Class: Langchain::LLM::GoogleVertexAI

Inherits:
Base
  • Object
show all
Defined in:
lib/langchain/llm/google_vertex_ai.rb

Overview

Wrapper around the Google Vertex AI APIs: cloud.google.com/vertex-ai

Gem requirements:

gem "googleauth"

Usage:

llm = Langchain::LLM::GoogleVertexAI.new(project_id: ENV["GOOGLE_VERTEX_AI_PROJECT_ID"], region: "us-central1")

Constant Summary collapse

DEFAULTS = 
{
  temperature: 0.1,
  max_output_tokens: 1000,
  top_p: 0.8,
  top_k: 40,
  dimensions: 768,
  embedding_model: "textembedding-gecko",
  chat_model: "gemini-1.0-pro"
}.freeze

Instance Attribute Summary collapse

Attributes inherited from Base

#client

Instance Method Summary collapse

Methods inherited from Base

#chat_parameters, #complete, #default_dimension, #default_dimensions, #summarize

Methods included from DependencyHelper

#depends_on

Constructor Details

#initialize(project_id:, region:, default_options: {}) ⇒ GoogleVertexAI

Returns a new instance of GoogleVertexAI.



28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 28

def initialize(project_id:, region:, default_options: {})
  depends_on "googleauth"

  @authorizer = ::Google::Auth.get_application_default(scope: [
    "https://www.googleapis.com/auth/cloud-platform",
    "https://www.googleapis.com/auth/generative-language.retriever"
  ])
  proj_id = project_id || @authorizer.project_id || @authorizer.quota_project_id
  @url = "https://#{region}-aiplatform.googleapis.com/v1/projects/#{proj_id}/locations/#{region}/publishers/google/models/"

  @defaults = DEFAULTS.merge(default_options)

  chat_parameters.update(
    model: {default: @defaults[:chat_model]},
    temperature: {default: @defaults[:temperature]},
    safety_settings: {default: @defaults[:safety_settings]}
  )
  chat_parameters.remap(
    messages: :contents,
    system: :system_instruction,
    tool_choice: :tool_config
  )
end

Instance Attribute Details

#authorizer ⇒ Object (readonly)

Google Cloud has a project id and a specific region of deployment. For GenAI-related things, a safe choice is us-central1.



26
27
28
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 26

def authorizer
  @authorizer
end

#defaults ⇒ Object (readonly)

Google Cloud has a project id and a specific region of deployment. For GenAI-related things, a safe choice is us-central1.



26
27
28
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 26

def defaults
  @defaults
end

#url ⇒ Object (readonly)

Google Cloud has a project id and a specific region of deployment. For GenAI-related things, a safe choice is us-central1.



26
27
28
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 26

def url
  @url
end

Instance Method Details

#chat(params = {}) ⇒ Langchain::LLM::GoogleGeminiResponse

Generate a chat completion for given messages

Parameters:

  • messages (Array<Hash>) —

    Input messages

  • model (String) —

    The model that will complete your prompt

  • tools (Array<Hash>) —

    The tools to use

  • tool_choice (String) —

    The tool choice to use

  • system (String) —

    The system instruction to use

Returns:

Raises:

  • (ArgumentError) 


80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 80

def chat(params = {})
  params[:system] = {parts: [{text: params[:system]}]} if params[:system]
  params[:tools] = {function_declarations: params[:tools]} if params[:tools]

  raise ArgumentError.new("messages argument is required") if Array(params[:messages]).empty?

  parameters = chat_parameters.to_params(params)
  parameters[:generation_config] = {temperature: parameters.delete(:temperature)} if parameters[:temperature]

  uri = URI("#{url}#{parameters[:model]}:generateContent")

  parsed_response = http_post(uri, parameters)

  wrapped_response = Langchain::LLM::GoogleGeminiResponse.new(parsed_response, model: parameters[:model])

  if wrapped_response.chat_completion || Array(wrapped_response.tool_calls).any?
    wrapped_response
  else
    raise StandardError.new(parsed_response)
  end
end

#embed(text:, model: ) ⇒ Langchain::LLM::GoogleGeminiResponse

Generate an embedding for a given text

Parameters:

  • text (String) —

    The text to generate an embedding for

  • model (String) (defaults to: ) —

    ID of the model to use

Returns:



59
60
61
62
63
64
65
66
67
68
69
70
 
# File 'lib/langchain/llm/google_vertex_ai.rb', line 59

def embed(
  text:,
  model: @defaults[:embedding_model]
)
  params = {instances: [{content: text}]}

  uri = URI("#{url}#{model}:predict")

  parsed_response = http_post(uri, params)

  Langchain::LLM::GoogleGeminiResponse.new(parsed_response, model: model)
end