Model - NomadicML

Overview

This document provides an overview of three key model classes used in the SDK: Model, OpenAIModel, and SagemakerModel. These classes are designed to facilitate the integration and usage of various LLMs in your applications. Each class inherits from the base Model class, which provides a consistent interface for setting up and running models with specific API keys and hyperparameters.

The Model class is an abstract class and cannot be instantiated directly. Subclasses like OpenAIModel and SagemakerModel provide specific implementations for their respective platforms.
The api_keys parameter is essential and must be provided for the model to function correctly.
Each subclass is responsible for handling the specific setup and execution of its respective platform’s model.

Fields (`Model` Class)

Parameter	Type	Default	Description	Required
`api_keys`	`Dict[str, str]`	N/A	API keys needed to run the model.	Yes
`llm`	`Optional[LLM]`	`None`, Nomadic creates the LlamaIndex LLM for you. However, you can also define any LlamaIndex LLM here.	The language model (LLM) instance used to run the experiment	No
`required_api_keys`	`ClassVar[Set[str]]`	`set()`	A set of API keys that the model expects to receive.	No
`hyperparameters`	`ClassVar[Dict]`	`DEFAULT_HYPERPARAMETER_SEARCH_SPACE`	A dictionary defining the hyperparameters that can be tuned for the model.	No
`name`	`str`	`"my model"`	The name of the model.	No
`client_id`	`Optional[str]`	`None`	ID of `Model` on the Nomadic Workspace, if synched	No

Methods

get_required_api_keys() -> Set[str]: Returns the set of expected API keys for the model.
get_hyperparameters() -> Dict: Returns the hyperparameters that can be tuned for the model.
run(**kwargs) -> RunResult: Abstract method that must be implemented by subclasses to execute the model with the provided parameters and return a RunResult object.

Supported Model Subclasses

Overview

The OpenAIModel class is a subclass of Model specifically designed to work with OpenAI’s GPT models. It handles the setup and execution of OpenAI models, allowing you to easily integrate them into your applications.

Class Variables

Parameter	Type	Default	Description	Required
`model`	`Optional[str]`	”gpt-4o-mini”	Name of the OpenAI model to use - from the OpenAI model list.	No
`required_api_keys`	`ClassVar[Set[str]]`	`{"OPENAI_API_KEY"}`	The set of expected API keys for OpenAI.	No
`hyperparameters`	`ClassVar[Dict]`	A default hyperparameter search space with `temperature`, `max_tokens`, `top_p`, and OpenAI-specific parameters `frequency_penalty`, `presence_penalty`	A dictionary defining the hyperparameters that can be tuned for the OpenAI model.	No

Example Usage

from nomadic.model import OpenAIModel
from nomadic.tuner import tune

experiment = Experiment(
    model=OpenAIModel(api_keys={"OPENAI_API_KEY": "<Your OpenAI API Key>"}),
    params={"temperature", "max_tokens"},
    evaluation_dataset=json.loads(
        requests.get(
            "https://example.com/evaluation_dataset.json"
        ).content
    ),
    evaluator=SemanticSimilarityEvaluator(embed_model=OpenAIEmbedding()),
)

results = experiment.run(
    param_dict={
        "temperature": tune.choice([0.1, 0.5, 0.9]),
        "max_tokens": tune.choice([50, 100, 200]),
    })
best_result = experiment.experiment_result.best_run_result

Overview

The SagemakerModel class is a subclass of Model designed to work with AWS SageMaker endpoints. It configures the SageMaker model using the provided AWS credentials and runs experiments using specified parameters.

Class Variables

Parameter	Type	Default	Description	Required
`required_api_keys`	`ClassVar[Set[str]]`	`{"AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY", "AWS_DEFAULT_REGION", "ENDPOINT_NAME"}`	A set of API keys expected for the SageMaker model.	Yes
`hyperparameters`	`ClassVar[Dict]`	A default hyperparameter search space with `temperature`, `max_tokens`, and `top_p`.	A dictionary of hyperparameters specific to the SageMaker model.	No

Example Usage

from nomadic.model import SagemakerModel
from nomadic.tuner import tune

experiment = Experiment(
    name = "Sample_Nomadic_Experiment",
    model = SagemakerModel(
        api_keys={
            "AWS_ACCESS_KEY_ID": "<...>",
            "AWS_SECRET_ACCESS_KEY": "<...>",
            "AWS_DEFAULT_REGION": "<...>",
            "ENDPOINT_NAME": "<...>",
        }),
    evaluator = BatchEvalRunner(...),
    hp_space = {
        'temperature': tune.choice([0.1, 0.3, 0.5, 0.7, 0.9])
        'top_k': tune.randint(3,5)
    }
    current_hp_values = {
        'temperature'=0.5,
        'top_k'=5,
        'top_p'=8
    },
    evaluation_dataset = {
        {
            'Context': "You are a helpful assistant writing a transcript for ...",
            'Instruction': "Absolutely do not hallucinate. Capture all relevant ...",
            'Answer': "The generated summary is shown below: ..."
        }
    }
)
results = experiment.run()

Overview

The TogetherAIModel class is a subclass of Model that is implemented to work with models hosted on Together.AI. It handles the setup and execution of Together.AI models, allowing you to easily integrate them with your applications.

Class Variables

Parameter	Type	Default	Description	Required
`model`	`Optional[str]`	N/A	Name of the model to use - from the Together.AI Chat Models list.	No
`required_api_keys`	`ClassVar[Set[str]]`	`{"TOGETHER_API_KEY"}`	The set of expected API keys for Together.AI.	No
`hyperparameters`	`ClassVar[Dict]`	A default hyperparameter search space with `temperature`, `max_tokens`, `top_p`.	A dictionary defining the hyperparameters that can be tuned for the Together.AI model.	No

Example Usage

from nomadic.model import TogetherAIModel
from nomadic.tuner import tune

experiment = Experiment(
    model=TogetherAIModel(api_keys={"TOGETHER_API_KEY": "<Your Together.AI API Key>"}),
    params={"temperature", "max_tokens"},
    evaluation_dataset=json.loads(
        requests.get(
            "https://example.com/evaluation_dataset.json"
        ).content
    ),
    evaluator=SemanticSimilarityEvaluator(embed_model=OpenAIEmbedding()),
)

results = experiment.run(
    param_dict={
        "temperature": tune.choice([0.1, 0.5, 0.9]),
        "max_tokens": tune.choice([50, 100, 200]),
    })
best_result = experiment.experiment_result.best_run_result

Overview

The VLLMModel class is a subclass of Model that is implemented to work with models served through vLLM. It handles the setup and execution, allowing you to easily integrate vLLM-served models them with your applications.

Class Variables

Parameter	Type	Default	Description	Required
`api_url`	`Optional[str]`	N/A	URL of model being served through vLLM.	No
`required_api_keys`	`ClassVar[Set[str]]`	`{}`	The set of expected API keys for model being served through vLLM.	No
`hyperparameters`	`ClassVar[Dict]`	A default hyperparameter search space with `temperature`, `max_tokens`, `top_p`.	A dictionary defining the hyperparameters that can be tuned for the model being served on vLLM.	No

Example Usage

from nomadic.model import VLLMModel
from nomadic.tuner import tune

experiment = Experiment(
    model=VLLMModel(api_url="http://localhost:8000/generate"),
    params={"temperature", "max_tokens"},
    evaluation_dataset=json.loads(
        requests.get(
            "https://example.com/evaluation_dataset.json"
        ).content
    ),
    evaluator=SemanticSimilarityEvaluator(embed_model=OpenAIEmbedding()),
)

results = experiment.run(
    param_dict={
        "temperature": tune.choice([0.1, 0.5, 0.9]),
        "max_tokens": tune.choice([50, 100, 200]),
    })
best_result = experiment.experiment_result.best_run_result

SDK Reference

​Overview

​Fields (Model Class)

​Methods

​Supported Model Subclasses

Overview

Class Variables

Example Usage

Overview

Class Variables

Example Usage

Overview

Class Variables

Example Usage

Overview

Class Variables

Example Usage

Overview

Fields (`Model` Class)

Methods

Supported Model Subclasses