Deployment configuration for Gemma language models. More...

Inheritance diagram for Mila::Dnn::GemmaModelConfig:

Public Member Functions
	GemmaModelConfig (dim_t context_length)
	Construct with a required context length.
std::string	toString () const
Public Member Functions inherited from Mila::Dnn::LanguageModelConfig< GemmaModelConfig >
std::string	baseToString () const
	Produce the base fields portion of a toString() summary.
dim_t	getContextLength () const noexcept
KvCacheCompression	getKvCacheCompression () const noexcept
dim_t	getLanguageModelHeadPositions () const noexcept
WeightQuantization	getWeightQuantization () const noexcept
GemmaModelConfig &	withContextLength (dim_t context_length)
	Set the maximum sequence length.
GemmaModelConfig &	withFP4Quantization ()
	FP4 quantization – FP4 weights, FP8 KV cache.
GemmaModelConfig &	withFP8Quantization ()
	FP8 quantization – FP8 weights, FP8 KV cache.
GemmaModelConfig &	withFullPrecision ()
	Full precision – BF16 weights, BF16 KV cache.
GemmaModelConfig &	withKvCacheCompression (KvCacheCompression kv)
	Set the KV cache compression mode independently.
GemmaModelConfig &	withLanguageModelHeadPositions (dim_t positions)
	Positions the language-model head evaluates per pass.
GemmaModelConfig &	withPrecisionPlan ()
	The family's designed per-role allocation, from a pre-quantized artifact.
GemmaModelConfig &	withWeightQuantization (WeightQuantization wq)
	Set the weight quantization mode independently.

Additional Inherited Members
Protected Attributes inherited from Mila::Dnn::LanguageModelConfig< GemmaModelConfig >
dim_t	context_length_
KvCacheCompression	kv_cache_compression_
dim_t	language_model_head_positions_
WeightQuantization	weight_quantization_

Detailed Description

Deployment configuration for Gemma language models.

Inherits all fluent setters and accessors from LanguageModelConfig<GemmaModelConfig>.

Constructor & Destructor Documentation

Mila::Dnn::GemmaModelConfig::GemmaModelConfig ( dim_t context_length )

inlineexplicit

Construct with a required context length.

Parameters

context_length Maximum sequence length in tokens. Must be > 0.

Exceptions

std::invalid_argument if context_length is zero.

The documentation for this struct was generated from the following file: