Function ModelEmbedding

core/backend/embeddings.go:41–146

(ctx context.Context, s string, tokens []int, loader *model.ModelLoader, modelConfig config.ModelConfig, appConfig *config.ApplicationConfig)

}

func ModelEmbedding(ctx context.Context, s string, tokens []int, loader *model.ModelLoader, modelConfig config.ModelConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {

	// model.WithContext(ctx) overrides the app-context default set in
	// ModelOptions so distributed routing decisions reach the request's
	// X-LocalAI-Node holder via distributedhdr.Stamp.
	opts := ModelOptions(modelConfig, appConfig, model.WithContext(ctx))

	inferenceModel, err := loader.Load(opts...)
	if err != nil {
		recordModelLoadFailure(appConfig, modelConfig.Name, modelConfig.Backend, err, nil)
		return nil, err
	}

	var fn func() ([]float32, error)
	switch model := inferenceModel.(type) {
	case grpc.Backend:
		fn = func() ([]float32, error) {
			predictOptions := gRPCPredictOpts(modelConfig, loader.ModelPath)
			if len(tokens) > 0 {
				embeds := []int32{}

				for _, t := range tokens {
					embeds = append(embeds, int32(t))
				}
				predictOptions.EmbeddingTokens = embeds

				res, err := model.Embeddings(appConfig.Context, predictOptions)
				if err != nil {
					return nil, err
				}

				return res.Embeddings, nil
			}
			predictOptions.Embeddings = s

			res, err := model.Embeddings(appConfig.Context, predictOptions)
			if err != nil {
				return nil, err
			}

			return res.Embeddings, nil
		}
	default:
		fn = func() ([]float32, error) {
			return nil, fmt.Errorf("embeddings not supported by the backend")
		}
	}

	wrappedFn := func() ([]float32, error) {
		embeds, err := fn()
		if err != nil {
			return embeds, err
		}
		// Return embeddings as-is to preserve full dimensionality
		// Trailing zeros may be valid values in some embedding models
		return embeds, nil
	}