LayerSpec

class ctranslate2.specs.LayerSpec

A layer specification declares the weights that should be set by the converters.

Inherits from: ctranslate2.specs.model_spec.FrozenAttr

Extended by:

• ctranslate2.specs.ModelSpec

• ctranslate2.specs.TransformerDecoderSpec

• ctranslate2.specs.TransformerEncoderSpec

Methods:

• optimize

• validate

• variables

optimize(quantization: Optional[str] = None) → None

Recursively applies some optimizations to this layer:

• Alias variables with the same shape and value.

• Quantize weights.

Parameters

quantization – Weight quantization scheme (possible values are: int8, int8_float32, int8_float16, int8_bfloat16, int16, float16, bfloat16, float32).

validate() → None

Verify that the required weights are set.

Raises

ValueError – If a required weight is not set in the specification.

variables(prefix: str = '', ordered: bool = False) → Dict[str, ndarray]

Recursively returns the weights from this layer and its children.

Parameters

• prefix – Prefix to prepend to all variable names.

• ordered – If set, an ordered list is returned instead.

Returns

Dictionary mapping variables name to value.