strickvl's picture
Add new SentenceTransformer model.
2b5fc8f verified
metadata
base_model: sentence-transformers/all-MiniLM-L6-v2
datasets: []
language:
  - en
library_name: sentence-transformers
license: apache-2.0
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@100
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:1490
  - loss:MatryoshkaLoss
  - loss:MultipleNegativesRankingLoss
widget:
  - source_sentence: >-
      How can I configure the orchestrator settings for each cloud provider in
      ZenML?
    sentences:
      - >-
        . If not set, the cluster will not be autostopped.down: Tear down the
        cluster after all jobs finish (successfully or abnormally). If
        idle_minutes_to_autostop is also set, the cluster will be torn down
        after the specified idle time. Note that if errors occur during
        provisioning/data syncing/setting up, the cluster will not be torn down
        for debugging purposes.


        stream_logs: If True, show the logs in the terminal as they are
        generated while the cluster is running.


        docker_run_args: Additional arguments to pass to the docker run command.
        For example, ['--gpus=all'] to use all GPUs available on the VM.


        The following code snippets show how to configure the orchestrator
        settings for each cloud provider:


        Code Example:


        from
        zenml.integrations.skypilot_aws.flavors.skypilot_orchestrator_aws_vm_flavor
        import SkypilotAWSOrchestratorSettings


        skypilot_settings = SkypilotAWSOrchestratorSettings(


        cpus="2",


        memory="16",


        accelerators="V100:2",


        accelerator_args={"tpu_vm": True, "runtime_version": "tpu-vm-base"},


        use_spot=True,


        spot_recovery="recovery_strategy",


        region="us-west-1",


        zone="us-west1-a",


        image_id="ami-1234567890abcdef0",


        disk_size=100,


        disk_tier="high",


        cluster_name="my_cluster",


        retry_until_up=True,


        idle_minutes_to_autostop=60,


        down=True,


        stream_logs=True


        docker_run_args=["--gpus=all"]


        @pipeline(


        settings={


        "orchestrator.vm_aws": skypilot_settings


        Code Example:


        from
        zenml.integrations.skypilot_gcp.flavors.skypilot_orchestrator_gcp_vm_flavor
        import SkypilotGCPOrchestratorSettings


        skypilot_settings = SkypilotGCPOrchestratorSettings(


        cpus="2",


        memory="16",


        accelerators="V100:2",


        accelerator_args={"tpu_vm": True, "runtime_version": "tpu-vm-base"},


        use_spot=True,


        spot_recovery="recovery_strategy",


        region="us-west1",


        zone="us-west1-a",


        image_id="ubuntu-pro-2004-focal-v20231101",


        disk_size=100,


        disk_tier="high",


        cluster_name="my_cluster",


        retry_until_up=True,


        idle_minutes_to_autostop=60,


        down=True,


        stream_logs=True


        @pipeline(


        settings={


        "orchestrator.vm_gcp": skypilot_settings
      - >-
        he Post-execution workflow has changed as follows:The get_pipelines and
        get_pipeline methods have been moved out of the Repository (i.e. the new
        Client ) class and lie directly in the post_execution module now. To use
        the user has to do:


        from zenml.post_execution import get_pipelines, get_pipeline


        New methods to directly get a run have been introduced: get_run and
        get_unlisted_runs method has been introduced to get unlisted runs.


        Usage remains largely similar. Please read the new docs for
        post-execution to inform yourself of what further has changed.


        How to migrate: Replace all post-execution workflows from the paradigm
        of Repository.get_pipelines or Repository.get_pipeline_run to the
        corresponding post_execution methods.


        πŸ“‘Future Changes


        While this rehaul is big and will break previous releases, we do have
        some more work left to do. However we also expect this to be the last
        big rehaul of ZenML before our 1.0.0 release, and no other release will
        be so hard breaking as this one. Currently planned future breaking
        changes are:


        Following the metadata store, the secrets manager stack component might
        move out of the stack.


        ZenML StepContext might be deprecated.


        🐞 Reporting Bugs


        While we have tried our best to document everything that has changed, we
        realize that mistakes can be made and smaller changes overlooked. If
        this is the case, or you encounter a bug at any time, the ZenML core
        team and community are available around the clock on the growing Slack
        community.


        For bug reports, please also consider submitting a GitHub Issue.


        Lastly, if the new changes have left you desiring a feature, then
        consider adding it to our public feature voting board. Before doing so,
        do check what is already on there and consider upvoting the features you
        desire the most.


        PreviousMigration guide


        NextMigration guide 0.23.0 β†’ 0.30.0


        Last updated 12 days ago
      - >-
        nML, namely an orchestrator and an artifact store.Keep in mind, that
        each one of these components is built on top of base abstractions and is
        completely extensible.


        Orchestrator


        An Orchestrator is a workhorse that coordinates all the steps to run in
        a pipeline. Since pipelines can be set up with complex combinations of
        steps with various asynchronous dependencies between them, the
        orchestrator acts as the component that decides what steps to run and
        when to run them.


        ZenML comes with a default local orchestrator designed to run on your
        local machine. This is useful, especially during the exploration phase
        of your project. You don't have to rent a cloud instance just to try out
        basic things.


        Artifact Store


        An Artifact Store is a component that houses all data that pass through
        the pipeline as inputs and outputs. Each artifact that gets stored in
        the artifact store is tracked and versioned and this allows for
        extremely useful features like data caching which speeds up your
        workflows.


        Similar to the orchestrator, ZenML comes with a default local artifact
        store designed to run on your local machine. This is useful, especially
        during the exploration phase of your project. You don't have to set up a
        cloud storage system to try out basic things.


        Flavor


        ZenML provides a dedicated base abstraction for each stack component
        type. These abstractions are used to develop solutions, called Flavors,
        tailored to specific use cases/tools. With ZenML installed, you get
        access to a variety of built-in and integrated Flavors for each
        component type, but users can also leverage the base abstractions to
        create their own custom flavors.


        Stack Switching


        When it comes to production-grade solutions, it is rarely enough to just
        run your workflow locally without including any cloud infrastructure.
  - source_sentence: How can I fetch artifacts from other pipelines within a step using ZenML?
    sentences:
      - >2-
                                                         ┃┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨

        ┃ EXPIRES IN       β”‚
        N/A                                                                     
        ┃


        ┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨


        ┃ OWNER            β”‚
        default                                                                 
        ┃


        ┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨


        ┃ WORKSPACE        β”‚
        default                                                                 
        ┃


        ┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨


        ┃ SHARED           β”‚
        βž–                                                                      
        ┃


        ┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨


        ┃ CREATED_AT       β”‚ 2023-05-19
        09:15:12.882929                                               ┃


        ┠──────────────────┼──────────────────────────────────────────────────────────────────────────┨


        ┃ UPDATED_AT       β”‚ 2023-05-19
        09:15:12.882930                                               ┃


        ┗━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛


        Configuration


        ┏━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━┓


        ┃ PROPERTY          β”‚ VALUE      ┃


        ┠───────────────────┼────────────┨


        ┃ project_id        β”‚ zenml-core ┃


        ┠───────────────────┼────────────┨


        ┃ user_account_json β”‚ [HIDDEN]   ┃


        ┗━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━┛


        Local client provisioning


        The local gcloud CLI, the Kubernetes kubectl CLI and the Docker CLI can
        be configured with credentials extracted from or generated by a
        compatible GCP Service Connector. Please note that unlike the
        configuration made possible through the GCP CLI, the Kubernetes and
        Docker credentials issued by the GCP Service Connector have a short
        lifetime and will need to be regularly refreshed. This is a byproduct of
        implementing a high-security profile.
      - >-
        gmax(prediction.numpy())


        return classes[maxindex]The custom predict function should get the model
        and the input data as arguments and return the model predictions. ZenML
        will automatically take care of loading the model into memory and
        starting the seldon-core-microservice that will be responsible for
        serving the model and running the predict function.


        After defining your custom predict function in code, you can use the
        seldon_custom_model_deployer_step to automatically build your function
        into a Docker image and deploy it as a model server by setting the
        predict_function argument to the path of your custom_predict function:


        from zenml.integrations.seldon.steps import
        seldon_custom_model_deployer_step


        from zenml.integrations.seldon.services import SeldonDeploymentConfig


        from zenml import pipeline


        @pipeline


        def seldon_deployment_pipeline():


        model = ...


        seldon_custom_model_deployer_step(


        model=model,


        predict_function="<PATH.TO.custom_predict>",  # TODO: path to custom
        code


        service_config=SeldonDeploymentConfig(


        model_name="<MODEL_NAME>",  # TODO: name of the deployed model


        replicas=1,


        implementation="custom",


        resources=SeldonResourceRequirements(


        limits={"cpu": "200m", "memory": "250Mi"}


        ),


        serviceAccountName="kubernetes-service-account",


        ),


        Advanced Custom Code Deployment with Seldon Core Integration


        Before creating your custom model class, you should take a look at the
        custom Python model section of the Seldon Core documentation.


        The built-in Seldon Core custom deployment step is a good starting point
        for deploying your custom models. However, if you want to deploy more
        than the trained model, you can create your own custom class and a
        custom step to achieve this.


        See the ZenML custom Seldon model class as a reference.


        PreviousMLflow


        NextBentoML


        Last updated 15 days ago
      - >-
        Get arbitrary artifacts in a step


        Not all artifacts need to come through the step interface from direct
        upstream steps.


        As described in the metadata guide, the metadata can be fetched with the
        client, and this is how you would use it to fetch it within a step. This
        allows you to fetch artifacts from other upstream steps or even
        completely different pipelines.


        from zenml.client import Client


        from zenml import step


        @step


        def my_step():


        client = Client()


        # Directly fetch an artifact


        output = client.get_artifact_version("my_dataset", "my_version")


        output.run_metadata["accuracy"].value


        This is one of the ways you can access artifacts that have already been
        created and stored in the artifact store. This can be useful when you
        want to use artifacts from other pipelines or steps that are not
        directly upstream.


        See Also


        Managing artifacts - learn about the ExternalArtifact type and how to
        pass artifacts between steps.


        PreviousOrganize data with tags


        NextHandle custom data types


        Last updated 15 days ago
  - source_sentence: Where can I find more information about using Feast in ZenML?
    sentences:
      - >-
        hat's described on the feast page: How to use it?.PreviousDevelop a
        Custom Model Registry


        NextFeast


        Last updated 1 year ago
      - >-
        other remote stack components also running in AWS.This method uses the
        implicit AWS authentication available in the environment where the ZenML
        code is running. On your local machine, this is the quickest way to
        configure an S3 Artifact Store. You don't need to supply credentials
        explicitly when you register the S3 Artifact Store, as it leverages the
        local credentials and configuration that the AWS CLI stores on your
        local machine. However, you will need to install and set up the AWS CLI
        on your machine as a prerequisite, as covered in the AWS CLI
        documentation, before you register the S3 Artifact Store.


        Certain dashboard functionality, such as visualizing or deleting
        artifacts, is not available when using an implicitly authenticated
        artifact store together with a deployed ZenML server because the ZenML
        server will not have permission to access the filesystem.


        The implicit authentication method also needs to be coordinated with
        other stack components that are highly dependent on the Artifact Store
        and need to interact with it directly to work. If these components are
        not running on your machine, they do not have access to the local AWS
        CLI configuration and will encounter authentication failures while
        trying to access the S3 Artifact Store:


        Orchestrators need to access the Artifact Store to manage pipeline
        artifacts


        Step Operators need to access the Artifact Store to manage step-level
        artifacts


        Model Deployers need to access the Artifact Store to load served models


        To enable these use-cases, it is recommended to use an AWS Service
        Connector to link your S3 Artifact Store to the remote S3 bucket.


        To set up the S3 Artifact Store to authenticate to AWS and access an S3
        bucket, it is recommended to leverage the many features provided by the
        AWS Service Connector such as auto-configuration, best security
        practices regarding long-lived credentials and fine-grained access
        control and reusing the same credentials across multiple stack
        components.
      - >2-
         us know!

        Configuration at pipeline or step levelWhen running your ZenML pipeline
        with the Sagemaker orchestrator, the configuration set when configuring
        the orchestrator as a ZenML component will be used by default. However,
        it is possible to provide additional configuration at the pipeline or
        step level. This allows you to run whole pipelines or individual steps
        with alternative configurations. For example, this allows you to run the
        training process with a heavier, GPU-enabled instance type, while
        running other steps with lighter instances.


        Additional configuration for the Sagemaker orchestrator can be passed
        via SagemakerOrchestratorSettings. Here, it is possible to configure
        processor_args, which is a dictionary of arguments for the Processor.
        For available arguments, see the Sagemaker documentation . Currently, it
        is not possible to provide custom configuration for the following
        attributes:


        image_uri


        instance_count


        sagemaker_session


        entrypoint


        base_job_name


        env


        For example, settings can be provided in the following way:


        sagemaker_orchestrator_settings = SagemakerOrchestratorSettings(


        processor_args={


        "instance_type": "ml.t3.medium",


        "volume_size_in_gb": 30


        They can then be applied to a step as follows:


        @step(settings={"orchestrator.sagemaker":
        sagemaker_orchestrator_settings})


        For example, if your ZenML component is configured to use ml.c5.xlarge
        with 400GB additional storage by default, all steps will use it except
        for the step above, which will use ml.t3.medium with 30GB additional
        storage.


        Check out this docs page for more information on how to specify settings
        in general.


        For more information and a full list of configurable attributes of the
        Sagemaker orchestrator, check out the SDK Docs .


        S3 data access in ZenML steps
  - source_sentence: How is the AWS region specified in the configuration for ZenML?
    sentences:
      - >-
        ge or if the ZenML version doesn't change at all).a backup file or
        database is created before every database migration attempt (i.e. during
        every Helm upgrade). If a backup already exists (i.e. persisted in a
        persistent volume or backup database), it is overwritten.


        the persistent backup file or database is cleaned up after the migration
        is completed successfully or if the database doesn't need to undergo a
        migration. This includes backups created by previous failed migration
        attempts.


        the persistent backup file or database is NOT cleaned up after a failed
        migration. This allows the user to manually inspect and/or apply the
        backup if the automatic recovery fails.


        The following example shows how to configure the ZenML server to use a
        persistent volume to store the database dump file:


        zenml:


        # ...


        database:


        url: "mysql://admin:[email protected]:3306/zenml"


        # Configure the database backup strategy


        backupStrategy: dump-file


        backupPVStorageSize: 1Gi


        podSecurityContext:


        fsGroup: 1000 # if you're using a PVC for backup, this should
        necessarily be set.


        PreviousDeploy with Docker


        NextDeploy using HuggingFace Spaces


        Last updated 15 days ago
      - >-
        🌲Control logging


        Configuring ZenML's default logging behavior


        ZenML produces various kinds of logs:


        The ZenML Server produces server logs (like any FastAPI server).


        The Client or Runner environment produces logs, for example after
        running a pipeline. These are steps that are typically before, after,
        and during the creation of a pipeline run.


        The Execution environment (on the orchestrator level) produces logs when
        it executes each step of a pipeline. These are logs that are typically
        written in your steps using the python logging module.


        This section talks about how users can control logging behavior in these
        various environments.


        PreviousTrain with GPUs


        NextView logs on the dashboard


        Last updated 19 days ago
      - >2-
                                                         ┃┠──────────────────┼─────────────────────────────────────────────────────────────────────┨

        ┃ SHARED           β”‚
        βž–                                                                  ┃


        ┠──────────────────┼─────────────────────────────────────────────────────────────────────┨


        ┃ CREATED_AT       β”‚ 2023-06-19
        18:12:42.066053                                          ┃


        ┠──────────────────┼─────────────────────────────────────────────────────────────────────┨


        ┃ UPDATED_AT       β”‚ 2023-06-19
        18:12:42.066055                                          ┃


        ┗━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛


        Configuration


        ┏━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━┓


        ┃ PROPERTY              β”‚ VALUE     ┃


        ┠───────────────────────┼───────────┨


        ┃ region                β”‚ us-east-1 ┃


        ┠───────────────────────┼───────────┨


        ┃ aws_access_key_id     β”‚ [HIDDEN]  ┃


        ┠───────────────────────┼───────────┨


        ┃ aws_secret_access_key β”‚ [HIDDEN]  ┃


        ┗━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━┛


        AWS Secret Key


        Long-lived AWS credentials consisting of an AWS access key ID and secret
        access key associated with an AWS IAM user or AWS account root user (not
        recommended).


        This method is preferred during development and testing due to its
        simplicity and ease of use. It is not recommended as a direct
        authentication method for production use cases because the clients have
        direct access to long-lived credentials and are granted the full set of
        permissions of the IAM user or AWS account root user associated with the
        credentials. For production, it is recommended to use the AWS IAM Role,
        AWS Session Token, or AWS Federation Token authentication method
        instead.


        An AWS region is required and the connector may only be used to access
        AWS resources in the specified region.


        If you already have the local AWS CLI set up with these credentials,
        they will be automatically picked up when auto-configuration is used
        (see the example below).
  - source_sentence: >-
      Can you explain how the `query_similar_docs` function handles document
      reranking?
    sentences:
      - >-
        ry_similar_docs(


        question: str,


        url_ending: str,use_reranking: bool = False,


        returned_sample_size: int = 5,


        ) -> Tuple[str, str, List[str]]:


        """Query similar documents for a given question and URL ending."""


        embedded_question = get_embeddings(question)


        db_conn = get_db_conn()


        num_docs = 20 if use_reranking else returned_sample_size


        # get (content, url) tuples for the top n similar documents


        top_similar_docs = get_topn_similar_docs(


        embedded_question, db_conn, n=num_docs, include_metadata=True


        if use_reranking:


        reranked_docs_and_urls = rerank_documents(question, top_similar_docs)[


        :returned_sample_size


        urls = [doc[1] for doc in reranked_docs_and_urls]


        else:


        urls = [doc[1] for doc in top_similar_docs]  # Unpacking URLs


        return (question, url_ending, urls)


        We get the embeddings for the question being passed into the function
        and connect to our PostgreSQL database. If we're using reranking, we get
        the top 20 documents similar to our query and rerank them using the
        rerank_documents helper function. We then extract the URLs from the
        reranked documents and return them. Note that we only return 5 URLs, but
        in the case of reranking we get a larger number of documents and URLs
        back from the database to pass to our reranker, but in the end we always
        choose the top five reranked documents to return.


        Now that we've added reranking to our pipeline, we can evaluate the
        performance of our reranker and see how it affects the quality of the
        retrieved documents.


        Code Example


        To explore the full code, visit the Complete Guide repository and for
        this section, particularly the eval_retrieval.py file.


        PreviousUnderstanding reranking


        NextEvaluating reranking performance


        Last updated 15 days ago
      - >-
        uter vision that expect a single dataset as input.model drift checks
        require two datasets and a mandatory model as input. This list includes
        a subset of the model evaluation checks provided by Deepchecks for
        tabular data and for computer vision that expect two datasets as input:
        target and reference.


        This structure is directly reflected in how Deepchecks can be used with
        ZenML: there are four different Deepchecks standard steps and four
        different ZenML enums for Deepchecks checks . The Deepchecks Data
        Validator API is also modeled to reflect this same structure.


        A notable characteristic of Deepchecks is that you don't need to
        customize the set of Deepchecks tests that are part of a test suite.
        Both ZenML and Deepchecks provide sane defaults that will run all
        available Deepchecks tests in a given category with their default
        conditions if a custom list of tests and conditions are not provided.


        There are three ways you can use Deepchecks in your ZenML pipelines that
        allow different levels of flexibility:


        instantiate, configure and insert one or more of the standard Deepchecks
        steps shipped with ZenML into your pipelines. This is the easiest way
        and the recommended approach, but can only be customized through the
        supported step configuration parameters.


        call the data validation methods provided by the Deepchecks Data
        Validator in your custom step implementation. This method allows for
        more flexibility concerning what can happen in the pipeline step, but
        you are still limited to the functionality implemented in the Data
        Validator.


        use the Deepchecks library directly in your custom step implementation.
        This gives you complete freedom in how you are using Deepchecks'
        features.


        You can visualize Deepchecks results in Jupyter notebooks or view them
        directly in the ZenML dashboard.


        Warning! Usage in remote orchestrators
      - >2-
         use for the database connection.
        database_ssl_ca:# The path to the client SSL certificate to use for the
        database connection.

        database_ssl_cert:


        # The path to the client SSL key to use for the database connection.

        database_ssl_key:


        # Whether to verify the database server SSL certificate.

        database_ssl_verify_server_cert:


        Run the deploy command and pass the config file above to it.Copyzenml
        deploy --config=/PATH/TO/FILENote To be able to run the deploy command,
        you should have your cloud provider's CLI configured locally with
        permissions to create resources like MySQL databases and networks.


        Configuration file templates


        Base configuration file


        Below is the general structure of a config file. Use this as a base and
        then add any cloud-specific parameters from the sections below.


        # Name of the server deployment.


        name:


        # The server provider type, one of aws, gcp or azure.


        provider:


        # The path to the kubectl config file to use for deployment.


        kubectl_config_path:


        # The Kubernetes namespace to deploy the ZenML server to.


        namespace: zenmlserver


        # The path to the ZenML server helm chart to use for deployment.


        helm_chart:


        # The repository and tag to use for the ZenML server Docker image.


        zenmlserver_image_repo: zenmldocker/zenml


        zenmlserver_image_tag: latest


        # Whether to deploy an nginx ingress controller as part of the
        deployment.


        create_ingress_controller: true


        # Whether to use TLS for the ingress.


        ingress_tls: true


        # Whether to generate self-signed TLS certificates for the ingress.


        ingress_tls_generate_certs: true


        # The name of the Kubernetes secret to use for the ingress.


        ingress_tls_secret_name: zenml-tls-certs


        # The ingress controller's IP address. The ZenML server will be exposed
        on a subdomain of this IP. For AWS, if you have a hostname instead, use
        the following command to get the IP address: `dig +short <hostname>`.


        ingress_controller_ip:


        # Whether to create a SQL database service as part of the recipe.


        deploy_db: true


        # The username and password for the database.
model-index:
  - name: strickvl/finetuned-all-MiniLM-L6-v2
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 384
          type: dim_384
        metrics:
          - type: cosine_accuracy@1
            value: 0.30120481927710846
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5421686746987951
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.6746987951807228
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.7409638554216867
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.30120481927710846
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.18072289156626503
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.13493975903614455
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.07409638554216866
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.30120481927710846
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.5421686746987951
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.6746987951807228
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.7409638554216867
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5191955019858888
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.44787244214955063
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.4579267717676669
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 256
          type: dim_256
        metrics:
          - type: cosine_accuracy@1
            value: 0.29518072289156627
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5301204819277109
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.6325301204819277
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.7349397590361446
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.29518072289156627
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.17670682730923695
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.12650602409638553
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.07349397590361445
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.29518072289156627
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.5301204819277109
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.6325301204819277
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.7349397590361446
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.5118888198675068
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.4409805890227577
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.45029464689656734
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 128
          type: dim_128
        metrics:
          - type: cosine_accuracy@1
            value: 0.2710843373493976
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.5120481927710844
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.6144578313253012
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.6987951807228916
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.2710843373493976
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.1706827309236948
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.12289156626506023
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.06987951807228915
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.2710843373493976
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.5120481927710844
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.6144578313253012
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.6987951807228916
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.4883715088201252
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.4208237712755786
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.4307910346351659
            name: Cosine Map@100
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: dim 64
          type: dim_64
        metrics:
          - type: cosine_accuracy@1
            value: 0.25301204819277107
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.4578313253012048
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.5542168674698795
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.6566265060240963
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.25301204819277107
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.15261044176706828
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.1108433734939759
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.06566265060240963
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.25301204819277107
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.4578313253012048
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.5542168674698795
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.6566265060240963
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.4465853836525359
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.380495792694588
            name: Cosine Mrr@10
          - type: cosine_map@100
            value: 0.39060460620612997
            name: Cosine Map@100

strickvl/finetuned-all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

Model Details

Model Description

  • Model Type: Sentence Transformer
  • Base model: sentence-transformers/all-MiniLM-L6-v2
  • Maximum Sequence Length: 256 tokens
  • Output Dimensionality: 384 tokens
  • Similarity Function: Cosine Similarity
  • Language: en
  • License: apache-2.0

Model Sources

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the πŸ€— Hub
model = SentenceTransformer("strickvl/finetuned-all-MiniLM-L6-v2")
# Run inference
sentences = [
    'Can you explain how the `query_similar_docs` function handles document reranking?',
    'ry_similar_docs(\n\nquestion: str,\n\nurl_ending: str,use_reranking: bool = False,\n\nreturned_sample_size: int = 5,\n\n) -> Tuple[str, str, List[str]]:\n\n"""Query similar documents for a given question and URL ending."""\n\nembedded_question = get_embeddings(question)\n\ndb_conn = get_db_conn()\n\nnum_docs = 20 if use_reranking else returned_sample_size\n\n# get (content, url) tuples for the top n similar documents\n\ntop_similar_docs = get_topn_similar_docs(\n\nembedded_question, db_conn, n=num_docs, include_metadata=True\n\nif use_reranking:\n\nreranked_docs_and_urls = rerank_documents(question, top_similar_docs)[\n\n:returned_sample_size\n\nurls = [doc[1] for doc in reranked_docs_and_urls]\n\nelse:\n\nurls = [doc[1] for doc in top_similar_docs]  # Unpacking URLs\n\nreturn (question, url_ending, urls)\n\nWe get the embeddings for the question being passed into the function and connect to our PostgreSQL database. If we\'re using reranking, we get the top 20 documents similar to our query and rerank them using the rerank_documents helper function. We then extract the URLs from the reranked documents and return them. Note that we only return 5 URLs, but in the case of reranking we get a larger number of documents and URLs back from the database to pass to our reranker, but in the end we always choose the top five reranked documents to return.\n\nNow that we\'ve added reranking to our pipeline, we can evaluate the performance of our reranker and see how it affects the quality of the retrieved documents.\n\nCode Example\n\nTo explore the full code, visit the Complete Guide repository and for this section, particularly the eval_retrieval.py file.\n\nPreviousUnderstanding reranking\n\nNextEvaluating reranking performance\n\nLast updated 15 days ago',
    " use for the database connection.\ndatabase_ssl_ca:# The path to the client SSL certificate to use for the database connection.\ndatabase_ssl_cert:\n\n# The path to the client SSL key to use for the database connection.\ndatabase_ssl_key:\n\n# Whether to verify the database server SSL certificate.\ndatabase_ssl_verify_server_cert:\n\nRun the deploy command and pass the config file above to it.Copyzenml deploy --config=/PATH/TO/FILENote To be able to run the deploy command, you should have your cloud provider's CLI configured locally with permissions to create resources like MySQL databases and networks.\n\nConfiguration file templates\n\nBase configuration file\n\nBelow is the general structure of a config file. Use this as a base and then add any cloud-specific parameters from the sections below.\n\n# Name of the server deployment.\n\nname:\n\n# The server provider type, one of aws, gcp or azure.\n\nprovider:\n\n# The path to the kubectl config file to use for deployment.\n\nkubectl_config_path:\n\n# The Kubernetes namespace to deploy the ZenML server to.\n\nnamespace: zenmlserver\n\n# The path to the ZenML server helm chart to use for deployment.\n\nhelm_chart:\n\n# The repository and tag to use for the ZenML server Docker image.\n\nzenmlserver_image_repo: zenmldocker/zenml\n\nzenmlserver_image_tag: latest\n\n# Whether to deploy an nginx ingress controller as part of the deployment.\n\ncreate_ingress_controller: true\n\n# Whether to use TLS for the ingress.\n\ningress_tls: true\n\n# Whether to generate self-signed TLS certificates for the ingress.\n\ningress_tls_generate_certs: true\n\n# The name of the Kubernetes secret to use for the ingress.\n\ningress_tls_secret_name: zenml-tls-certs\n\n# The ingress controller's IP address. The ZenML server will be exposed on a subdomain of this IP. For AWS, if you have a hostname instead, use the following command to get the IP address: `dig +short <hostname>`.\n\ningress_controller_ip:\n\n# Whether to create a SQL database service as part of the recipe.\n\ndeploy_db: true\n\n# The username and password for the database.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Information Retrieval

Metric Value
cosine_accuracy@1 0.3012
cosine_accuracy@3 0.5422
cosine_accuracy@5 0.6747
cosine_accuracy@10 0.741
cosine_precision@1 0.3012
cosine_precision@3 0.1807
cosine_precision@5 0.1349
cosine_precision@10 0.0741
cosine_recall@1 0.3012
cosine_recall@3 0.5422
cosine_recall@5 0.6747
cosine_recall@10 0.741
cosine_ndcg@10 0.5192
cosine_mrr@10 0.4479
cosine_map@100 0.4579

Information Retrieval

Metric Value
cosine_accuracy@1 0.2952
cosine_accuracy@3 0.5301
cosine_accuracy@5 0.6325
cosine_accuracy@10 0.7349
cosine_precision@1 0.2952
cosine_precision@3 0.1767
cosine_precision@5 0.1265
cosine_precision@10 0.0735
cosine_recall@1 0.2952
cosine_recall@3 0.5301
cosine_recall@5 0.6325
cosine_recall@10 0.7349
cosine_ndcg@10 0.5119
cosine_mrr@10 0.441
cosine_map@100 0.4503

Information Retrieval

Metric Value
cosine_accuracy@1 0.2711
cosine_accuracy@3 0.512
cosine_accuracy@5 0.6145
cosine_accuracy@10 0.6988
cosine_precision@1 0.2711
cosine_precision@3 0.1707
cosine_precision@5 0.1229
cosine_precision@10 0.0699
cosine_recall@1 0.2711
cosine_recall@3 0.512
cosine_recall@5 0.6145
cosine_recall@10 0.6988
cosine_ndcg@10 0.4884
cosine_mrr@10 0.4208
cosine_map@100 0.4308

Information Retrieval

Metric Value
cosine_accuracy@1 0.253
cosine_accuracy@3 0.4578
cosine_accuracy@5 0.5542
cosine_accuracy@10 0.6566
cosine_precision@1 0.253
cosine_precision@3 0.1526
cosine_precision@5 0.1108
cosine_precision@10 0.0657
cosine_recall@1 0.253
cosine_recall@3 0.4578
cosine_recall@5 0.5542
cosine_recall@10 0.6566
cosine_ndcg@10 0.4466
cosine_mrr@10 0.3805
cosine_map@100 0.3906

Training Details

Training Dataset

Unnamed Dataset

  • Size: 1,490 training samples
  • Columns: positive and anchor
  • Approximate statistics based on the first 1000 samples:
    positive anchor
    type string string
    details
    • min: 9 tokens
    • mean: 21.12 tokens
    • max: 49 tokens
    • min: 21 tokens
    • mean: 240.72 tokens
    • max: 256 tokens
  • Samples:
    positive anchor
    Can you provide the details for the Azure service principal with the ID 273d2812-2643-4446-82e6-6098b8ccdaa4? ┃┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ ID β”‚ 273d2812-2643-4446-82e6-6098b8ccdaa4 ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ NAME β”‚ azure-service-principal ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ TYPE β”‚ πŸ‡¦ azure ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ AUTH METHOD β”‚ service-principal ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ RESOURCE TYPES β”‚ πŸ‡¦ azure-generic, πŸ“¦ blob-container, πŸŒ€ kubernetes-cluster, 🐳 docker-registry ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ RESOURCE NAME β”‚ ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ SECRET ID β”‚ 50d9f230-c4ea-400e-b2d7-6b52ba2a6f90 ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ SESSION DURATION β”‚ N/A ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨

    ┃ EXPIRES IN β”‚ N/A ┃

    ┠──────────────────┼────────────────────────────────────────────────────────────────────────────────┨
    What are the new features introduced in ZenML 0.20.0 regarding the Metadata Store? ed to update the way they are registered in ZenML.the updated ZenML server provides a new and improved collaborative experience. When connected to a ZenML server, you can now share your ZenML Stacks and Stack Components with other users. If you were previously using the ZenML Profiles or the ZenML server to share your ZenML Stacks, you should switch to the new ZenML server and Dashboard and update your existing workflows to reflect the new features.

    ZenML takes over the Metadata Store role

    ZenML can now run as a server that can be accessed via a REST API and also comes with a visual user interface (called the ZenML Dashboard). This server can be deployed in arbitrary environments (local, on-prem, via Docker, on AWS, GCP, Azure etc.) and supports user management, workspace scoping, and more.

    The release introduces a series of commands to facilitate managing the lifecycle of the ZenML server and to access the pipeline and pipeline run information:

    zenml connect / disconnect / down / up / logs / status can be used to configure your client to connect to a ZenML server, to start a local ZenML Dashboard or to deploy a ZenML server to a cloud environment. For more information on how to use these commands, see the ZenML deployment documentation.

    zenml pipeline list / runs / delete can be used to display information and about and manage your pipelines and pipeline runs.

    In ZenML 0.13.2 and earlier versions, information about pipelines and pipeline runs used to be stored in a separate stack component called the Metadata Store. Starting with 0.20.0, the role of the Metadata Store is now taken over by ZenML itself. This means that the Metadata Store is no longer a separate component in the ZenML architecture, but rather a part of the ZenML core, located wherever ZenML is deployed: locally on your machine or running remotely as a server.
    Which environment variables should I set to use the Azure Service Connector authentication method in ZenML? -client-id","client_secret": "my-client-secret"}).Note: The remaining configuration options are deprecated and may be removed in a future release. Instead, you should set the ZENML_SECRETS_STORE_AUTH_METHOD and ZENML_SECRETS_STORE_AUTH_CONFIG variables to use the Azure Service Connector authentication method.

    ZENML_SECRETS_STORE_AZURE_CLIENT_ID: The Azure application service principal client ID to use to authenticate with the Azure Key Vault API. If you are running the ZenML server hosted in Azure and are using a managed identity to access the Azure Key Vault service, you can omit this variable.

    ZENML_SECRETS_STORE_AZURE_CLIENT_SECRET: The Azure application service principal client secret to use to authenticate with the Azure Key Vault API. If you are running the ZenML server hosted in Azure and are using a managed identity to access the Azure Key Vault service, you can omit this variable.

    ZENML_SECRETS_STORE_AZURE_TENANT_ID: The Azure application service principal tenant ID to use to authenticate with the Azure Key Vault API. If you are running the ZenML server hosted in Azure and are using a managed identity to access the Azure Key Vault service, you can omit this variable.

    These configuration options are only relevant if you're using Hashicorp Vault as the secrets store backend.

    ZENML_SECRETS_STORE_TYPE: Set this to hashicorp in order to set this type of secret store.

    ZENML_SECRETS_STORE_VAULT_ADDR: The URL of the HashiCorp Vault server to connect to. NOTE: this is the same as setting the VAULT_ADDR environment variable.

    ZENML_SECRETS_STORE_VAULT_TOKEN: The token to use to authenticate with the HashiCorp Vault server. NOTE: this is the same as setting the VAULT_TOKEN environment variable.

    ZENML_SECRETS_STORE_VAULT_NAMESPACE: The Vault Enterprise namespace. Not required for Vault OSS. NOTE: this is the same as setting the VAULT_NAMESPACE environment variable.
  • Loss: MatryoshkaLoss with these parameters:
    {
        "loss": "MultipleNegativesRankingLoss",
        "matryoshka_dims": [
            384,
            256,
            128,
            64
        ],
        "matryoshka_weights": [
            1,
            1,
            1,
            1
        ],
        "n_dims_per_step": -1
    }
    

Training Hyperparameters

Non-Default Hyperparameters

  • eval_strategy: epoch
  • per_device_train_batch_size: 32
  • per_device_eval_batch_size: 16
  • gradient_accumulation_steps: 16
  • learning_rate: 2e-05
  • num_train_epochs: 4
  • lr_scheduler_type: cosine
  • warmup_ratio: 0.1
  • bf16: True
  • tf32: True
  • load_best_model_at_end: True
  • optim: adamw_torch_fused
  • batch_sampler: no_duplicates

All Hyperparameters

Click to expand
  • overwrite_output_dir: False
  • do_predict: False
  • eval_strategy: epoch
  • prediction_loss_only: True
  • per_device_train_batch_size: 32
  • per_device_eval_batch_size: 16
  • per_gpu_train_batch_size: None
  • per_gpu_eval_batch_size: None
  • gradient_accumulation_steps: 16
  • eval_accumulation_steps: None
  • learning_rate: 2e-05
  • weight_decay: 0.0
  • adam_beta1: 0.9
  • adam_beta2: 0.999
  • adam_epsilon: 1e-08
  • max_grad_norm: 1.0
  • num_train_epochs: 4
  • max_steps: -1
  • lr_scheduler_type: cosine
  • lr_scheduler_kwargs: {}
  • warmup_ratio: 0.1
  • warmup_steps: 0
  • log_level: passive
  • log_level_replica: warning
  • log_on_each_node: True
  • logging_nan_inf_filter: True
  • save_safetensors: True
  • save_on_each_node: False
  • save_only_model: False
  • restore_callback_states_from_checkpoint: False
  • no_cuda: False
  • use_cpu: False
  • use_mps_device: False
  • seed: 42
  • data_seed: None
  • jit_mode_eval: False
  • use_ipex: False
  • bf16: True
  • fp16: False
  • fp16_opt_level: O1
  • half_precision_backend: auto
  • bf16_full_eval: False
  • fp16_full_eval: False
  • tf32: True
  • local_rank: 0
  • ddp_backend: None
  • tpu_num_cores: None
  • tpu_metrics_debug: False
  • debug: []
  • dataloader_drop_last: False
  • dataloader_num_workers: 0
  • dataloader_prefetch_factor: None
  • past_index: -1
  • disable_tqdm: True
  • remove_unused_columns: True
  • label_names: None
  • load_best_model_at_end: True
  • ignore_data_skip: False
  • fsdp: []
  • fsdp_min_num_params: 0
  • fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
  • fsdp_transformer_layer_cls_to_wrap: None
  • accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
  • deepspeed: None
  • label_smoothing_factor: 0.0
  • optim: adamw_torch_fused
  • optim_args: None
  • adafactor: False
  • group_by_length: False
  • length_column_name: length
  • ddp_find_unused_parameters: None
  • ddp_bucket_cap_mb: None
  • ddp_broadcast_buffers: False
  • dataloader_pin_memory: True
  • dataloader_persistent_workers: False
  • skip_memory_metrics: True
  • use_legacy_prediction_loop: False
  • push_to_hub: False
  • resume_from_checkpoint: None
  • hub_model_id: None
  • hub_strategy: every_save
  • hub_private_repo: False
  • hub_always_push: False
  • gradient_checkpointing: False
  • gradient_checkpointing_kwargs: None
  • include_inputs_for_metrics: False
  • eval_do_concat_batches: True
  • fp16_backend: auto
  • push_to_hub_model_id: None
  • push_to_hub_organization: None
  • mp_parameters:
  • auto_find_batch_size: False
  • full_determinism: False
  • torchdynamo: None
  • ray_scope: last
  • ddp_timeout: 1800
  • torch_compile: False
  • torch_compile_backend: None
  • torch_compile_mode: None
  • dispatch_batches: None
  • split_batches: None
  • include_tokens_per_second: False
  • include_num_input_tokens_seen: False
  • neftune_noise_alpha: None
  • optim_target_modules: None
  • batch_eval_metrics: False
  • batch_sampler: no_duplicates
  • multi_dataset_batch_sampler: proportional

Training Logs

Epoch Step dim_128_cosine_map@100 dim_256_cosine_map@100 dim_384_cosine_map@100 dim_64_cosine_map@100
0.6667 1 0.3800 0.3986 0.4149 0.3471
2.0 3 0.4194 0.4473 0.4557 0.3762
2.6667 4 0.4308 0.4503 0.4579 0.3906
  • The bold row denotes the saved checkpoint.

Framework Versions

  • Python: 3.10.14
  • Sentence Transformers: 3.0.1
  • Transformers: 4.41.2
  • PyTorch: 2.3.1+cu121
  • Accelerate: 0.31.0
  • Datasets: 2.19.1
  • Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning}, 
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply}, 
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}