legion-llm 0.6.3 → 0.6.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 126a28a6fc47bfabc448da8efcc20361012d8cfbc00e7041be92ef5aa76f5363
-  data.tar.gz: fa334053810d6e7850eba2b9604347bd4c0d087f4ee529fbbc4a52c2a80ea15b
+  metadata.gz: eac9a164ca981a40e76eb183e9a1a6809a407d28c8e898fbdbcb38d65e26406a
+  data.tar.gz: 993e3c5f37a40aced5ba745e967b5e5c662bea7b734f2fe2343d248d9b0a5e7c
 SHA512:
-  metadata.gz: a18685a5074155d67f345ae29b22913336eebde0931309ed47ce2cfa4a2a99d81b74e0a8ac2fb8763fc438474992cf7830d8453650bf433f66b4abaf878c608d
-  data.tar.gz: 3b1de6afbcf73da75064feb0aa8ed4182461f3cf24b05dd40b17f49c0a0f2379c4d8f2d3bb39a4d8397266c5cd6593ef7957a02c1fe4cb54b120640d4feace88
+  metadata.gz: 46fcb27d23b7d531d3b7a9a14d9248f9c61ae3eaeca46a46b8d892d33f3ae1bbc7eff4073cd3cd4ad54aee4170a44a3d186f7d820770d8b3610f4baccc4e2a73
+  data.tar.gz: b005e7b350db1aef999dada991407da94bc44ca0911a4cf959edfb333d60ca5b90aa87f9772f0cba8a3c85726951351dcc124621eb79df29511cd98d0d50bf82

data/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [0.6.4] - 2026-04-01
+
+### Added
+- Direct Azure OpenAI embedding provider with SNI host header injection, bypasses ruby_llm and DNS — connects to private endpoint IP with correct Host header
+- Azure embedding supports single and batch requests, dimension enforcement, and settings-driven IP override (`llm.embedding.azure.ip`)
+- Default embedding fallback chain: azure -> ollama -> bedrock -> openai
+
 ## [0.6.3] - 2026-03-31
 
 ### Changed

data/lib/legion/llm/embeddings.rb

@@ -38,6 +38,7 @@ module Legion
           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
 
           response   = RubyLLM.embed(text, **build_opts(model, provider, dimensions))
           vector     = apply_dimension_enforcement(response.vectors.first, provider)
@@ -57,6 +58,7 @@ module Legion
           texts      = texts.map { |t| apply_prefix(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
 
           response = RubyLLM.embed(texts, **build_opts(model, provider, dimensions))
           response.vectors.each_with_index.map do |vec, i|
@@ -284,6 +286,71 @@ module Legion
           OLLAMA_CONTEXT_CHARS[base] || OLLAMA_DEFAULT_CONTEXT_CHARS
         end
 
+        # ── Azure OpenAI (direct HTTP with SNI, bypasses ruby_llm) ──
+
+        def generate_azure(text:, model:, dimensions: nil)
+          result = azure_embed_request(model: model, input: text, dimensions: dimensions)
+          vector = result.dig('data', 0, 'embedding')
+          vector = apply_dimension_enforcement(vector, :azure) if vector
+          return dimension_error(model, :azure, vector) if vector.is_a?(String)
+
+          tokens = result.dig('usage', 'total_tokens') || 0
+          { vector: vector, model: model, provider: :azure, dimensions: vector&.size || 0, tokens: tokens }
+        end
+
+        def generate_azure_batch(texts:, model:, dimensions: nil)
+          result = azure_embed_request(model: model, input: texts, dimensions: dimensions)
+          (result['data'] || []).each_with_index.map do |entry, i|
+            build_batch_entry(entry['embedding'], model, :azure, i)
+          end
+        rescue StandardError => e
+          Legion::Logging.warn("Azure batch embedding failed: #{e.message}") if defined?(Legion::Logging)
+          texts.map { |_| { vector: nil, model: model, provider: :azure, error: e.message } }
+        end
+
+        def azure_embed_request(model:, input:, dimensions: nil)
+          settings = azure_embedding_settings
+          api_base = settings[:api_base]
+          api_key  = settings[:api_key]
+          ip       = settings[:ip]
+          raise 'Azure OpenAI embedding not configured (llm.providers.azure.api_base required)' unless api_base
+
+          host = URI.parse(api_base).host
+          target = ip || host
+          path = "/openai/deployments/#{model}/embeddings?api-version=2024-02-01"
+
+          require 'net/http'
+          http = Net::HTTP.new(target, 443)
+          http.use_ssl = true
+          http.open_timeout = 5
+          http.read_timeout = 30
+
+          req = Net::HTTP::Post.new(path)
+          req['Content-Type'] = 'application/json'
+          req['Host'] = host
+          req['api-key'] = api_key
+          body = { input: input }
+          body[:dimensions] = dimensions || TARGET_DIMENSION
+          req.body = ::JSON.dump(body)
+
+          response = http.request(req)
+          raise "Azure embed failed: #{response.code} #{response.body}" unless response.is_a?(Net::HTTPSuccess)
+
+          ::JSON.parse(response.body)
+        end
+
+        def azure_embedding_settings
+          base = Legion::Settings.dig(:llm, :providers, :azure) || {}
+          embed = Legion::Settings.dig(:llm, :embedding, :azure) || {}
+          {
+            api_base: embed[:api_base] || base[:api_base],
+            api_key:  embed[:api_key] || base[:api_key] || base[:auth_token],
+            ip:       embed[:ip]
+          }
+        end
+
+        # ── Ollama (direct HTTP, bypasses ruby_llm) ──
+
         def ollama_embed_request(model:, input:)
           base_url = Legion::Settings.dig(:llm, :providers, :ollama, :base_url) || 'http://localhost:11434'
           conn = Faraday.new(url: base_url) do |f|

data/lib/legion/llm/settings.rb

@@ -165,9 +165,10 @@ module Legion
         {
           dimension:         1024,
           enforce_dimension: true,
-          provider_fallback: %w[ollama bedrock openai],
+          provider_fallback: %w[azure ollama bedrock openai],
           provider_models:   {
             ollama:  'mxbai-embed-large',
+            azure:   'text-embedding-3-small',
             bedrock: 'amazon.titan-embed-text-v2:0',
             openai:  'text-embedding-3-small'
           },

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.6.3'
+    VERSION = '0.6.4'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.6.3
+  version: 0.6.4
 platform: ruby
 authors:
 - Esity