Module: Legion::LLM::API::Translators::AnthropicResponse

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/api/translators/anthropic_response.rb

Class Method Summary collapse

Class Method Details

.extract_tool_calls(pipeline_response) ⇒ Object 



275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 275

def self.extract_tool_calls(pipeline_response)
  return [] unless pipeline_response.respond_to?(:tools)

  Array(pipeline_response.tools).map do |tc|
    source = tc.respond_to?(:source) ? tc.source : (tc[:source] || tc['source'] || {})
    is_legionio = legionio_tool_source?(source)

    {
      id:        if tc.respond_to?(:id)
                   tc.id
                 else
                   tc[:id] || tc['id'] || (is_legionio ? "srvtoolu_#{SecureRandom.hex(10)}" : "toolu_#{SecureRandom.hex(10)}")
                 end,
      name:      tc.respond_to?(:name)      ? tc.name      : (tc[:name] || tc['name'] || ''),
      arguments: tc.respond_to?(:arguments) ? tc.arguments : (tc[:arguments] || tc['arguments'] || {}),
      result:    tc.respond_to?(:result)    ? tc.result    : (tc[:result] || tc['result']),
      source:    source,
      legionio:  is_legionio
    }
  end
end

.format(pipeline_response, model:, request_id: nil) ⇒ Object

Format internal pipeline response into Anthropic Messages API shape.



14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 14

def self.format(pipeline_response, model:, request_id: nil)
  log.debug("[llm][api][anthropic] action=format request_id=#{request_id} model=#{model}")

  msg     = pipeline_response.message
  content = extract_content(msg, pipeline_response)
  tokens  = pipeline_response.respond_to?(:tokens) ? pipeline_response.tokens : nil
  routing = pipeline_response.respond_to?(:routing) ? (pipeline_response.routing || {}) : {}

  resolved_model = routing[:model] || routing['model'] || model

  {
    id:            request_id || "msg_#{SecureRandom.hex(12)}",
    type:          'message',
    role:          'assistant',
    content:       content,
    model:         resolved_model.to_s,
    stop_reason:   format_stop_reason(pipeline_response),
    stop_sequence: nil,
    usage:         format_usage(tokens)
  }
end

.format_chunk(text, index: 0) ⇒ Object

Emit Anthropic streaming events for a single text chunk. Returns the SSE lines for the delta event.



38
39
40
41
42
43
44
45
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 38

def self.format_chunk(text, index: 0)
  log.debug("[llm][api][anthropic] action=format_chunk index=#{index} text=#{text[0, 80]}")
  {
    type:  'content_block_delta',
    index: index,
    delta: { type: 'text_delta', text: text }
  }
end

.format_stop_reason(pipeline_response) ⇒ Object 



304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 304

def self.format_stop_reason(pipeline_response)
  tool_calls = extract_tool_calls(pipeline_response)

  # Only return 'pause_turn' for LegionIO tools that are still pending
  # (i.e., executed server-side but result not yet available).
  # If all LegionIO tools have results, the tool loop completed and
  # the turn should end normally.
  pending_legionio = tool_calls.select { |tc| tc[:legionio] == true && tc[:result].nil? }
  return 'pause_turn' if pending_legionio.any?

  # Client-side (passthrough) tool_use blocks without results mean
  # the client needs to execute them.
  return 'tool_use' if tool_calls.any? { |tc| tc[:legionio] != true && tc[:result].nil? }

  return 'end_turn' unless pipeline_response.respond_to?(:stop)

  stop = pipeline_response.stop
  reason = stop.is_a?(Hash) ? (stop[:reason] || stop['reason']) : stop.to_s

  case reason.to_s
  when 'tool_use'       then 'tool_use'
  when 'max_tokens'     then 'max_tokens'
  when 'content_filter' then 'content_filter'
  when 'stop'
    pipeline_response.respond_to?(:stop_sequence) && pipeline_response.stop_sequence ? 'stop_sequence' : 'end_turn'
  else 'end_turn'
  end
end

.normalize_thinking_payload(value) ⇒ Object 



247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 247

def self.normalize_thinking_payload(value)
  return nil unless value

  if value.is_a?(Hash)
    normalized = value.transform_keys { |key| key.respond_to?(:to_sym) ? key.to_sym : key }
    content = normalized[:content] || normalized[:text] || normalized[:thinking]
    signature = normalized[:signature]
    data = normalized[:data]
    redacted = normalized[:redacted] || normalized[:type].to_s == 'redacted_thinking'
    type = normalized[:type]
  else
    content = value.respond_to?(:content) ? value.content : nil
    content = value.text if content.nil? && value.respond_to?(:text)
    content = value unless value.respond_to?(:content) || value.respond_to?(:text)
    signature = value.respond_to?(:signature) ? value.signature : nil
    data = value.respond_to?(:data) ? value.data : nil
    redacted = false
    type = nil
  end

  content = content.to_s unless content.nil?
  signature = signature.to_s unless signature.nil?
  data = data.to_s unless data.nil?
  return nil if content.to_s.empty? && signature.to_s.empty? && data.to_s.empty?

  { content: content, signature: signature, data: data, redacted: redacted, type: type }.compact
end

.streaming_events(pipeline_response, model:, request_id: nil, full_text: '') ⇒ Object

Ordered sequence of SSE event hashes for a complete streaming response. Caller emits each via emit_sse_event.



49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 49

def self.streaming_events(pipeline_response, model:, request_id: nil, full_text: '')
  log.debug("[llm][api][anthropic] action=streaming_events request_id=#{request_id} text_length=#{full_text.length}")

  tokens  = pipeline_response.respond_to?(:tokens) ? pipeline_response.tokens : nil
  routing = pipeline_response.respond_to?(:routing) ? (pipeline_response.routing || {}) : {}
  resolved_model = routing[:model] || routing['model'] || model

  tool_calls = extract_tool_calls(pipeline_response)
  content_index = 0

  events = []

  events << ['message_start', {
    type:    'message_start',
    message: {
      id:            request_id || "msg_#{SecureRandom.hex(12)}",
      type:          'message',
      role:          'assistant',
      content:       [],
      model:         resolved_model.to_s,
      stop_reason:   nil,
      stop_sequence: nil,
      usage:         { input_tokens: token_count(tokens, :input), output_tokens: 0 }
    }
  }]

  events << ['ping', { type: 'ping' }]

  # Emit text block only when there's real text content.
  # When tool calls are present and the text is just JSON arguments
  # (common with vLLM/qwen forced tool choices), skip the text block
  # so the Anthropic-format SSE emits only tool_use blocks with
  # stop_reason=tool_use.
  text_is_tool_arguments = tool_calls.any? && text_looks_like_tool_json?(full_text)
  unless full_text.empty? || text_is_tool_arguments
    events << ['content_block_start', {
      type:          'content_block_start',
      index:         content_index,
      content_block: { type: 'text', text: '' }
    }]

    events << ['content_block_delta', {
      type:  'content_block_delta',
      index: content_index,
      delta: { type: 'text_delta', text: full_text }
    }]

    events << ['content_block_stop', { type: 'content_block_stop', index: content_index }]
    content_index += 1
  end

  # Emit tool_use/server_tool_use blocks (index may have been bumped above)
  tool_calls.each do |tc|
    if tc[:legionio] == true
      # Server tool: emit call + result together
      events << ['content_block_start', {
        type:          'content_block_start',
        index:         content_index,
        content_block: { type: 'server_tool_use', id: tc[:id], name: tc[:name], input: tc[:arguments] || {} }
      }]
      events << ['content_block_delta', {
        type:  'content_block_delta',
        index: content_index,
        delta: { type: 'input_json_delta', partial_json: Legion::JSON.dump(tc[:arguments] || {}) }
      }]
      events << ['content_block_stop', { type: 'content_block_stop', index: content_index }]
      content_index += 1

      # Emit result
      result = tc[:result]
      if result
        result_str = if result.is_a?(String)
                       result
                     else
                       begin
                         Legion::JSON.dump(result)
                       rescue StandardError
                         result.to_s
                       end
                     end
        events << ['content_block_start', {
          type:          'content_block_start',
          index:         content_index,
          content_block: { type: 'server_tool_result', id: tc[:id], content: [] }
        }]
        events << ['content_block_delta', {
          type:  'content_block_delta',
          index: content_index,
          delta: { type: 'content_block_delta', content: [{ type: 'text', text: result_str }] }
        }]
        events << ['content_block_stop', { type: 'content_block_stop', index: content_index }]
        content_index += 1
      end
    else
      events << ['content_block_start', {
        type:          'content_block_start',
        index:         content_index,
        content_block: { type: 'tool_use', id: tc[:id], name: tc[:name], input: tc[:arguments] || {} }
      }]
      events << ['content_block_delta', {
        type:  'content_block_delta',
        index: content_index,
        delta: { type: 'input_json_delta', partial_json: Legion::JSON.dump(tc[:arguments] || {}) }
      }]
      events << ['content_block_stop', { type: 'content_block_stop', index: content_index }]
      content_index += 1
    end
  end

  stop_reason = format_stop_reason(pipeline_response)
  events << ['message_delta', {
    type:  'message_delta',
    delta: { stop_reason: stop_reason, stop_sequence: nil },
    usage: { output_tokens: token_count(tokens, :output) }
  }]

  events << ['message_stop', { type: 'message_stop' }]

  events
end

.thinking_content_block(pipeline_response) ⇒ Object 



220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 220

def self.thinking_content_block(pipeline_response)
  thinking = thinking_payload(pipeline_response)
  return nil unless thinking

  if thinking[:redacted] || thinking[:type].to_s == 'redacted_thinking'
    data = thinking[:data] || thinking[:signature]
    return nil if data.to_s.empty?

    return { type: 'redacted_thinking', data: data.to_s }
  end

  block = { type: 'thinking', thinking: thinking[:content].to_s }
  block[:signature] = thinking[:signature].to_s unless thinking[:signature].to_s.empty?
  block
end

.thinking_payload(pipeline_response) ⇒ Object 



236
237
238
239
240
241
242
243
244
245
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 236

def self.thinking_payload(pipeline_response)
  return nil unless pipeline_response.respond_to?(:thinking)

  thinking_data = begin
    pipeline_response.thinking
  rescue Exception # rubocop:disable Style/RescueException
    nil
  end
  normalize_thinking_payload(thinking_data)
end

.token_count(tokens, key) ⇒ Object 



340
341
342
343
344
345
346
347
348
 
# File 'lib/legion/llm/api/translators/anthropic_response.rb', line 340

def self.token_count(tokens, key)
  return 0 if tokens.nil?
  return tokens[key] || tokens[key.to_s] || 0 if tokens.is_a?(Hash)

  method_name = { input: :input_tokens, output: :output_tokens }[key]
  return tokens.public_send(method_name) if method_name && tokens.respond_to?(method_name)

  0
end