skainet-io-gguf/sk.ainet.io.gguf.gemma

Package-level declarations

Types

Gemma3nConfigParser

object Gemma3nConfigParser

Parser for HuggingFace Gemma 3n config.json files.

Gemma3nGgufTensorNames

object Gemma3nGgufTensorNames : TensorNameMapper

GGUF tensor naming for Gemma 3n models.

Gemma3nLayerWeights

data class Gemma3nLayerWeights<T : DType>(val inputLayernorm: Tensor<T, Float>, val wq: Tensor<T, Float>, val wk: Tensor<T, Float>, val wv: Tensor<T, Float>, val wo: Tensor<T, Float>, val postAttentionLayernorm: Tensor<T, Float>, val gateProj: Tensor<T, Float>, val upProj: Tensor<T, Float>, val downProj: Tensor<T, Float>, val perLayerInput: Tensor<T, Float>?, val perLayerOutput: Tensor<T, Float>?)

Weights for a single Gemma 3n transformer layer.

Gemma3nModelMetadata

data class Gemma3nModelMetadata(val architecture: String, val embeddingLength: Int, val perLayerEmbeddingLength: Int, val contextLength: Int, val blockCount: Int, val headCount: Int, val kvHeadCount: Int, val feedForwardLengths: List<Int>, val headDim: Int, val vocabSize: Int, val slidingWindow: Int, val ropeBaseLocal: Float, val ropeBaseGlobal: Float, val kvSharedLayers: Int, val layerPattern: List<String>)

Metadata for Gemma 3n models extracted from GGUF files.

Gemma3nRuntimeWeights

data class Gemma3nRuntimeWeights<T : DType>(val metadata: Gemma3nModelMetadata, val tokenEmbedding: Tensor<T, Float>, val ropeFreqReal: Tensor<T, Float>?, val ropeFreqImag: Tensor<T, Float>?, val layers: List<Gemma3nLayerWeights<T>>, val finalNorm: Tensor<T, Float>, val lmHead: Tensor<T, Float>, val quantTypes: Map<String, GGMLQuantizationType> = emptyMap())

Complete runtime weights for Gemma 3n model.

Gemma3nSafeTensorsWeightLoader

class Gemma3nSafeTensorsWeightLoader(indexPath: String)

Loads Gemma 3n weights from HuggingFace SafeTensors format.

Gemma3nTensorNames

object Gemma3nTensorNames

Tensor name constants for Gemma 3n GGUF format.

Gemma3nWeightLoader

class Gemma3nWeightLoader

Adapter that loads Gemma 3n weights from GGUF files.

Gemma3nWeightMapper

object Gemma3nWeightMapper

Maps raw weights to runtime structure with shape validation.

Gemma3nWeights

data class Gemma3nWeights<T : DType, V>(val metadata: Gemma3nModelMetadata, val tensors: Map<String, Tensor<T, V>>, val quantTypes: Map<String, GGMLQuantizationType> = emptyMap())

Raw weights loaded from GGUF, before mapping to runtime structure.

LayerType

enum LayerType : Enum<LayerType>

Type of attention layer in Gemma 3n.