applies dynamic quantize-dequantize to activations.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicDenseMixin[source]

Mixin that adds dynamic quantization to dense-like layers.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicDense[source]

Dynamically quantized Dense layer.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicEinsumDense[source]

Dynamically quantized EinsumDense layer.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicMultiHeadAttention[source]

Dynamically quantized MultiHeadAttention layer.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicCachedGemma3Attention[source]

Dynamically quantized CachedGemma3Attention layer.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicGemma3VisionAttention[source]

Dynamically quantized Gemma3VisionAttention layer.

class neural_compressor.jax.quantization.layers_dynamic.QDynamicReversibleEmbedding[source]

Dynamically quantized ReversibleEmbedding layer.