Module stella_en_v5 

Stella v5 model implementation.

Stella is a dense text embedding model optimized for retrieval and similarity tasks. This implementation provides support for multiple embedding dimensions.

Key characteristics:

• Dense text embeddings optimized for similarity search
• Multiple output dimension support (256 to 8192)
• Grouped query attention (GQA)
• RMSNorm for layer normalization
• Rotary positional embeddings (RoPE)

References:

• MRL Framework
• Model Card

Structs

Config

EmbedHead

EmbeddingModel

Embeddings

Model

Enums

EmbedDim

An enum variant representing the Embedding head dimensions stella is trained on As the model-card suggests, D1024 is good enough for most cases

ModelVariant