Class: Legion::Extensions::Llm::Vllm::Provider

Inherits:
Provider
  • Object
show all
Includes:
Provider::OpenAICompatible, Logging::Helper
Defined in:
lib/legion/extensions/llm/vllm/provider.rb

Overview

vLLM provider implementation for the Legion::Extensions::Llm base provider contract.

Defined Under Namespace

Modules: Capabilities

Class Method Summary collapse

Instance Method Summary collapse

Class Method Details

.capabilities ⇒ Object 



21
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 21

def capabilities = Capabilities

.configuration_options ⇒ Object 



19
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 19

def configuration_options = %i[vllm_api_base vllm_api_key]

.configuration_requirements ⇒ Object 



20
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 20

def configuration_requirements = []

.local? ⇒ Boolean

Returns:

  • (Boolean) 


18
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 18

def local? = false

.registry_publisher ⇒ Object 



23
24
25
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 23

def registry_publisher
  Vllm.registry_publisher
end

.slug ⇒ Object 



17
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 17

def slug = 'vllm'

Instance Method Details

#api_base ⇒ Object 



54
55
56
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 54

def api_base
  normalize_url(config.vllm_api_base || 'localhost:8000')
end

#discover_offerings(live: false) ⇒ Object 



92
93
94
95
96
97
98
99
100
101
102
103
104
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 92

def discover_offerings(live: false, **)
  models = if live
             @cached_models = list_models
           else
             Array(@cached_models)
           end
  models.map { |model_info| offering_from_model(model_info) }.tap do |offerings|
    log.debug { "built #{offerings.size} vLLM offering(s) live=#{live}" }
  end
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: 'vllm.discover_offerings')
  []
end

#headers ⇒ Object 



58
59
60
61
62
63
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 58

def headers
  token = config.vllm_api_key
  return {} if token.nil? || token.to_s.empty?

  { 'Authorization' => "Bearer #{token}" }
end

#health(live: false) ⇒ Object 



72
73
74
75
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 72

def health(live: false)
  log.info { "checking health live=#{live} at #{api_base}#{health_url}" }
  connection.get(health_url).body
end

#health_url ⇒ Object 



65
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 65

def health_url = '/health'

#list_models ⇒ Object 



84
85
86
87
88
89
90
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 84

def list_models
  log.info { "discovering models from #{api_base}#{models_url}" }
  super.tap do |models|
    log.info { "discovered #{models.size} model(s) from vLLM" }
    self.class.registry_publisher.publish_models_async(models, readiness: readiness(live: false))
  end
end

#readiness(live: false) ⇒ Object 



77
78
79
80
81
82
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 77

def readiness(live: false)
  log.info { "checking readiness live=#{live} at #{api_base}" }
  super.tap do |metadata|
    self.class.registry_publisher.publish_readiness_async(metadata) if live
  end
end

#reset_mm_cache ⇒ Object 



119
120
121
122
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 119

def reset_mm_cache
  log.debug { 'resetting vLLM multimodal cache' }
  connection.post(reset_mm_cache_url, {}).body
end

#reset_mm_cache_url ⇒ Object 



68
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 68

def reset_mm_cache_url = '/reset_mm_cache'

#reset_prefix_cache(reset_running_requests: nil, reset_external: nil) ⇒ Object 



111
112
113
114
115
116
117
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 111

def reset_prefix_cache(reset_running_requests: nil, reset_external: nil)
  log.debug do
    "resetting vLLM prefix cache reset_running_requests=#{reset_running_requests.inspect} " \
      "reset_external=#{reset_external.inspect}"
  end
  connection.post(with_query(reset_prefix_cache_url, reset_running_requests:, reset_external:), {}).body
end

#reset_prefix_cache_url ⇒ Object 



67
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 67

def reset_prefix_cache_url = '/reset_prefix_cache'

#settings ⇒ Object 



50
51
52
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 50

def settings
  Vllm.default_settings
end

#sleep(level: 1) ⇒ Object 



124
125
126
127
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 124

def sleep(level: 1)
  log.debug { "putting vLLM worker to sleep level=#{level.inspect}" }
  connection.post(with_query(sleep_url, level:), {}).body
end

#sleep_url ⇒ Object 



69
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 69

def sleep_url = '/sleep'

#stream_usage_supported? ⇒ Boolean

Returns:

  • (Boolean) 


48
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 48

def stream_usage_supported? = true

#version ⇒ Object 



106
107
108
109
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 106

def version
  log.info { "fetching version from #{api_base}#{version_url}" }
  connection.get(version_url).body
end

#version_url ⇒ Object 



66
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 66

def version_url = '/version'

#wake_up(tags: nil) ⇒ Object 



129
130
131
132
133
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 129

def wake_up(tags: nil)
  log.debug { "waking vLLM worker tags=#{Array(tags).inspect}" }
  query = Array(tags).map { |tag| ['tags', tag] }
  connection.post(with_query(wake_up_url, query), {}).body
end

#wake_up_url ⇒ Object 



70
 
# File 'lib/legion/extensions/llm/vllm/provider.rb', line 70

def wake_up_url = '/wake_up'