Is Cybertron compatible with nomic-ai models?

[faelp22]

Hello everyone, how are you?

Cybertron is compatible with the nomic-ai/nomic-embed-text-v1.5 model. Does anyone have an example of how I can use it?

I adapted this example to use the model

import (
	"bufio"
	"context"
	"fmt"
	"io"
	"os"

	"github.com/nlpodyssey/cybertron/pkg/models/bert"
	"github.com/nlpodyssey/cybertron/pkg/tasks"
	"github.com/nlpodyssey/cybertron/pkg/tasks/textencoding"
	"github.com/rs/zerolog"
	"github.com/rs/zerolog/log"
)

const limit = 10 // number of dimensions to show

// https://github.com/nlpodyssey/cybertron/tree/main/examples/textencoding

func main() {

	zerolog.SetGlobalLevel(zerolog.DebugLevel)

	modelsDir := "llms"
	modelName := "nomic-ai/nomic-embed-text-v1.5"

	m, err := tasks.Load[textencoding.Interface](&tasks.Config{ModelsDir: modelsDir, ModelName: modelName})
	if err != nil {
		log.Fatal().Err(err).Send()
	}

	fn := func(text string) error {
		result, err := m.Encode(context.Background(), text, int(bert.MeanPooling))
		if err != nil {
			return err
		}
		fmt.Println(result.Vector.Data().F64()[:limit])
		return nil
	}

	err = ForEachInput(os.Stdin, fn)
	if err != nil {
		log.Fatal().Err(err).Send()
	}
}

// ForEachInput calls the given callback function for each line of input.
func ForEachInput(r io.Reader, callback func(text string) error) error {
	scanner := bufio.NewScanner(r)
	for {
		fmt.Print("> ")
		scanner.Scan()
		text := scanner.Text()
		if text == "" {
			break
		}
		if err := callback(text); err != nil {
			return err
		}
	}
	return nil
}

Logs

{"level":"debug","file":"llms/nomic-ai/nomic-embed-text-v1.5/config.json","time":"2024-10-31T16:21:52-03:00","message":"model file already exists, skipping download"}
{"level":"fatal","error":"unsupported model type for download: \"nomic_bert\"","time":"2024-10-31T16:21:52-03:00"}

However, it is giving an error, I don't understand very well.

thank you for your attention

[matteo-grella]

I can implement the nomic_bert model as it would require minimal effort. Are you still interested in having this implemented?

[faelp22]

yes, it would be nice to have that

[faelp22]

@matteo-grella If you can teach me I can make the adjustment to the code.