Module model 

GGUF-quantized EmbeddingGemma model internals.

Implements the full forward pass: token embedding → transformer layers (bidirectional attention) → mean pooling → dense projections → L2 normalization.

Structs

EmbeddingGemmaModel

The full EmbeddingGemma model loaded from GGUF + Dense layer safetensors.

Functions

l2_normalize

L2-normalizes a tensor along the last dimension.

mean_pool

Mean-pools token embeddings over the sequence dimension, respecting an attention mask.