Class: EmbeddingUtil::Providers::SelfHosted

Inherits:
EmbeddingUtil::Provider show all
Defined in:
lib/embedding_util/providers/self_hosted.rb

Constant Summary collapse

EMBEDDING_BATCH_SIZE = 
32

Instance Attribute Summary

Attributes inherited from EmbeddingUtil::Provider

#config

Instance Method Summary collapse

Methods inherited from EmbeddingUtil::Provider

#initialize, provider_name, #provider_name, supported?

Constructor Details

This class inherits a constructor from EmbeddingUtil::Provider

Instance Method Details

#embed(texts, profile: config.resolved_profile) ⇒ Object 



26
27
28
29
30
31
32
 
# File 'lib/embedding_util/providers/self_hosted.rb', line 26

def embed(texts, profile: config.resolved_profile)
  manager = ServerManager.new(config: config)
  endpoint = manager.ensure_server(:embedding, profile: profile)
  manager.track_activity(:embedding, profile: profile) do
    embed_batches(endpoint, texts, profile)
  end
end

#rerank(query, documents, profile: config.resolved_profile) ⇒ Object 



34
35
36
37
38
39
40
41
42
43
44
 
# File 'lib/embedding_util/providers/self_hosted.rb', line 34

def rerank(query, documents, profile: config.resolved_profile)
  manager = ServerManager.new(config: config)
  endpoint = manager.ensure_server(:reranker, profile: profile)
  rerank_with_activity(manager, endpoint, query, documents, profile)
rescue EndpointError => e
  raise unless retryable_reranker_error?(e) && can_escalate_reranker_ubatch?

  config.reranker_ubatch_size = config.reranker_max_ubatch_size
  endpoint = manager.restart_server(:reranker, profile: profile)
  rerank_with_activity(manager, endpoint, query, documents, profile)
end

#support ⇒ Object 



16
17
18
19
20
21
22
23
24
 
# File 'lib/embedding_util/providers/self_hosted.rb', line 16

def support
  {
    provider: provider_name,
    supported: supported?,
    runtime: RuntimeCommand.resolve(config.runtime),
    shutdown_idle: config.shutdown_idle,
    state_dir: config.state_dir
  }
end

#supported? ⇒ Boolean

Returns:

  • (Boolean) 


12
13
14
 
# File 'lib/embedding_util/providers/self_hosted.rb', line 12

def supported?
  ServerManager.supported?(config)
end