Module: Legion::LLM::Call::Embeddings

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/call/embeddings.rb

Class Method Summary collapse

Class Method Details

.default_model ⇒ Object 



73
74
75
 
# File 'lib/legion/llm/call/embeddings.rb', line 73

def default_model
  resolve_model(resolve_provider)
end

.generate(text:, model: nil, provider: nil, dimensions: nil, task: :document) ⇒ Object 



13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
 
# File 'lib/legion/llm/call/embeddings.rb', line 13

def generate(text:, model: nil, provider: nil, dimensions: nil, task: :document)
  return { vector: nil, model: model, provider: provider, error: 'LLM not started' } unless LLM.started?

  provider ||= resolve_provider
  return embedding_unavailable_result(model, provider) unless provider
  return { vector: nil, model: model, provider: provider, error: "provider #{provider} is disabled" } if provider_disabled?(provider)

  model ||= resolve_model(provider)
  text    = coerce_text_input(text)
  text    = apply_prefix(text, model: model, task: task)

  return generate_ollama(text: text, model: model) if provider&.to_sym == :ollama
  return generate_azure(text: text, model: model, dimensions: dimensions) if provider&.to_sym == :azure
  return { vector: nil, model: model, provider: provider, error: "provider #{provider} does not support embeddings" } \
    if provider && !provider_supports_embeddings?(provider)

  response = dispatch_native_embedding(text: text, model: model, provider: provider, dimensions: dimensions)
  tokens = extract_input_tokens(response)
  emit_embedding_metering(provider: provider, model: model, tokens: tokens)
  vector = normalize_vectors_first(response[:result])
  vector = apply_dimension_enforcement(vector, provider)
  return dimension_error(model, provider, vector) if vector.is_a?(String)

  { vector: vector, model: model, provider: provider, dimensions: vector&.size || 0, tokens: tokens }
rescue StandardError => e
  handle_exception(e, level: :warn)
  handle_embed_failure(e, text: text, failed_provider: provider, failed_model: model)
end

.generate_batch(texts:, model: nil, provider: nil, dimensions: nil, task: :document) ⇒ Object 



42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
 
# File 'lib/legion/llm/call/embeddings.rb', line 42

def generate_batch(texts:, model: nil, provider: nil, dimensions: nil, task: :document)
  return texts.map { |_| { vector: nil, error: 'LLM not started' } } unless LLM.started?

  provider ||= resolve_provider
  return unavailable_batch_result(texts, provider, model) unless provider

  disabled_result = disabled_batch_result(texts, provider, model)
  return disabled_result if disabled_result

  model  ||= resolve_model(provider)
  texts    = texts.map { |t| apply_prefix(coerce_text_input(t), model: model, task: task) }

  return generate_ollama_batch(texts: texts, model: model) if provider&.to_sym == :ollama
  return generate_azure_batch(texts: texts, model: model, dimensions: dimensions) if provider&.to_sym == :azure

  unless !provider || provider_supports_embeddings?(provider)
    return texts.each_with_index.map do |_, i|
      { vector: nil, model: model, provider: provider, dimensions: 0, index: i,
        error: "provider #{provider} does not support embeddings" }
    end
  end

  response = dispatch_native_embedding(text: texts, model: model, provider: provider, dimensions: dimensions)
  normalize_vectors(response[:result]).each_with_index.map do |vec, i|
    build_batch_entry(vec, model, provider, i)
  end
rescue StandardError => e
  handle_exception(e, level: :warn)
  texts.map { |_| { vector: nil, model: model, provider: provider, error: e.message } }
end