text2vec-transformers

Overview

The text2vec-transformers module enables Weaviate to obtain vectors locally from text using a transformers-based model.

text2vec-transformers encapsulates models in Docker containers, which allows independent scaling on GPU-enabled hardware while keeping Weaviate on CPU-only hardware, as Weaviate is CPU-optimized.

Key notes:

This module is not available on Weaviate Cloud Services (WCS).
Enabling this module will enable the nearText search operator.
This module is only compatible with models encapsulated in a Docker container.
Pre-built images are available with popular models.
You can also use other models, such as:
- By building an image for any publicly available model from the Hugging Face model hub.
- By building an image for any model compatible with Hugging Face's AutoModel and AutoTokenizer.

Do you have GPU acceleration?

Transformer model inference speeds are usually about ten times faster with GPUs. If you have a GPU, use one of the GPU enabled models.

If you use text2vec-transformers without GPU acceleration, imports or nearText queries may become bottlenecks. The ONNX-enabled images can use ONNX Runtime for faster inference processing on CPUs. Look for the -onnx suffix in the image name.

Alternatively, consider one of the following options:

an API-based module such as text2vec-cohere or text2vec-openai
a local inference container such as text2vec-contextionary or text2vec-gpt4all

Weaviate instance configuration

Not applicable to WCS

This module is not available on Weaviate Cloud Services.

Docker Compose file

To use text2vec-transformers, you must enable it in your Docker Compose file (e.g. docker-compose.yml).

Use the configuration tool

While you can do so manually, we recommend using the Weaviate configuration tool to generate the Docker Compose file.

Parameters

Weaviate:

ENABLE_MODULES (Required): The modules to enable. Include text2vec-transformers to enable the module.
DEFAULT_VECTORIZER_MODULE (Optional): The default vectorizer module. You can set this to text2vec-transformers to make it the default for all classes.
TRANSFORMERS_INFERENCE_API (Required): The URL of the inference container.

Inference container:

image (Required): The image name of the inference container.
ENABLE_CUDA (Optional): Set to 1 to enable GPU usage. Default is 0 (CPU only).

Example

This configuration enables text2vec-transformers, sets it as the default vectorizer, and sets the parameters for the Transformers Docker container, including setting it to use sentence-transformers-multi-qa-MiniLM-L6-cos-v1 image and to disable CUDA acceleration.

version: '3.4'
services:
  weaviate:
    image: semitechnologies/weaviate:1.23.3
    restart: on-failure:0
    ports:
     - 8080:8080
     - 50051:50051
    environment:
      QUERY_DEFAULTS_LIMIT: 20
      AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: 'true'
      PERSISTENCE_DATA_PATH: "./data"
      ENABLE_MODULES: text2vec-transformers
      DEFAULT_VECTORIZER_MODULE: text2vec-transformers
      TRANSFORMERS_INFERENCE_API: http://t2v-transformers:8080
      CLUSTER_HOSTNAME: 'node1'
  t2v-transformers:
    image: semitechnologies/transformers-inference:sentence-transformers-multi-qa-MiniLM-L6-cos-v1
    environment:
      ENABLE_CUDA: 0 # set to 1 to enable
...

Have you enabled CUDA?

Make sure to enable CUDA if you have a compatible GPU available (ENABLE_CUDA=1) to take advantage of GPU acceleration.

Alternative: Run a separate container

As an alternative, you can run the inference container independently from Weaviate. To do so, you can:

Enable text2vec-transformers in your Docker Compose file,
Omit t2v-transformers parameters,
Run the inference container separately, e.g. using Docker, and
Set TRANSFORMERS_INFERENCE_API to the URL of the inference container.

For example, choose any of our pre-built transformers models and spin it up - for example using:

docker run -itp "8000:8080" semitechnologies/transformers-inference:sentence-transformers-multi-qa-MiniLM-L6-cos-v1

Then, for example if Weaviate is running outside of Docker, set TRANSFORMERS_INFERENCE_API="http://localhost:8000". Alternatively if Weaviate is part of the same Docker network, e.g. because they are part of the same docker-compose.yml file, you can use Docker networking/DNS, such as TRANSFORMERS_INFERENCE_API=http://t2v-transformers:8080.

Class configuration

You can configure how the module will behave in each class through the Weaviate schema.

Vectorization settings

You can set vectorizer behavior using the moduleConfig section under each class and property:

Class-level

vectorizer - what module to use to vectorize the data.
vectorizeClassName – whether to vectorize the class name. Default: true.
poolingStrategy – the pooling strategy to use. Default: masked_mean. Allowed values: masked_mean or cls. (Read more on this topic.)

Property-level

skip – whether to skip vectorizing the property altogether. Default: false
vectorizePropertyName – whether to vectorize the property name. Default: false

Example

{
  "classes": [
    {
      "class": "Document",
      "description": "A class called document",
      "vectorizer": "text2vec-transformers",
      "moduleConfig": {
        "text2vec-transformers": {
          "vectorizeClassName": false
        }
      },
      "properties": [
        {
          "name": "content",
          "dataType": [
            "text"
          ],
          "description": "Content that will be vectorized",
          "moduleConfig": {
            "text2vec-transformers": {
              "skip": false,
              "vectorizePropertyName": false
            }
          }
        }
      ],
    }
  ]
}

Model selection

To select a model, please point text2vec-transformers to the appropriate Docker container.

You can use one of our pre-built Docker images, or build your own (with just a few lines of code).

This allows you to use any suitable model from the Hugging Face model hub or your own custom model.

Option 1: Pre-built images

We have built images from publicly available models that in our opinion are well suited for semantic search. You can use any of the following:

Model Name	Image Name
`distilbert-base-uncased` (Info)	`semitechnologies/transformers-inference:distilbert-base-uncased`
`sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-paraphrase-multilingual-MiniLM-L12-v2`
`sentence-transformers/multi-qa-MiniLM-L6-cos-v1` (Info)	`semitechnologies/transformers-inference:sentence-transformers-multi-qa-MiniLM-L6-cos-v1`
`sentence-transformers/multi-qa-mpnet-base-cos-v1` (Info)	`semitechnologies/transformers-inference:sentence-transformers-multi-qa-mpnet-base-cos-v1`
`sentence-transformers/all-mpnet-base-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-all-mpnet-base-v2`
`sentence-transformers/all-MiniLM-L12-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-all-MiniLM-L12-v2`
`sentence-transformers/paraphrase-multilingual-mpnet-base-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-paraphrase-multilingual-mpnet-base-v2`
`sentence-transformers/all-MiniLM-L6-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-all-MiniLM-L6-v2`
`sentence-transformers/all-MiniLM-L6-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-all-MiniLM-L6-v2-onnx`
`sentence-transformers/multi-qa-distilbert-cos-v1` (Info)	`semitechnologies/transformers-inference:sentence-transformers-multi-qa-distilbert-cos-v1`
`sentence-transformers/gtr-t5-base` (Info)	`semitechnologies/transformers-inference:sentence-transformers-gtr-t5-base`
`sentence-transformers/gtr-t5-large` (Info)	`semitechnologies/transformers-inference:sentence-transformers-gtr-t5-large`
`google/flan-t5-base` (Info)	`semitechnologies/transformers-inference:google-flan-t5-base`
`google/flan-t5-large` (Info)	`semitechnologies/transformers-inference:google-flan-t5-large`
`BAAI/bge-small-en-v1.5` (Info)	`semitechnologies/transformers-inference:baai-bge-small-en-v1.5`
`BAAI/bge-base-en-v1.5` (Info)	`semitechnologies/transformers-inference:baai-bge-base-en-v1.5`
DPR Models
`facebook/dpr-ctx_encoder-single-nq-base` (Info)	`semitechnologies/transformers-inference:facebook-dpr-ctx_encoder-single-nq-base`
`facebook/dpr-question_encoder-single-nq-base` (Info)	`semitechnologies/transformers-inference:facebook-dpr-question_encoder-single-nq-base`
`vblagoje/dpr-ctx_encoder-single-lfqa-wiki` (Info)	`semitechnologies/transformers-inference:vblagoje-dpr-ctx_encoder-single-lfqa-wiki`
`vblagoje/dpr-question_encoder-single-lfqa-wiki` (Info)	`semitechnologies/transformers-inference:vblagoje-dpr-question_encoder-single-lfqa-wiki`
Bar-Ilan University NLP Lab Models
`biu-nlp/abstract-sim-sentence` (Info)	`semitechnologies/transformers-inference:biu-nlp-abstract-sim-sentence`
`biu-nlp/abstract-sim-query` (Info)	`semitechnologies/transformers-inference:biu-nlp-abstract-sim-query`

ONNX-enabled images (CPU only)

We also provide ONNX-enabled images for some models. These images use ONNX Runtime for faster inference on CPUs. They are quantized for ARM64 and AMD64 (AVX2) hardware.

Look for the -onnx suffix in the image name.

Model Name	Image Name
`sentence-transformers/all-MiniLM-L6-v2` (Info)	`semitechnologies/transformers-inference:sentence-transformers-all-MiniLM-L6-v2-onnx`
`BAAI/bge-small-en-v1.5` (Info)	`semitechnologies/transformers-inference:baai-bge-small-en-v1.5-onnx`
`BAAI/bge-base-en-v1.5` (Info)	[`semitechnologies/transformers-inference:baai-bge-base-en-v1.5-onnx`

Is your preferred model missing?

If your preferred model is missing, please open an issue to include it or build a custom image as outlined below.

How to set the version

You can explicitly set the version through a suffix.

Use -1.0.0 to pin to a specific version. E.g. semitechnologies/transformers-inference:distilbert-base-uncased-1.0.0 will always use the version with git tag 1.0.0 of the distilbert-base-uncased repository.
You can explicitly set -latest to always use the latest version, however this is the default behavior.

Option 2: A Hugging Face model

You can build a Docker image to use any public model from the Hugging Face model hub with a two-line Dockerfile. In the following example, we are going to build a custom image for the distilroberta-base model.

Step 1: Create a `Dockerfile`

Create a new Dockerfile. We will name it distilroberta.Dockerfile. Add the following lines to it:

FROM semitechnologies/transformers-inference:custom
RUN MODEL_NAME=distilroberta-base ./download.py

Step 2: Build and tag your Dockerfile.

We will tag our Dockerfile as distilroberta-inference:

docker build -f distilroberta.Dockerfile -t distilroberta-inference .

Step 3: Use the image

You can now push your image to your favorite registry or reference it locally in your Weaviate docker-compose.yml using the Docker tag distilroberta-inference.

Option 3: A private or local model

You can build a Docker image which supports any model which is compatible with Hugging Face's AutoModel and AutoTokenizer.

In the following example, we are going to build a custom image for a non-public model which we have locally stored at ./my-model.

Create a new Dockerfile (you do not need to clone this repository, any folder on your machine is fine), we will name it my-model.Dockerfile. Add the following lines to it:

FROM semitechnologies/transformers-inference:custom
COPY ./my-model /app/models/model

The above will make sure that your model end ups in the image at /app/models/model. This path is important, so that the application can find the model.

Now you just need to build and tag your Dockerfile, we will tag it as my-model-inference:

docker build -f my-model.Dockerfile -t my-model-inference .

That's it! You can now push your image to your favorite registry or reference it locally in your Weaviate docker-compose.yml using the Docker tag my-model-inference.

To debug and test if your inference container is working correctly, you can send queries to the vectorizer module's inference container directly, so you can see exactly what vectors it would produce for which input.

To do so – you need to expose the inference container in your Docker Compose file – add something like this:

ports:
  - "9090:8080"

to your text2vec-transformers.

Then you can send REST requests to it directly, e.g.:

curl localhost:9090/vectors -H 'Content-Type: application/json' -d '{"text": "foo bar"}'

and it will print the created vector directly.

Usage

Example

import weaviate
import weaviate.classes as wvc
from weaviate.collections.classes.grpc import Move
import os

client = weaviate.connect_to_local()

publications = client.collections.get("Publication")

response = publications.query.near_text(
    query="fashion",
    distance=0.6,
    move_to=Move(force=0.85, concepts="haute couture"),
    move_away=Move(force=0.45, concepts="finance"),
    return_metadata=wvc.MetadataQuery(distance=True),
    limit=2
)

for o in response.objects:
    print(o.properties)
    print(o.metadata)

import weaviate

client = weaviate.Client("http://localhost:8080")

nearText = {
  "concepts": ["fashion"],
  "distance": 0.6, # prior to v1.14 use "certainty" instead of "distance"
  "moveAwayFrom": {
    "concepts": ["finance"],
    "force": 0.45
  },
  "moveTo": {
    "concepts": ["haute couture"],
    "force": 0.85
  }
}

result = (
  client.query
  .get("Publication", "name")
  .with_additional(["certainty OR distance"]) # note that certainty is only supported if distance==cosine
  .with_near_text(nearText)
  .do()
)

print(result)

import weaviate from 'weaviate-ts-client';

const client = weaviate.client({
  scheme: 'http',
  host: 'localhost:8080',
});

const response = await client.graphql
  .get()
  .withClassName('Publication')
  .withFields('name _additional{certainty distance}') // note that certainty is only supported if distance==cosine
  .withNearText({
    concepts: ['fashion'],
    distance: 0.6,  // prior to v1.14 use certainty instead of distance
    moveAwayFrom: {
      concepts: ['finance'],
      force: 0.45,
    },
    moveTo: {
      concepts: ['haute couture'],
      force: 0.85,
    },
  })
  .do();
console.log(response);

package main

import (
  "context"
  "fmt"

  "github.com/weaviate/weaviate-go-client/v4/weaviate"
  "github.com/weaviate/weaviate-go-client/v4/weaviate/graphql"
)

func main() {
  cfg := weaviate.Config{
    Host:   "localhost:8080",
    Scheme: "http",
  }
  client, err := weaviate.NewClient(cfg)
  if err != nil {
    panic(err)
  }

  className := "Publication"

  name := graphql.Field{Name: "name"}
  _additional := graphql.Field{
    Name: "_additional", Fields: []graphql.Field{
      {Name: "certainty"}, // only supported if distance==cosine
      {Name: "distance"},  // always supported
    },
  }

  concepts := []string{"fashion"}
  distance := float32(0.6)
  moveAwayFrom := &graphql.MoveParameters{
    Concepts: []string{"finance"},
    Force:    0.45,
  }
  moveTo := &graphql.MoveParameters{
    Concepts: []string{"haute couture"},
    Force:    0.85,
  }
  nearText := client.GraphQL().NearTextArgBuilder().
    WithConcepts(concepts).
    WithDistance(distance). // use WithCertainty(certainty) prior to v1.14
    WithMoveTo(moveTo).
    WithMoveAwayFrom(moveAwayFrom)

  ctx := context.Background()

  result, err := client.GraphQL().Get().
    WithClassName(className).
    WithFields(name, _additional).
    WithNearText(nearText).
    Do(ctx)

  if err != nil {
    panic(err)
  }
  fmt.Printf("%v", result)
}

package io.weaviate;

import io.weaviate.client.Config;
import io.weaviate.client.WeaviateClient;
import io.weaviate.client.base.Result;
import io.weaviate.client.v1.graphql.model.GraphQLResponse;
import io.weaviate.client.v1.graphql.query.argument.NearTextArgument;
import io.weaviate.client.v1.graphql.query.argument.NearTextMoveParameters;
import io.weaviate.client.v1.graphql.query.fields.Field;

public class App {
  public static void main(String[] args) {
    Config config = new Config("http", "localhost:8080");
    WeaviateClient client = new WeaviateClient(config);

    NearTextMoveParameters moveTo = NearTextMoveParameters.builder()
      .concepts(new String[]{ "haute couture" }).force(0.85f).build();

    NearTextMoveParameters moveAway = NearTextMoveParameters.builder()
      .concepts(new String[]{ "finance" }).force(0.45f)
      .build();

    NearTextArgument nearText = client.graphQL().arguments().nearTextArgBuilder()
      .concepts(new String[]{ "fashion" })
      .distance(0.6f) // use .certainty(0.7f) prior to v1.14
      .moveTo(moveTo)
      .moveAwayFrom(moveAway)
      .build();

    Field name = Field.builder().name("name").build();
    Field _additional = Field.builder()
      .name("_additional")
      .fields(new Field[]{
        Field.builder().name("certainty").build(), // only supported if distance==cosine
        Field.builder().name("distance").build(),  // always supported
      }).build();

    Result<GraphQLResponse> result = client.graphQL().get()
      .withClassName("Publication")
      .withFields(name, _additional)
      .withNearText(nearText)
      .run();

    if (result.hasErrors()) {
      System.out.println(result.getError());
      return;
    }
    System.out.println(result.getResult());
  }
}

# Note: Under nearText, use `certainty` instead of distance prior to v1.14
# Under _additional, `certainty` is only supported if distance==cosine, but `distance` is always supported
echo '{
  "query": "{
    Get {
      Publication(
        nearText: {
          concepts: [\"fashion\"],
          distance: 0.6,
          moveAwayFrom: {
            concepts: [\"finance\"],
            force: 0.45
          },
          moveTo: {
            concepts: [\"haute couture\"],
            force: 0.85
          }
        }
      ) {
        name
        _additional {
          certainty
          distance
        }
      }
    }
  }"
}' | curl \
    -X POST \
    -H 'Content-Type: application/json' \
    -H 'Authorization: Bearer learn-weaviate' \
    -H "X-OpenAI-Api-Key: $OPENAI_API_KEY" \
    -d @- \
    https://edu-demo.weaviate.network/v1/graphql

{
  Get{
    Publication(
      nearText: {
        concepts: ["fashion"],
        distance: 0.6 # prior to v1.14 use "certainty" instead of "distance"
        moveAwayFrom: {
          concepts: ["finance"],
          force: 0.45
        },
        moveTo: {
          concepts: ["haute couture"],
          force: 0.85
        }
      }
    ){
      name
      _additional {
        certainty # only supported if distance==cosine.
        distance  # always supported
      }
    }
  }
}

Chunking

The text2vec-transformers module can automatically chunk text based on the model's maximum token length before it is passed to the model. It will then return the pooled vectors.

See HuggingFaceVectorizer.vectorizer() for the exact implementation.

Model licenses

The text2vec-transformers module is compatible with various models. Each of the models has its own license. For detailed information, please review the license for the model you are using in the Hugging Face Model Hub.

It is your responsibility to evaluate whether the terms of its license(s), if any, are appropriate for your intended use.

Release notes

The release notes are here.

Overview​

Weaviate instance configuration​

Docker Compose file​

Parameters​

Example​

Alternative: Run a separate container​

Class configuration​

Vectorization settings​

Class-level​

Property-level​

Example​

Model selection​

Option 1: Pre-built images​

ONNX-enabled images (CPU only)​

Is your preferred model missing?​

How to set the version​

Option 2: A Hugging Face model​

Step 1: Create a Dockerfile​

Step 2: Build and tag your Dockerfile.​

Step 3: Use the image​

Option 3: A private or local model​

Usage​

Example​

Chunking​

Model licenses​

Release notes​

Overview

Weaviate instance configuration

Docker Compose file

Parameters

Example

Alternative: Run a separate container

Class configuration

Vectorization settings

Class-level

Property-level

Example

Model selection

Option 1: Pre-built images

ONNX-enabled images (CPU only)

Is your preferred model missing?

How to set the version

Option 2: A Hugging Face model

Step 1: Create a `Dockerfile`

Step 2: Build and tag your Dockerfile.

Step 3: Use the image

Option 3: A private or local model

Usage

Example

Chunking

Model licenses

Release notes