WeaviateEx.API.GenerativeConfig (WeaviateEx v0.7.4)

View Source

Configuration builders for Weaviate generative search (RAG) providers.

This module provides helper functions to configure generative AI modules for collection schema definitions. These configurations enable RAG (Retrieval Augmented Generation) capabilities on collections.

Supported Providers

  • OpenAI - GPT-4, GPT-3.5, O1/O3 models
  • Azure OpenAI - Azure-hosted OpenAI models
  • Anthropic - Claude 3.5 Sonnet, Claude 3 Opus/Haiku
  • Cohere - Command R/R+ models
  • Mistral - Mistral Large, Medium models
  • Google - Gemini Pro, PaLM models
  • AWS - Bedrock/SageMaker hosted models
  • Ollama - Self-hosted models
  • Databricks - Foundation Model APIs
  • NVIDIA - NIM models
  • FriendliAI - Serverless inference
  • XAI - Grok models
  • Anyscale - Endpoints
  • ContextualAI - Grounded generation

Examples

# Configure OpenAI for a collection
config = GenerativeConfig.openai(model: "gpt-4")
Collections.create("Article", %{
  properties: [...],
  generative_config: config
})

# Configure Anthropic with options
config = GenerativeConfig.anthropic(
  model: "claude-3-5-sonnet-latest",
  temperature: 0.7,
  max_tokens: 4096
)

# Configure Ollama for local inference
config = GenerativeConfig.ollama(
  model: "llama3",
  api_endpoint: "http://localhost:11434"
)

Summary

Types

config()
provider()

Functions

anthropic(opts \\ [])

Create Anthropic generative configuration.

anyscale(opts \\ [])

Create Anyscale generative configuration.

aws(opts \\ [])

Create AWS (Bedrock/SageMaker) generative configuration.

azure_openai(opts \\ [])

Create Azure OpenAI generative configuration.

cohere(opts \\ [])

Create Cohere generative configuration.

contextualai(opts \\ [])

Create ContextualAI generative configuration.

custom(name, opts)

Create a configuration for a custom/unlisted generative provider.

databricks(opts \\ [])

Create Databricks generative configuration.

friendliai(opts \\ [])

Create FriendliAI generative configuration.

google(opts \\ [])

Create Google (Gemini/PaLM) generative configuration.

mistral(opts \\ [])

Create Mistral generative configuration.

nvidia(opts \\ [])

Create NVIDIA NIM generative configuration.

ollama(opts \\ [])

Create Ollama generative configuration.

openai(opts \\ [])

Create OpenAI generative configuration.

provider_module(provider)

Get the Weaviate module name for a provider.

providers()

List all supported generative providers.

xai(opts \\ [])

Create XAI (Grok) generative configuration.

Types

config()

@type config() :: %{required(String.t()) => map()}

provider()

@type provider() ::
  :openai
  | :anthropic
  | :cohere
  | :mistral
  | :google
  | :aws
  | :ollama
  | :databricks
  | :nvidia
  | :friendliai
  | :xai
  | :anyscale
  | :contextualai

Functions

anthropic(opts \\ [])

@spec anthropic(keyword()) :: config()

Create Anthropic generative configuration.

Options

  • :model - Model name (e.g., "claude-3-5-sonnet-latest")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :top_k - Top-k sampling parameter
  • :top_p - Top-p (nucleus) sampling parameter

Examples

GenerativeConfig.anthropic(model: "claude-3-5-sonnet-latest")

GenerativeConfig.anthropic(
  model: "claude-3-opus-20240229",
  temperature: 0.7,
  max_tokens: 4096
)

anyscale(opts \\ [])

@spec anyscale(keyword()) :: config()

Create Anyscale generative configuration.

Options

  • :model - Model name (e.g., "meta-llama/Llama-3-70b-chat-hf")
  • :base_url - Custom API endpoint URL
  • :temperature - Sampling temperature

Examples

GenerativeConfig.anyscale(model: "meta-llama/Llama-3-70b-chat-hf")

aws(opts \\ [])

@spec aws(keyword()) :: config()

Create AWS (Bedrock/SageMaker) generative configuration.

Options

  • :model - Model ID (e.g., "anthropic.claude-v2")
  • :region - AWS region
  • :service - AWS service ("bedrock" or "sagemaker")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate

Examples

GenerativeConfig.aws(
  model: "anthropic.claude-v2",
  region: "us-east-1"
)

GenerativeConfig.aws(
  model: "anthropic.claude-v2",
  region: "us-west-2",
  service: "bedrock"
)

azure_openai(opts \\ [])

@spec azure_openai(keyword()) :: config()

Create Azure OpenAI generative configuration.

Options

  • :resource_name - Azure resource name (required)
  • :deployment_id - Deployment ID (required)
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate

Examples

GenerativeConfig.azure_openai(
  resource_name: "my-resource",
  deployment_id: "gpt-4-deployment"
)

cohere(opts \\ [])

@spec cohere(keyword()) :: config()

Create Cohere generative configuration.

Options

  • :model - Model name (e.g., "command-r-plus", "command-r")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :k - Top-k sampling parameter
  • :p - Top-p (nucleus) sampling parameter

Examples

GenerativeConfig.cohere(model: "command-r-plus")

GenerativeConfig.cohere(
  model: "command-r",
  temperature: 0.8,
  k: 5
)

contextualai(opts \\ [])

@spec contextualai(keyword()) :: config()

Create ContextualAI generative configuration.

Options

  • :model - Model name (optional)

Examples

GenerativeConfig.contextualai()

GenerativeConfig.contextualai(model: "grounded-generation")

custom(name, opts)

@spec custom(
  String.t(),
  keyword()
) :: config()

Create a configuration for a custom/unlisted generative provider.

Use this when connecting to a generative AI provider not natively supported by Weaviate. All options are passed through with automatic camelCase conversion.

Arguments

  • name - Identifier for the custom provider (for your reference)
  • opts - Provider configuration options

Common Options

  • :api_endpoint - API endpoint URL (required for most providers)
  • :model - Model identifier
  • :api_key_header - Header name for API key authentication
  • :base_url - Base URL for the API

Any additional options are passed through with snake_case to camelCase conversion.

Examples

# Custom LLM endpoint
GenerativeConfig.custom("my-llm",
  api_endpoint: "https://my-llm.example.com/v1/generate",
  model: "custom-model-v1"
)

# Local LLM
GenerativeConfig.custom("local-ollama",
  api_endpoint: "http://localhost:11434",
  model: "llama2"
)

# Enterprise LLM with custom auth header
GenerativeConfig.custom("enterprise-llm",
  api_endpoint: "https://llm.enterprise.com/api",
  model: "enterprise-gpt",
  api_key_header: "X-API-Key"
)

databricks(opts \\ [])

@spec databricks(keyword()) :: config()

Create Databricks generative configuration.

Options

  • :endpoint - Databricks serving endpoint URL (required)
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :top_k - Top-k sampling parameter
  • :top_p - Top-p (nucleus) sampling parameter

Examples

GenerativeConfig.databricks(
  endpoint: "https://my-workspace.cloud.databricks.com"
)

friendliai(opts \\ [])

@spec friendliai(keyword()) :: config()

Create FriendliAI generative configuration.

Options

  • :model - Model name
  • :base_url - Custom API endpoint URL
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate

Examples

GenerativeConfig.friendliai(model: "mixtral-8x7b")

google(opts \\ [])

@spec google(keyword()) :: config()

Create Google (Gemini/PaLM) generative configuration.

Options

  • :model - Model name (e.g., "gemini-pro", "gemini-1.5-pro")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :project_id - GCP project ID

Examples

GenerativeConfig.google(model: "gemini-1.5-pro")

GenerativeConfig.google(
  model: "gemini-pro",
  project_id: "my-gcp-project"
)

mistral(opts \\ [])

@spec mistral(keyword()) :: config()

Create Mistral generative configuration.

Options

  • :model - Model name (e.g., "mistral-large-latest")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :base_url - Custom API endpoint URL

Examples

GenerativeConfig.mistral(model: "mistral-large-latest")

nvidia(opts \\ [])

@spec nvidia(keyword()) :: config()

Create NVIDIA NIM generative configuration.

Options

  • :model - Model name (e.g., "meta/llama-3.1-405b-instruct")
  • :base_url - Custom API endpoint URL
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate

Examples

GenerativeConfig.nvidia(model: "meta/llama-3.1-405b-instruct")

ollama(opts \\ [])

@spec ollama(keyword()) :: config()

Create Ollama generative configuration.

Options

  • :model - Model name (e.g., "llama3", "mistral")
  • :api_endpoint - Ollama API endpoint URL
  • :temperature - Sampling temperature

Examples

GenerativeConfig.ollama(model: "llama3")

GenerativeConfig.ollama(
  model: "llama3:70b",
  api_endpoint: "http://localhost:11434"
)

openai(opts \\ [])

@spec openai(keyword()) :: config()

Create OpenAI generative configuration.

Options

  • :model - Model name (e.g., "gpt-4", "gpt-3.5-turbo")
  • :temperature - Sampling temperature (0.0-2.0)
  • :max_tokens - Maximum tokens to generate
  • :base_url - Custom API endpoint URL

Examples

GenerativeConfig.openai(model: "gpt-4")

GenerativeConfig.openai(
  model: "gpt-4-turbo",
  temperature: 0.7,
  max_tokens: 1000
)

provider_module(provider)

@spec provider_module(provider()) :: String.t()

Get the Weaviate module name for a provider.

Examples

GenerativeConfig.provider_module(:openai)
# => "generative-openai"

GenerativeConfig.provider_module(:google)
# => "generative-palm"

providers()

@spec providers() :: [provider()]

List all supported generative providers.

Examples

GenerativeConfig.providers()
# => [:openai, :anthropic, :cohere, ...]

xai(opts \\ [])

@spec xai(keyword()) :: config()

Create XAI (Grok) generative configuration.

Options

  • :model - Model name (e.g., "grok-2")
  • :temperature - Sampling temperature
  • :max_tokens - Maximum tokens to generate
  • :base_url - Custom API endpoint URL

Examples

GenerativeConfig.xai(model: "grok-2")